2024
2024 Jan 5 | 2024 Jan 11 | 2024 Jan 25 |
2024 Feb 9 | 2024 Feb 28 | 2024 Mar 21 |
2024 Apr 12 | 2024 Apr 24 | 2024 May 8 |
2024 May 22 | 2024 Jun 10 | 2024 Jun 24 |
2024 Jul 5 | 2024 Jul 8 | 2024 Jul 22 |
2024 Aug 12 |
2023 Jan 6 | 2023 Jan 11 | 2023 Jan 25 |
2023 Feb 22 | 2023 Feb 28 | 2023 Mar 21 |
2023 Apr 11 | 2023 Apr 24 | 2023 May 8 |
2023 May 22 | 2023 Jun 9 | 2023 Jun 23 |
2023 Jul 5 | 2023 Jul 7 | 2023 Jul 21 |
2023 Aug 11 | 2023 Aug 25 | 2023 Sep 8 |
2023 Sep 22 | 2023 Oct 13 | 2023 Oct 27 |
2023 Nov 10 | 2023 Nov 27 | 2023 Dec 11 |
2022 Jan 7 | 2022 Jan 11 | 2022 Jan 25 |
2022 Feb 22 | 2022 Feb 28 | 2022 Mar 21 |
2022 Apr 8 | 2022 Apr 22 | 2022 May 6 |
2022 May 20 | 2022 Jun 10 | 2022 Jun 24 |
2022 Jul 5 | 2022 Jul 8 | 2022 Jul 22 |
2022 Aug 12 | 2022 Aug 26 | 2022 Sep 9 |
2022 Sep 23 | 2022 Oct 14 | 2022 Oct 28 |
2022 Nov 11 | 2022 Nov 28 | 2022 Dec 9 |
2021 Jan 8 | 2021 Jan 11 | 2021 Jan 25 |
2021 Feb 22 | 2021 Feb 26 | 2021 Mar 18 |
2021 Apr 6 | 2021 Apr 19 | 2021 May 7 |
2021 May 21 | 2021 Jun 4 | 2021 Jun 21 |
2021 Jul 2 | 2021 Jul 9 | 2021 Jul 21 |
2021 Aug 20 | 2021 Sep 7 | 2021 Sep 20 |
2021 Sep 27 | 2021 Oct 11 | 2021 Nov 8 |
2021 Nov 18 | 2021 Dec 2 | 2021 Dec 16 |
2020 Jan 8 | 2020 Jan 10 | 2020 Jan 24 |
2020 Feb 20 | 2020 Feb 28 | 2020 Mar 18 |
2020 Apr 6 | 2020 Apr 17 | 2020 May 7 |
2020 May 21 | 2020 Jun 4 | 2020 Jun 19 |
2020 Jul 2 | 2020 Jul 9 | 2020 Jul 27 |
2020 Aug 20 | 2020 Sep 4 | 2020 Sep 18 |
2020 Sep 25 | 2020 Oct 9 | 2020 Nov 4 |
2020 Nov 18 | 2020 Dec 2 | 2020 Dec 16 |
2019 Jan 8 | 2019 Jan 10 | 2019 Jan 25 |
2019 Feb 20 | 2019 Feb 28 | 2019 Mar 18 |
2019 Apr 8 | 2019 Apr 23 | 2019 May 7 |
2019 May 21 | 2019 Jun 4 | 2019 Jun 20 |
2019 Jul 2 | 2019 Jul 9 | 2019 Jul 26 |
2019 Aug 22 | 2019 Sep 5 | 2019 Sep 19 |
2019 Sep 26 | 2019 Oct 10 | 2019 Nov 4 |
2019 Nov 18 | 2019 Dec 2 | 2019 Dec 16 |
2018 Jan 4 | 2018 Jan 9 | 2018 Jan 25 |
2018 Feb 16 | 2018 Feb 28 | 2018 Mar 16 |
2018 Apr 6 | 2018 Apr 20 | 2018 May 4 |
2018 May 18 | 2018 Jun 1 | 2018 Jun 20 |
2018 Jul 2 | 2018 Jul 6 | 2018 Jul 27 |
2018 Aug 21 | 2018 Sep 6 | 2018 Sep 19 |
2018 Sep 28 | 2018 Oct 12 | 2018 Nov 2 |
2018 Nov 16 | 2018 Nov 30 | 2018 Dec 14 |
2017 Jan 4 | 2017 Jan 9 | 2017 Jan 25 |
2017 Feb 17 | 2017 Feb 28 | 2017 Mar 17 |
2017 Apr 7 | 2017 Apr 20 | 2017 May 5 |
2017 May 19 | 2017 Jun 2 | 2017 Jun 20 |
2017 Jul 3 | 2017 Jul 7 | 2017 Jul 27 |
2017 Aug 21 | 2017 Sep 6 | 2017 Sep 19 |
2017 Sep 29 | 2017 Oct 13 | 2017 Nov 2 |
2017 Nov 16 | 2017 Dec 1 | 2017 Dec 15 |
2016 Jan 4 | 2016 Jan 8 | 2016 Jan 29 |
2016 Feb 18 | 2016 Feb 29 | 2016 Mar 18 |
2016 Apr 7 | 2016 Apr 20 | 2016 May 5 |
2016 May 19 | 2016 Jun 2 | 2016 Jun 20 |
2016 Jul 8 | 2016 Jul 8 | 2016 Jul 27 |
2016 Aug 19 | 2016 Sep 6 | 2016 Sep 19 |
2016 Sep 30 | 2016 Oct 14 | 2016 Nov 2 |
2016 Nov 16 | 2016 Dec 1 | 2016 Dec 15 |
2015 Jan 9 | 2015 Jan 28 | 2015 Jan 30 |
2015 Feb 18 | 2015 Feb 27 | 2015 Mar 11 |
2015 Mar 31 | 2015 Apr 20 | 2015 Apr 30 |
2015 May 19 | 2015 May 26 | 2015 Jun 23 |
2015 Jul 1 | 2015 Jul 13 | 2015 Jul 27 |
2015 Aug 18 | 2015 Sep 3 | 2015 Sep 18 |
2015 Sep 30 | 2015 Oct 15 | 2015 Oct 30 |
2015 Nov 16 | 2015 Dec 2 | 2015 Dec 15 |
2014 Jan 1 | 2014 Jan | 2014 Jan 1 |
2014 Feb | 2014 Feb | 2014 Mar |
2014 Apr | 2014 Apr | 2014 Apr |
2014 May 1 | 2014 Jun 1 | 2014 Jul 1 |
2014 Jul 1 | 2014 Aug 1 | 2014 Aug 18 |
2014 Sep 1 | 2014 Sep 2 | 2014 Sep 15 |
2014 Sep 29 | 2014 Oct 13 | 2014 Oct 29 |
2014 Nov 10 | 2014 Dec 1 | 2014 Dec 16 |
2013 Jan | 2013 Jan | 2013 Jan |
2013 Feb | 2013 Feb | 2013 Mar |
2013 Apr | 2013 Apr | 2013 Apr |
2013 May | 2013 May | 2013 Jun |
2013 Jul | 2013 Jul | 2013 Jul |
2013 Aug | 2013 Aug | 2013 Sep |
2013 Sep | 2013 Oct | 2013 Oct |
2013 Nov | 2013 Nov | 2013 Dec |
2012 Jan | 2012 Jan | 2012 Jan |
2012 Feb | 2012 Feb | 2012 Mar |
2012 Mar | 2012 Apr | 2012 Apr |
2012 May | 2012 May | 2012 Jun |
2012 Jul | 2012 Jul | 2012 Jul |
2012 Aug | 2012 Aug | 2012 Sep |
2012 Sep | 2012 Oct | 2012 Oct |
2012 Nov | 2012 Nov | 2012 Dec |
2011 Jan | 2011 Jan | 2011 Jan |
2011 Feb | 2011 Mar | 2011 Mar |
2011 Mar | 2011 Apr | 2011 Apr |
2011 May | 2011 May | 2011 Jun |
2011 Jul | 2011 Jul | 2011 Jul 1 |
2011 Aug | 2011 Aug | 2011 Sep |
2011 Sep | 2011 Oct | 2011 Oct |
2011 Nov | 2011 Nov | 2011 Dec |
2010 Jan | 2010 Jan | 2010 Jan |
2010 Jan | 2010 Mar | 2010 Mar |
2010 Apr | 2010 Apr | 2010 May |
2010 May | 2010 Jun | 2010 Jun |
2010 Jul | 2010 Jul | 2010 Jul 1 |
2010 Aug | 2010 Aug | 2010 Sep |
2010 Sep | 2010 Oct | 2010 Oct |
2010 Nov | 2010 Nov | 2010 Dec |
2009 Jan | 2009 Jan | 2009 Feb |
2009 Feb | 2009 Mar | 2009 Apr |
2009 Apr | 2009 Apr | 2009 May |
2009 May | 2009 Jun | 2009 Jun |
2009 Jul | 2009 Jul | 2009 Jul 1 |
2009 Aug | 2009 Aug | 2009 Sep |
2009 Sep | 2009 Oct | 2009 Oct |
2009 Nov | 2009 Nov | 2009 Dec |
2008 Jan | 2008 Jan | 2008 Feb |
2008 Feb | 2008 Mar | 2008 Mar |
2008 Apr | 2008 Apr | 2008 May |
2008 May | 2008 Jun | 2008 Jun |
2008 Jul | 2008 Jul 1 | 2008 Aug |
2008 Aug | 2008 Sep | 2008 Sep |
2008 Oct | 2008 Oct | 2008 Nov |
2008 Nov | 2008 Dec | 2008 Dec |
2007 Jan | 2007 Jan | 2007 Jan |
2007 Feb | 2007 Feb | 2007 Mar |
2007 Mar | 2007 Apr | 2007 Apr |
2007 May | 2007 May | 2007 Jun |
2007 Jun | 2007 Jul | 2007 Jul |
2007 Jul | 2007 Aug | 2007 Aug |
2007 Sep | 2007 Sep | 2007 Oct |
2007 Nov | 2007 Dec | 2007 Dec |
2006 Jan 1 | 2006 | 2006 |
2006 | 2006 | 2006 |
2006 | 2006 | 2006 |
2006 | 2006 | 2006 |
2006 Jul 1 | 2006 | 2006 |
2006 | 2006 Sep | 2006 Sep |
2006 Oct | 2006 Oct | 2006 Nov |
2006 Nov | 2006 Dec | 2006 Dec |
2005 Jan 1 | 2005 | 2005 |
2005 | 2005 | 2005 |
2005 | 2005 | 2005 |
2005 | 2005 | 2005 |
2005 Jul 1 | 2005 | 2005 |
2005 | 2005 | 2005 |
2005 | 2005 | 2005 |
2005 | 2005 | 2005 |
2004 Jan 1 | 2004 | 2004 |
2004 | 2004 | 2004 |
2004 | 2004 | 2004 |
2004 | 2004 | 2004 |
2004 Jul 1 | 2004 | 2004 |
2004 | 2004 | 2004 |
2004 | 2004 | 2004 |
2004 | 2004 | 2004 |
2003 Jan 1 | 2003 Jan 15 | 2003 Feb 1 |
2003 Feb 15 | 2003 Mar 1 | 2003 Mar 15 |
2003 Apr 1 | 2003 Apr 15 | 2003 May 1 |
2003 May 15 | 2003 Jun 1 | 2003 Jun 15 |
2003 Jul 1 | 2003 Jul 15 | 2003 Aug 1 |
2003 Aug 15 | 2003 Sep 1 | 2003 Sep 15 |
2003 Oct 1 | 2003 Oct 15 | 2003 Nov 1 |
2003 Nov 15 | 2003 Dec 1 | 2003 Dec 15 |
2002 Jan 1 | 2002 Jan 15 | 2002 Feb 1 |
2002 Feb 15 | 2002 Mar 1 | 2002 Mar 15 |
2002 Apr 1 | 2002 Apr 15 | 2002 May 1 |
2002 May 15 | 2002 Jun 1 | 2002 Jun 15 |
2002 Jul 1 | 2002 Jul 15 | 2002 Aug 1 |
2002 Aug 15 | 2002 Sep 1 | 2002 Sep 15 |
2002 Oct 1 | 2002 Oct 15 | 2002 Nov 1 |
2002 Nov 15 | 2002 Dec 1 | 2002 Dec 15 |
2001 Jan 1 | 2001 Jan 15 | 2001 Feb 1 |
2001 Feb 15 | 2001 Mar 1 | 2001 Mar 15 |
2001 Apr 1 | 2001 Apr 15 | 2001 May 1 |
2001 May 15 | 2001 Jun 1 | 2001 Jun 15 |
2001 Jul 1 | 2001 Jul 15 | 2001 Aug 1 |
2001 Aug 15 | 2001 Sep 1 | 2001 Sep 15 |
2001 Oct 1 | 2001 Oct 15 | 2001 Nov 1 |
2001 Nov 15 | 2001 Dec 1 | 2001 Dec 15 |
2000 Jan 1 | 2000 Jan 15 | 2000 Feb 1 |
2000 Feb 15 | 2000 Mar 1 | 2000 Mar 15 |
2000 Apr 1 | 2000 Apr 15 | 2000 May 1 |
2000 May 15 | 2000 Jun 1 | 2000 Jun 15 |
2000 Jul 1 | 2000 Jul 15 | 2000 Aug 1 |
2000 Aug 15 | 2000 Sep 1 | 2000 Sep 15 |
2000 Oct 1 | 2000 Oct 15 | 2000 Nov 1 |
2000 Nov 15 | 2000 Dec 1 | 2000 Dec 15 |
1999 Jan 1 | 1999 Jan 15 | 1999 Feb 1 |
1999 Feb 15 | 1999 Mar 1 | 1999 Mar 15 |
1999 Apr 1 | 1999 Apr 15 | 1999 May 1 |
1999 May 15 | 1999 Jun 1 | 1999 Jun 15 |
1999 Jul 1 | 1999 Jul 15 | 1999 Aug 1 |
1999 Aug 15 | 1999 Sep 1 | 1999 Sep 15 |
1999 Oct 1 | 1999 Oct 15 | 1999 Nov 1 |
1999 Nov 15 | 1999 Dec 1 | 1999 Dec 15 |
1998 Jan 1 | 1998 Jan 15 | 1998 Feb 1 |
1998 Feb 15 | 1998 Mar 1 | 1998 Mar 15 |
1998 Apr 1 | 1998 Apr 15 | 1998 May 1 |
1998 May 15 | 1998 Jun 1 | 1998 Jun 15 |
1998 Jul 1 | 1998 Jul 15 | 1998 Aug 1 |
1998 Aug 15 | 1998 Sep 1 | 1998 Sep 15 |
1998 Oct 1 | 1998 Oct 15 | 1998 Nov 1 |
1998 Nov 15 | 1998 Dec 1 | 1998 Dec 15 |
1997 Jan 1 | 1997 Jan 15 | 1997 Feb 1 |
1997 Feb 15 | 1997 Mar 1 | 1997 Mar 15 |
1997 Apr 1 | 1997 Apr 15 | 1997 May 1 |
1997 May 15 | 1997 Jun 1 | 1997 Jun 15 |
1997 Jul 1 | 1997 Jul 15 | 1997 Aug 1 |
1997 Aug 15 | 1997 Sep 1 | 1997 Sep 15 |
1997 Oct 1 | 1997 Oct 15 | 1997 Nov 1 |
1997 Nov 15 | 1997 Dec 1 | 1997 Dec 15 |
1996 Jan 1 | 1996 Jan 15 | 1996 Feb 1 |
1996 Feb 15 | 1996 Mar 1 | 1996 Mar 15 |
1996 Apr 1 | 1996 Apr 15 | 1996 May 1 |
1996 May 15 | 1996 Jun 1 | 1996 Jun 15 |
1996 Jul 1 | 1996 Jul 15 | 1996 Aug 1 |
1996 Aug 15 | 1996 Sep 1 | 1996 Sep 15 |
1996 Oct 1 | 1996 Oct 15 | 1996 Nov 1 |
1996 Nov 15 | 1996 Dec 1 | 1996 Dec 15 |
1995 Jan 11 | 1995 Jan 25 | 1995 Feb 11 |
1995 Feb 25 | 1995 Mar 11 | 1995 Mar 25 |
1995 Apr 11 | 1995 Apr 25 | 1995 May 11 |
1995 May 25 | 1995 Jun 11 | 1995 Jun 25 |
1995 Jul 11 | 1995 Jul 25 | 1995 Aug 11 |
1995 Aug 25 | 1995 Sep 11 | 1995 Sep 25 |
1995 Oct 11 | 1995 Oct 25 | 1995 Nov 11 |
1995 Nov 25 | 1995 Dec 11 | 1995 Dec 25 |
1994 Jan 11 | 1994 Jan 25 | 1994 Feb 11 |
1994 Feb 25 | 1994 Mar 11 | 1994 Mar 25 |
1994 Apr 11 | 1994 Apr 25 | 1994 May 11 |
1994 May 25 | 1994 Jun 11 | 1994 Jun 25 |
1994 Jul 11 | 1994 Jul 25 | 1994 Aug 11 |
1994 Aug 25 | 1994 Sep | 1994 Sep 11 |
1994 Sep 25 | 1994 Oct 11 | 1994 Oct 25 |
1994 Nov 11 | 1994 Nov 25 | 1994 Dec 11 |
1994 Dec 25 |
1993 Jan 11 | 1993 Jan 25 | 1993 Feb 11 |
1993 Feb 25 | 1993 Mar 11 | 1993 Mar 25 |
1993 Apr 11 | 1993 Apr 25 | 1993 May 11 |
1993 May 25 | 1993 Jun 11 | 1993 Jun 25 |
1993 Jul 1 | 1993 Jul 11 | 1993 Jul 25 |
1993 Aug 11 | 1993 Aug 25 | 1993 Sep 11 |
1993 Sep 25 | 1993 Oct 11 | 1993 Oct 25 |
1993 Nov 11 | 1993 Nov 25 | 1993 Dec 11 |
1993 Dec 25 |
1992 Jan 11 | 1992 Jan 25 | 1992 Feb 11 |
1992 Feb 25 | 1992 Mar 11 | 1992 Mar 25 |
1992 Apr 11 | 1992 Apr 25 | 1992 May 11 |
1992 May 11 | 1992 May 25 | 1992 Jun 11 |
1992 Jun 25 | 1992 Jul 11 | 1992 Jul 25 |
1992 Aug 11 | 1992 Aug 25 | 1992 Sep 11 |
1992 Sep 25 | 1992 Oct 11 | 1992 Oct 25 |
1992 Nov 11 | 1992 Nov 25 | 1992 Dec 11 |
1992 Dec 25 |
1991 Jan 11 | 1991 Jan 25 | 1991 Feb 11 |
1991 Feb 25 | 1991 Mar 11 | 1991 Mar 25 |
1991 Apr 11 | 1991 Apr 25 | 1991 Apr 25 |
1991 May 11 | 1991 May 25 | 1991 Jun 11 |
1991 Jun 25 | 1991 Jul 11 | 1991 Jul 25 |
1991 Aug 11 | 1991 Aug 25 | 1991 Sep 11 |
1991 Sep 25 | 1991 Oct 11 | 1991 Oct 25 |
1991 Nov 11 | 1991 Nov 25 | 1991 Dec |
1991 Dec 11 | 1991 Dec 25 |
1990 Jan 11 | 1990 Jan 25 | 1990 Feb 11 |
1990 Feb 25 | 1990 Mar 11 | 1990 Mar 25 |
1990 Apr 11 | 1990 Apr 25 | 1990 Apr 25 |
1990 May 11 | 1990 May 25 | 1990 Jun 11 |
1990 Jun 25 | 1990 Jul 11 | 1990 Jul 25 |
1990 Aug 11 | 1990 Aug 25 | 1990 Sep 11 |
1990 Sep 25 | 1990 Oct 11 | 1990 Oct 25 |
1990 Nov 11 | 1990 Nov 25 | 1990 Dec 11 |
1990 Dec 25 |
1989 Jan 11 | 1989 Jan 25 | 1989 Feb 11 |
1989 Feb 25 | 1989 Mar 11 | 1989 Mar 25 |
1989 Apr 11 | 1989 Apr 25 | 1989 May 11 |
1989 May 25 | 1989 Jun 12 | 1989 Jun 26 |
1989 Jul 11 | 1989 Jul 25 | 1989 Aug 11 |
1989 Aug 25 | 1989 Sep 12 | 1989 Sep 25 |
1989 Oct 11 | 1989 Oct 25 | 1989 Nov 11 |
1989 Nov 25 | 1989 Dec 11 | 1989 Dec 25 |
1989 |
1988 Jan 11 | 1988 Jan 25 | 1988 Feb 11 |
1988 Feb 25 | 1988 Mar 11 | 1988 Mar 25 |
1988 Mar 25 | 1988 Apr 11 | 1988 Apr 25 |
1988 May 11 | 1988 May 25 | 1988 Jun 10 |
1988 Jun 24 | 1988 Jul 11 | 1988 Jul 25 |
1988 Jul 25 | 1988 Aug 11 | 1988 Aug 25 |
1988 Sep 12 | 1988 Sep 26 | 1988 Oct 11 |
1988 Oct 25 | 1988 Nov 11 | 1988 Nov 25 |
1988 Dec 9 | 1988 Dec 23 | 1988 |
1987 Jan 12 | 1987 Jan 26 | 1987 Feb 11 |
1987 Feb 25 | 1987 Mar 11 | 1987 Mar 25 |
1987 Apr 10 | 1987 Apr 24 | 1987 May 11 |
1987 May 26 | 1987 Jun 11 | 1987 Jun 25 |
1987 Jul 10 | 1987 Jul 24 | 1987 Aug 11 |
1987 Aug 25 | 1987 Sep 11 | 1987 Sep 25 |
1987 Oct 12 | 1987 Oct 26 | 1987 Nov 11 |
1987 Nov 25 | 1987 Dec 10 | 1987 Dec 23 |
1987 |
1986 Jan 10 | 1986 Jan 24 | 1986 Feb 11 |
1986 Feb 25 | 1986 Mar 11 | 1986 Mar 25 |
1986 Apr 11 | 1986 Apr 25 | 1986 May 12 |
1986 May 27 | 1986 Jun 11 | 1986 Jun 25 |
1986 Jul 11 | 1986 Jul 25 | 1986 Aug 11 |
1986 Aug 26 | 1986 Sep 11 | 1986 Sep 25 |
1986 Oct 10 | 1986 Oct 24 | 1986 Nov 11 |
1986 Nov 25 | 1986 Dec 9 | 1986 Dec 22 |
1986 |
1985 Jan 11 | 1985 Jan 25 | 1985 Feb 11 |
1985 Feb 25 | 1985 Mar 11 | 1985 Mar 25 |
1985 Apr 11 | 1985 Apr 25 | 1985 May 10 |
1985 May 24 | 1985 Jun 11 | 1985 Jun 25 |
1985 Jul 11 | 1985 Jul 25 | 1985 Aug 12 |
1985 Aug 26 | 1985 Sep 11 | 1985 Sep 25 |
1985 Oct 11 | 1985 Oct 25 | 1985 Nov 11 |
1985 Nov 25 | 1985 Dec 9 | 1985 Dec 20 |
1985 |
1984 Jan 11 | 1984 Jan 11 | 1984 Jan 25 |
1984 Feb 10 | 1984 Feb 24 | 1984 Mar 12 |
1984 Mar 26 | 1984 Apr 11 | 1984 Apr 25 |
1984 May 11 | 1984 May 25 | 1984 Jun 11 |
1984 Jun 25 | 1984 Jul 11 | 1984 Jul 25 |
1984 Aug 10 | 1984 Aug 24 | 1984 Sep 11 |
1984 Sep 25 | 1984 Oct 11 | 1984 Oct 25 |
1984 Nov 12 | 1984 Nov 26 | 1984 Dec 11 |
1984 Dec 21 | 1984 |
1983 Jan 11 | 1983 Jan 25 | 1983 Feb 11 |
1983 Feb 25 | 1983 Mar 11 | 1983 Mar 25 |
1983 Apr 11 | 1983 Apr 25 | 1983 May 11 |
1983 May 25 | 1983 Jun 11 | 1983 Jun 25 |
1983 Jul 11 | 1983 Jul 25 | 1983 Aug 11 |
1983 Aug 25 | 1983 Sep 10 | 1983 Sep 24 |
1983 Oct 11 | 1983 Oct 25 | 1983 Nov 11 |
1983 Nov 25 | 1983 Dec 10 | 1983 Dec 20 |
1982 Jan 11 | 1982 Jan 22 | 1982 Feb 11 |
1982 Feb 25 | 1982 Mar 11 | 1982 Mar 25 |
1982 Apr 10 | 1982 Apr 24 | 1982 May 11 |
1982 May 25 | 1982 Jun 11 | 1982 Jun 25 |
1982 Jul 10 | 1982 Jul 24 | 1982 Aug 11 |
1982 Aug 25 | 1982 Sep 11 | 1982 Sep 25 |
1982 Oct 11 | 1982 Oct 25 | 1982 Nov 11 |
1982 Nov 25 | 1982 Dec 11 | 1982 Dec 20 |
1981 Jan 10 | 1981 Jan 24 | 1981 Feb 11 |
1981 Feb 25 | 1981 Mar 11 | 1981 Mar 25 |
1981 Apr 10 | 1981 Apr 24 | 1981 May 11 |
1981 May 25 | 1981 Jun 11 | 1981 Jun 25 |
1981 Jul 10 | 1981 Jul 24 | 1981 Aug 11 |
1981 Aug 25 | 1981 Sep 11 | 1981 Sep 25 |
1981 Oct 10 | 1981 Oct 24 | 1981 Nov 11 |
1981 Nov 25 | 1981 Dec 11 | 1981 Dec 21 |
1980 Jan 11 | 1980 Jan 25 | 1980 Feb 11 |
1980 Feb 25 | 1980 Mar 11 | 1980 Mar 25 |
1980 Apr 11 | 1980 Apr 25 | 1980 May 10 |
1980 May 24 | 1980 Jun 11 | 1980 Jun 25 |
1980 Jul 11 | 1980 Jul 25 | 1980 Aug 11 |
1980 Aug 25 | 1980 Sep 11 | 1980 Sep 25 |
1980 Oct 10 | 1980 Oct 24 | 1980 Nov 11 |
1980 Nov 25 | 1980 Dec 11 | 1980 Dec 20 |
1979 Jan | 1979 | 1979 |
1979 Feb | 1979 Mar | 1979 Apr |
1979 Jun 11 | 1979 Jun 25 | 1979 Jul 11 |
1979 Jul 25 | 1979 Aug 10 | 1979 Aug 24 |
1979 Sep 11 | 1979 Sep 25 | 1979 Oct 10 |
1979 Oct 25 | 1979 Nov 10 | 1979 Nov 24 |
1979 Dec 11 | 1979 Dec 20 |
1978 Jan | 1978 Jan 1 | 1978 Feb |
1978 Mar | 1978 Apr | 1978 May |
1978 Jun | 1978 Jul | 1978 Jul 1 |
1978 Aug | 1978 Sep | 1978 Oct |
1978 Nov | 1978 Dec |
1977 Jan | 1977 Feb | 1977 Mar |
1977 Apr | 1977 May | 1977 Jun |
1977 Jul | 1977 Aug | 1977 Sep |
1977 Oct | 1977 Nov | 1977 Dec |
1976 Jan | 1976 Feb | 1976 Mar |
1976 Apr | 1976 May | 1976 Jun |
1976 Jul | 1976 Aug | 1976 Sep |
1976 Oct | 1976 Nov | 1976 Dec |
1975 Jan | 1975 Feb | 1975 Mar |
1975 Apr | 1975 May | 1975 Jun |
1975 Jul | 1975 Aug | 1975 Sep |
1975 Oct | 1975 Nov | 1975 Dec |
1974 Jan | 1974 Feb | 1974 Mar |
1974 Apr | 1974 May | 1974 Jun |
1974 Jul | 1974 Aug | 1974 Sep |
1974 Oct | 1974 Nov | 1974 Dec |
research news
The UB research is applicable to two types of small DNA viruses: papillomaviruses, such as HPV, and polyomaviruses, such as the one shown here in a 3D print. Polyomaviruses infect most people without causing serious disease but they can lead to serious diseases and some cancers in immunologically weakened individuals. Photo: NIAID
By ELLEN GOLDBAUM
Published August 22, 2024
New research reveals that triggering a cell’s DNA damage response could be a promising avenue for developing novel treatments against several rare but devastating viruses for which no antiviral treatments exist, possibly including human papilloma virus (HPV), which causes cancer.
Published online on Aug. 10 in Nucleic Acids Research , the paper focuses on the DNA damage response pathway and demonstrates how this pathway can reduce the function of a viral enzyme, a helicase, resulting in suppressing viral replication.
“This research is significant both for understanding how cells respond to DNA damage, to prevent them from becoming cancerous in the first place, how targeting this pathway can be used in new cancer treatments, and because it now opens up possibilities for new approaches to treating some rare but devastating viral infections,” says Thomas Melendy, senior author on the paper and associate professor of microbiology and immunology in the Jacobs School of Medicine and Biomedical Sciences at UB.
The research focuses on a process called the DNA damage response, part of which has evolved to stop or slow DNA synthesis whenever cellular DNA damage occurs. “These pathways are important for preventing exacerbation of DNA damage that can lead to either cell death or cancer,” explains Rama Dey-Rao, research assistant professor of microbiology and immunology in the Jacobs School and joint first author on the paper with Caleb Hominski, a previous student in the lab.
When these pathways are activated, DNA replication is suppressed at sites in the genome called origins; at the same time, the progression of DNA replication forks also slows down. Replication forks, so-called because their structure resembles a fork, are where large groups of proteins coordinate genome replication through the unwinding and synthesis of DNA.
Melendy says that while quite a bit is known about how DNA damage response causes cells to stop DNA replication origins from “firing,” it’s been much harder to figure out how the progression of replication forks slows down in response to DNA damage.
“Researchers have been very interested in how that slowing occurs because it’s so dramatic,” says Melendy. “DNA damage response pathways cause replication forks to slow down progression by about ten-fold. This ten-fold slowdown means that synthesis of the cell’s genome, which usually takes about 12 hours, would take nearly five days, greatly increasing the time cells have to repair DNA damage.”
For years, Melendy and his colleagues have been studying two types of small DNA viruses: papillomaviruses, such as HPV, and polyomaviruses, which infect most people without causing serious disease but can lead to serious diseases and some cancers in immunologically weakened individuals. A rare cancer caused by a polyomavirus caused the death of musician Jimmy Buffett in 2023.
“We previously showed that in response to DNA damage, HPV does not stop or slow its DNA replication, while polyomaviruses do stop or slow their DNA replication,” says Melendy, “so by comparing and contrasting these two virus types we can gain insights into how polyomavirus DNA replication is slowed in response to DNA damage, which in turn provides us insights into how human cells slow replication forks.”
In the current research, they demonstrate that a phosphorylation site — where a phosphate is added to a molecule — on the major polyomavirus DNA replication and transcription protein is highly conserved in polyomaviruses across many animal species.
“The conservation of this phosphorylation/modification across polyomaviruses that have evolved to infect many different species of mammals suggested it was likely important,” says Melendy.
To study the effects of this, the UB researchers made a mutation at the specific amino acid residue on the viral protein where this phosphorylation occurs to mimic the addition of a phosphate group being there.
When they expressed this mutant viral protein in human cells using a system to evaluate polyomavirus DNA replication, they found the virus’s genome replication was decreased by 10-fold. However, viral transcription was unaffected, indicating that phosphorylation on that amino acid residue has a highly specific effect on viral DNA replication, but didn’t affect other functions of that protein.
In comparing the wild-type and mutant proteins, they found the only function it was compromised for was the ability to act as a DNA helicase, unwinding DNA strands to facilitate entry of DNA synthesis enzymes.
“This is the first demonstration that it might be possible to use phosphorylation as a ‘switch’ on a DNA helicase to dial down replication speed,” explains Melendy.
Evidence suggests a similar phosphorylation can occur in human DNA helicases as well.
“For many cancers, if we selectively inhibit the DNA damage checkpoints they still retain, and simultaneously treat with lower than normal amounts of DNA-damaging chemotherapeutics, then we might be able to selectively damage cancer cells while leaving non-cancerous cells intact, greatly enhancing cancer cell killing while simultaneously reducing toxic side effects.”
This is an ongoing area of study by the UB researchers with their collaborators at Roswell Park Comprehensive Cancer Center.
Based on the current study, these DNA damage checkpoints may now be relevant to treating viral infections of the small DNA viruses under investigation at UB.
“Because they rely almost exclusively on host cell enzymes to synthesize their viral genomes, these small DNA viruses have been very resistant to anti-viral therapeutics,” says Melendy. “We currently have no antiviral treatments for HPV or polyomaviruses. By triggering the DNA damage response in a patient, this could dramatically slow viral DNA replication, suppressing the infection, providing us with a novel avenue for possible antiviral treatments of these as-of-yet untreatable viral infections.”
Shichen Shen and Jun Qu, both of the School of Pharmacy and Pharmaceutical Sciences, are co-authors. The work was funded by the National Institutes of Health and the NIH Training Grant in Microbial Pathogenesis.
Introduction, materials and methods, data availability, supplementary data, methods for constructing and evaluating consensus genomic interval sets.
Julia Rymuza, Yuchen Sun, Guangtao Zheng, Nathan J LeRoy, Maria Murach, Neil Phan, Aidong Zhang, Nathan C Sheffield, Methods for constructing and evaluating consensus genomic interval sets, Nucleic Acids Research , 2024;, gkae685, https://doi.org/10.1093/nar/gkae685
The amount of genomic region data continues to increase. Integrating across diverse genomic region sets requires consensus regions, which enable comparing regions across experiments, but also by necessity lose precision in region definitions. We require methods to assess this loss of precision and build optimal consensus region sets. Here, we introduce the concept of flexible intervals and propose three novel methods for building consensus region sets, or universes: a coverage cutoff method, a likelihood method, and a Hidden Markov Model. We then propose three novel measures for evaluating how well a proposed universe fits a collection of region sets: a base-level overlap score, a region boundary distance score, and a likelihood score. We apply our methods and evaluation approaches to several collections of region sets and show how these methods can be used to evaluate fit of universes and build optimal universes. We describe scenarios where the common approach of merging regions to create consensus leads to undesirable outcomes and provide principled alternatives that provide interoperability of interval data while minimizing loss of resolution.
Advancements in high-throughput sequencing technologies have resulted in a vast amount of diverse epigenomic data that has given us tremendous insight into genome function. Epigenomic data are often summarized into genomic region sets stored in BED files. Through the work of hundreds of individual labs and projects such as ENCODE ( 1 ), the NCBI Gene Expression Omnibus ( 2 ) now contains almost 100 000 BED files ( 3 ). The volume of data has made integration challenging.
For an analysis that spans several genomic interval sets, one of the first steps is to define a consensus region set, or region universe , upon which the diverse sets can be interpreted ( 4–10 ). Such universe region sets have many common practical use cases. For example, they define genomic intervals for differential peak analysis ( 11 ); they form the regions of interest in a count matrix in single-cell epigenome analysis ( 11 , 12 ); they are used as a background for statistical region enrichment analysis ( 13–16 ); and they are a region vocabulary in vector representation approaches ( 17–20 ).
For many tools, the choice of universe is critical. It defines the features to which data will be projected. Currently, there are different ways of choosing a universe for analysis. Simple approaches include tiling the genome into fixed-size bins ( 12 , 15 ), or using intersection or union operations on a collection of region sets ( 11 ). Some methods have been developed to create better-fitting universes for specific downstream use cases ( 21 , 22 ). An alternative is to use a predefined universe from an external source; for example, the ENCODE consortium curated a registry of candidate cis-regulatory elements accessible through the SCREEN webserver ( 1 ), and the Ensemble Regulatory Build is a central, reusable source of regulatory region definitions ( 23 ). The choice of universe matters because universes can be a poor fit to data, and if a universe does not fit the data well, it can lead to incomplete or incorrect results ( 5 , 14 , 15 ). However, despite the importance of selecting a universe, it is often done ad hoc , and there are few approaches to assess the fit of a universe to a collection of region sets.
Here, we address these limitations by introducing novel concepts for constructing and evaluating region universes. First, we introduce the idea of flexible genomic intervals, which represent region boundaries by intervals instead of points, allowing us to summarize many fixed regions into fewer flexible regions without loss of information. Next, we propose three methods for constructing flexible region universes: a coverage cutoff universe, a maximum likelihood universe, and a Hidden Markov Model (HMM). Finally, we propose three methods to evaluate the fit of a universe to a collection of region sets: (i) the base-level F 10 -score; (ii) a Region Boundary Distance score ( RBD ); and (iii) a likelihood model score that assesses the likelihood that the proposed universe was drawn from the given distribution of region sets.
To assess our universes and evaluation methods, we compared our methods against alternatives and predefined universes. We show that flexible universes can capture information from complex data collections into one well-defined universe. Moreover, we show how our assessment metrics provide complementary measures of assessing universe fit and we prove the relevance of these measures. We show that the union universe has many downsides and propose the HMM universe as a generally useful approach for defining well-fit universes. To demonstrate how these universes could affect downstream analysis, we conclude with an application of region set enrichment analysis, where we show how the results are affected by choice of universe. Overall, our results demonstrate the importance of considering region universe and provide promising new tools to construct better-fitting universes for a variety of use cases.
To integrate genomic interval data, we first require a consensus set of intervals, or a universe. We may select a predefined universe from an external source or define one from a collection of input region sets using a consensus algorithm (Figure 1A ). With a universe in hand, we can then ‘tokenize’ the original regions (Figure 1B ). Tokenization redefines them into universe regions, normalizing differences in region boundaries to transform similar regions into a single representation. The most basic tokenization method is simple interval overlap. This approach works well if a universe approximates the original data well; otherwise, this may result in loss of precision. A universe may not be a good fit to a collection for a variety of reasons (Figure 1C ); for example, (i) a region can be shifted; (ii) two neighboring regions may be merged, making them indistinguishable in downstream analysis ( 5 ); (iii) a universe may omit important intervals, leading to loss of information ( 5 ); or (iv) a universe may contain extraneous regions that do not reflect genome coverage, adding noise and compute time ( 14 ). If a universe is a poor fit, it can affect downstream analysis negatively; for example, a differential accessibility analysis wouldn’t even test a locus that had been dropped from the universe. It could also miss a significant locus if it were merged with an abutting locus that lacked differential signal. Or, a motif analysis based on universe regions that had been shifted could miss enriched sequence features that were present in the part of the region left out of the universe.
Overview of the concept of a universe. ( A ) A consensus algorithm takes a region set collection |$\mathbb {R} = [\mathcal {R}_1, \mathcal {R}_2, ...]$| and builds a consensus representation called universe |$\mathcal {U}$| . ( B ) We ‘tokenize’ raw regions into universe |$\mathcal {U}$| by redefining them as universe regions, creating a more uniform collection. ( C ) Universes may poorly represent region sets by shifting, merging, dropping or adding extraneous regions.
To address these issues, we developed three new approaches for constructing a universe that is a good fit to the original data: first the ‘coverage cutoff universe’; second, the ‘maximum likelihood universe’; and finally, an ‘HMM universe’. To assess them, we also developed three universe fit metrics. Finally, we applied these to a variety of real datasets.
The coverage cutoff universe.
The simplest example of a universe built from the data is a ‘union universe’, in which a collection of region sets is merged. This method is often done for differential analysis of ATAC-Seq data ( 11 ). The union universe by definition covers all bases from the original collection; however, it can also lead to very large regions, particularly if the number of input region sets is large. Another simple alternative is using an intersection operation, which would include only bases covered in every region set in the collection, but this has the opposite problem: it leads to very sparse universes.
We reasoned that a hybrid approach may achieve a better result. First, we conceptualize a collection of region sets as a coverage signal track across all input region sets. Then, similar to a peak calling approach, we choose a cutoff x such that universe includes only positions with coverage greater or equal to x (Figure 2A ). Setting the cutoff to one corresponds to a union universe, and setting the cutoff equal to the number of input region sets corresponds to an intersection universe. Setting a cutoff in between the two balances these extremes and provides a tunable parameter that may be adjusted depending on the needs of downstream tasks. We call the resulting universe a coverage cutoff (CC) universe. A principled approach to selecting the cutoff is to use a simple likelihood model that calculates the probability of appearing in a collection. With this model, we can calculate an optimal cutoff according to Eq. ( 1 ) (see Supplementary methods for details).
Different approaches to building universes. ( A ) Coverage-based universes are derived from the genome coverage of a collection of region sets. Examples include intersection |$\mathcal {U}_{int}$| , coverage cutoff |$\mathcal {U}_{CC}$| , and union universe |$\mathcal {U}_{union}$| . ( B ) A flexible region in contrast to fixed region can represent boundaries of many variable regions. ( C ) The flexible coverage cutoff (CCF) universe is based on coverage of the genome by a collection. It uses two cutoff values: the lower defines flexible boundaries and the upper defines the region core. ( D ) A collection of genomic region sets is aggregated, and region starts, core (overlap), and ends are counted, creating signal tracks. ( E ) Maximum likelihood universe is derived from three signal tracks. Using a likelihood model, we build a scoring matrix that assesses the probability of each position being a given part of a flexible region. Next, we find the most likely path, which represents the maximum likelihood universe. ( F ) The HMM universe treats signal tracks representing genome coverage by different parts of a region as emissions of hidden states that correspond to different parts of flexible regions.
Here, S c is a sum of genome coverage by collection and g is the size of the genome.
We realized that, in a sense, the CC universe is a point estimate of a more complex distribution of possible universes. We reasoned that we may gain some insight by modeling the boundaries of the consensus regions as intervals, rather than points. To do this, we developed a new concept of a genomic region we call a flexible region . In contrast to fixed interval that is defined by two fixed points ( start and end ), a flexible interval is defined by four ( start start , start end , end start and end end ) (Figure 2B ). A flexible interval can model many region variations into one well-defined flexible interval.
A simple approach to constructing a flexible region universe is to define two cutoff values instead of one: the looser represents the cutoff for boundaries, and the stricter for the region core (Figure 2C ). This way, positions with coverage between those two points will be assigned to flexible region boundaries and positions with coverage higher than the second cutoff will be assigned to core of the region. Using this idea, we built a confidence interval around the optimal cutoff value, which extends the CC universe into the coverage cutoff flexible (CCF) universe.
While flexible intervals more naturally represent collections of overlapping region sets than fixed intervals, we reasoned that they still suffer from the possibility of merging neighboring regions when collections are large. To address this issue, we need information about not just the coverage of regions in a collection, but also the region start and end positions. To compute this information, we developed a fast plane-sweep algorithm (see Supplementary methods for details). With this tool we can quickly and efficiently calculate three tracks representing aggregate start, end, and coverage values of a region set collection at base-pair resolution (Figure 2D ). We reasoned that a model that could incorporate all of these signals may improve universe resolution.
We can conceptualize a flexible universe as a path through the genome that assigns either start, core, or end state to each position. Using a universe scoring model and optimization algorithm, we can build the best path through the genome (Figure 2E ). As a scoring model, we next developed a complex likelihood model, an extension of the simple likelihood model introduced earlier for CC universe, which considers not just the coverage (core), but also in region start and end signal tracks. This model describes for each position the probability of it being a region start, core, or end. We thus build the maximum likelihood universe (LH) in 3 steps: (i) compute the three signal tracks; (ii) use a likelihood model to build a scoring matrix; (iii) find the maximum likelihood path through the genome (see Supplementary methods for details).
The maximum likelihood universe provides a simple and principled model for optimal flexible universes. However, a disadvantage is that it provides no tunability, since the likelihood scores are determined purely from the data. We reasoned that this may lead to results depending on input collection. To address this, we sought a more tunable model using a Hidden Markov Model (HMM).
An HMM models a hidden processes using (i) a matrix of transition probabilities between hidden states and (ii) emission probabilities of observations from hidden states. In our model, there are three observed sequences: the number of starts, overlaps, and ends at a given position. The hidden variable corresponds to the different parts of the flexible segment (Figure 2F and Supplementary Figure S1 ). We can tune transition probabilities, which can be chosen in a way that will prevent unnecessary segmentation, and emission matrix, which describes the relationship between observations and hidden states (see Supplementary methods for details).
Having developed several new approaches to construct universes, we next sought to evaluate these universes and compare them to other common approaches. Because the choice of universe can dramatically affect downstream analyses, it is important to choose a universe deliberately. However, there are no well-established methods for assessing universe fit to data. Furthermore, different analyses may be better served by a different types of universe, indicating that there really is no generally optimal universe, but the idea of what makes a ‘good’ universe depends on the downstream analysis. For example, a differential accessibility analysis should prioritize sensitivity over specificity; in this case, it is not a major problem if the universe includes many regions present in only a few samples, since the cost of extra comparisons is lower. On the other hand, the cost of excluding a region that could be a significant differential locus would be high. In contrast, a word-based deep learning task that trains a model with input dimensions equal to the size of the universe may elect a more specific universe, at the cost of discarding some regions that are present in a few of the samples, because otherwise the training could be intractable. Thus, the question of universe optimality depends on the use case, and therefore, we require methods of evaluating universe fit that can be tuned to a research question. This problem is similar to the comparison of two generic interval sets, for which several methods have been developed ( 24 ), with two key differences: first, we want to compare a universe region set not only to one other region set, but to a collection of them; and second, the question is not symmetric: it is generally more important that a universe not miss information (regions), even at the cost of some extra regions – and the desired level of asymmetry can vary. Therefore, we developed three methods for assessing universe fit to data: (i) a base-level overlap score; (ii) a region boundary distance score and (iii) the universe likelihood.
Our first metric is based on base coverage. We consider the universe as a prediction of whether a given genomic position is present in a given region set from the collection. Treating each region set from the collection as a query, we can then conceptualize matches and mismatches as true positives (covered in both universe and query), false positives (covered in universe, but not in query), or false negatives (covered in query, but not in universe) (Figure 3A ). This allows us to calculate common classification evaluation measures such as precision and recall. Precision counts the number of true positives, so a low precision indicates presence of unimportant positions in the universe; recall measures how much of the universe is in a query. To combine precision and recall, we use the F 10 -score, a weighted version of the traditional F -score that pays 10 more times attention to recall than precision. This asymmetry captures our goal to prioritize sensitivity over specificity: by prioritizing recall, we indicate that it’s better to have a few extra, noisy regions that to exclude something important. The F 10 -score results in values between 0 and 1, with a perfect fit approaching 1. An alternative approach to base-level overlap score would be Jaccard similarity, however it does not account for asymmetry of the comparison.
Different approaches to assess how well the universe represents the data. ( A ) The base-level overlap measure considers the universe as a prediction of a region set and based on that it calculates number of false positives (FP), true positives (TP), and false negatives (FN), and from that derives recall ( R ) and precision ( P ), which are combined into the F 10 -score. ( B ) The region boundary distance (RBD) score assesses how well a universe represents start and end positions, by calculating distance from region set to universe, and from universe to region set; those two metrics are combined into a region boundary score by calculating their reciprocal, weighted harmonic mean. ( C ) Likelihood assessment uses a likelihood model based on signal tracks representing genome coverage by different parts of a region to calculate universe likelihood as a combination of likelihoods of all three signals tracks. D) A complete analysis example comparing a collection of region sets against 4 proposed universes: |$\mathcal {U}_1$| a precise universe, |$\mathcal {U}_2$| a sensitive universe, |$\mathcal {U}_3$| a fragmented universe, and |$\mathcal {U}_4$| well-fit universe. For each universe, all 3 metrics are calculated. The F 10 -score and RBD score assess individual region sets. The final score for a collection is their average. In contrast, the likelihood is calculated directly for the whole collection.
One disadvantage of the base overlap score is that it is unaware of region boundaries. A universe region that covers two abutting query regions would get a perfect score. This can be highly problematic in downstream applications; for example, in a differential analysis, lumping two distinct loci together could dilute differential signal. To address this, we sought a measure that would consider region starts and ends (Figure 3B ). We calculate the distance between each boundary of each region in the query and the closest corresponding boundary in the universe. Universes with boundaries that are near the query boundaries would have shorter distances, indicating better universe fit. However, highly fragmented universes with many unnecessary boundaries would have very small distances from query to universe. To account for this, we also calculate the inverse distance: from boundaries in universe to the nearest boundaries in the query. Finally, we combine those two metrics into a region boundary distance score ( RBD ) by taking their reciprocal, weighted harmonic means. With this score we describe the universe’s ability to conserve information about starts and ends, with a score of one representing a perfect representation of boundary locations. However, we do not incorporate any information about collection coverage in this score.
For fixed universes, the start and end point are well-defined, however for flexible regions they are intervals. Therefore, for flexible universes, we modify the RBD score to set distance equal to zero if a boundary query region is inside the universe’s boundary interval.
Finally, we sought a metric that incorporates both information about region boundaries as well as genome coverage. We propose here a universe likelihood score (Figure 3C ). We first calculate three signal tracks representing genome coverage by start, core, and end of the regions in the collection. Then for each signal track we make a separate model, which results in three separate models for different parts of a region. Each of these models describes the probability of a given position being a given part of a region, depending on the signal strength (see Supplementary methods for details). That results in a complex, probabilistic description of a region set collection. Next, we use this model to calculate the likelihood of the universe, which we can compare between universes. We make two versions of likelihood calculations, one suited for fixed universes and one for flexible universes. We use log likelihood, so our values range from minus infinity (low) to zero (high). Finally, to increase interpretability, we normalize the scores by subtracting the likelihood of an empty universe (one that contains no regions). Thus, a positive final score reflects a given universe that is more likely than having no regions at all, while a negative score means the universe is less likely than the empty universe.
To accommodate flexible universes, we adapted the likelihood score by calculating the boundary likelihood as if the whole flexible interval could contain a boundary position, rather than a single fixed point (see Supplementary methods for details).
Having developed three assessment methods, we can use them to compare competing universes to assess which universe is the best fit for a collection of region sets. We do this by computing the scores for each universe and comparing among universes (Figure 3D ). The scores assess different aspects of the universe fit; the F 10 -score promotes sensitive universes over specific ones, RBD score penalizes sparse universes, and likelihood provides complex universe assessment. Although the likelihood score incorporates information about boundaries as well as how well the universe covers the collection, it can penalize sensitive universes because it is not intentionally biased toward asymmetry the way the previous scores are. Thus, by computing all these scores, we reason that we get a complete picture that can guide decisions for selecting a universe for a collection of region sets.
Next, we developed an evaluation strategy to test our universe building and assessment methods on real data. We assembled five diverse collections of region sets representing different biological problems (Figure 4A ): (i) CTCF ChIP small , a small random collection of CTCF region sets ( n = 40) from the ENCODE database ( 1 ); (ii) CTCF ChIP large , CTCF ChIP-seq datasets ( n = 877); (iii) TF ChIP , ChIP-seq experiments for diverse transcription factors (TFs) ( n = 8503); (iv) B-LCL ATAC , a small set of ATAC-seq files from B-Lymphoblastoid cell lines (B-LCL; n = 400) from ChIP-Atlas ( 25 ); (v) a Random ATAC , random ATAC-seq results ( n = 5000). These datasets vary in data type, collection size, and level of heterogeneity of input regions across region sets.
Overview of evaluation approach. ( A ) Five collections representing different biological problems used for assessment. ( B ) For each collection, we compared it to five data-driven universes and three predefined universes. The data-driven universes are tailored to the input collection, but the predefined universes do not vary by collection. ( C ) We assessed the fit of each universe to each collection using our three assessment methods.
We also assembled universes to assess. First, we obtained three universes that do not depend on analyzed data, which we call predefined universes: (i) the tiles universe, which bins the genome into non-overlapping 1000 bp tiles; (ii) the SCREEN universe, which consists of predefined cis-regulatory elements from ENCODE ( 1 ); and (iii) the Regulatory Build (RB) universe, consisting of pre-defined regulatory elements from Ensembl ( 23 ). In addition to these three external universes, we also built 5 data-driven universes that are specific to each region set collection. These include: (i) the union universe; (ii) the CC universe; (iii) the CCF universe; (iv) the LH universe; and (v) the Hidden Markov Model (HMM) universe. This led to 28 universes and 40 pairwise comparisons of universe-to-collection (Figure 4B ). For each comparison, we computed our three assessment methods (Figure 4C ). This gives us a comprehensive evaluation of both externally sourced and data-driven universes, tested on diverse query region set collections.
To explore the differences in our five region set collections, we first computed general coverage statistics ( Supplementary Table S1 ). The smallest region set, B-LCL ATAC , contains ≈700 000 regions and covers 0.2% of the genome, whereas the largest, TF ChIP , contains ≈1.5 billion regions and covers 91% of genome. We also observed that the ATAC-seq collections have smaller regions on average than the ChIP-seq collections.
The universes also have very different characteristics, with some requiring additional filtering based on region likelihood and size (see Supplementary methods for details, Supplementary Figures S2 and S3 ). For example, for the CTCF ChIP large collection, the eight universes have different levels of precision and fragmentation (Figure 5A ). The universes differed in average region size, number of regions, and percent of genome covered ( Supplementary Figure S4 , Supplementary Table S2 ). Having assembled the universes, we next computed our three assessment methods.
Universes overview and results of base-level overlap score. ( A ) Example of universes assessed for the CTCF ChIP large collection, including the 3 constant external universes, and 5 data-driven universes built from the input collection. ( B ) Different universes represent genome coverage by the collection to a different extent, example from the Random ATAC collection. Collection |$\mathbb {R}$| consists of many different files, which are represented by the core signal track. Regions in |$\mathcal {R}_1, \mathcal {R}_2, \mathcal {R}_3$| are best represented by CC, CCF and LH universes in terms of overlap. ( C ) Precision and recall distribution for each collection and universes assessment. ( D ) Average F 10 -score for each collection and universes assessment.
Region sets in a collection can differ widely; our first assessment method assesses fit by quantifying the degree of overlap between each region set and the universe. In example data from the Random ATAC collection, we observe that some universes cover many bases present in only few of the collection’s region sets, while other universes are more stringent (Figure 5B ). To assess this globally, we first computed precision and recall for each comparison (Figure 5C ). We observed that the tiling universe and union universe both have perfect recall for all tested collections, consistent with how these universes are constructed; the tiling universe covers the entire (mappable) genome, and the union universe by definition covers every base contained in the collection. In contrast, recall is lower for the more stringent data-driven universes; the CC, CCF and LH universes exclude positions with low coverage, especially for large collections; the HMM universe has higher recall in general, indicating that it contains most positions covered by the collection. Finally, the lowest recall scores are assigned to the external universes, SCREEN and RB, which is consistent with these universes being built from other data sources. This highlights an advantage of building bespoke universes tailored to a collection: recall is superior. On the precision side, the worst performer overall is the tiles universe, consistent with many tiles in the universe that do not reflect coverage in the collection. In contrast, SCREEN and RB had generally higher precision, especially for ChIP-Seq collections.
In general, data-driven universes tend to represent collections well with good precision and recall. This is most apparent for the B-LCL ATAC collection for which data-driven universes are much better than predefined. For likelihood universes (CC, CCF, and LH universes) built from large ChIP-Seq collections ( CTCF ChIP large and TF ChIP ) we observe the worst recall among data-driven universes. This is the consequence of Eq. ( 1 ), from which we observed that, for ChIP-Seq collections, the optimal cutoff value is higher. On the other hand, both the HMM universe and the union universe have high recall and low precision. In general, we see that universes with high recall have lower precision, reflecting the delicate balance between including complete information in universes without adding too much noise.
To propose a balance between precision and recall, we next calculated the F 10 -score, which assigns more weight to recall than precision (Figure 5D ). For predefined universes, we see that the tiles universe scores well for ChIP-seq collections, which cover more of the genome, but poorly for ATAC-seq collections, which have lower coverage. In general, both RB and SCREEN are outperformed by data-driven universes, especially for the B-LCL ATAC collection, for which they contain too much noise. In general, for ChIP-Seq collections, the union universe is the best for these metrics, consistent with the weighting we chose that gives 10 times the weight to recall. Interestingly, for the TF ChIP collection, the HMM universe outperforms likelihood universes; on the other hand, for ATAC-Seq collections, the CCF universe outperforms both union and HMM universes. Overall, we conclude that computing precision, recall, and F 10 provide useful insight into assessing universe fit. They provide a way to quantify the advantages of a data-driven universe and assess how much information is lost by an external universe.
Next, we sought to address the major weakness we see in the base-level overlap score: that it does not consider region boundaries. Assessing boundaries is important because of how it affects downstream analysis. If two regulatory elements with distinct behavior are merged into a single region in a universe, then all downstream analyses will essentially evaluate the average of the two signals. But different universes have different sensitivities to boundary points (Figure 6A ); for example, anecdotally, the union universe is not sensitive at all: it contains few boundaries, especially for larger collections. It will clearly merge together many neighboring regions, even if they have distinct patterns across input sets. The CC and CCF universes are more sensitive but still miss out on many boundaries for bigger collections; the LH universe is very sensitive to boundaries, but has other weaknesses (it tends to exclude positions with low coverage); all of those problem are solved with HMM universe, which is very sensitive and also is able to represent regions with low coverage. To assess boundaries globally, we turned to the region boundary distance score. First, for each region set, for each region, for each boundary, we calculated the distance to the nearest corresponding universe boundary (See Methods; Figure 6B ). For all collections, the distance from collection to RB universe was very high. Similarly, the union universe performs very poorly in this metric, particularly for larger collections, consistent with intuition that the union regions lose boundary precision as the number of regions increases. We also computed the inverse: distances from query to universe (Figure 6C ). Interestingly, for ATAC-Seq collections, we observe a small distance from query to universe but a high distance from universe to query. This indicates that all universes have many boundaries that are not present in the raw data, but at the same time boundaries present in the queries are well-reflected by universes.
Results of region boundary distance score. ( A ) Different universes represent region boundaries to a different extent. Three signal tracks provide summarized description of the whole collection. Both LH and HMM universe are most sensitive to region boundaries. ( B ) Distribution of median distances from query to universe. ( C ) Distribution of median of distance from universe to query. ( D ) Average RBD score for each collection and universe comparison.
To summarize both distance directions, we calculated their reciprocal, weighted harmonic mean, the Region Boundary Distance score ( RBD ) (Figure 6D ). The average RBD score shows that the RB universe is a very poor fit to all collections, likely because it contains few, large regions. On the other hand, tiles universe has similar scores for all collections, which is good for big ChIP-Seq collections compared to other universes, but bad for ATAC-Seq collections. Interestingly, the SCREEN universe seems to be a good fit for all collections, and the best fit for ATAC-Seq collections, even outperforming the data-driven ones for this metric. Among data-driven universes, the HMM universe performs the best, with LH in second place for all collections except B-LCL ATAC . However, for this collection all data-driven universes have similar scores. Coverage-based universes (CC, CCF) perform well for ATAC-Seq collections, but not for large ChIP-Seq collections ( CTCF ChIP large and TF ChIP ) compared to other data-driven universes. As expected, the RBD score reflects the poor performance of the union universe for large ChIP-Seq collections; the merging leads to poor reflection of interval boundaries.
Finally, we calculated the likelihood score for each comparison (Figure 7A ). In general, predefined universes are a worse fit to the collections than an empty universe, with exception of SCREEN for CTCF ChIP large and TF ChIP collections. Among data-driven universes, the union universe performs very poorly, achieving negative scores for all collections. As expected, CC, CCF, and LH universes outperform the empty universe for all collections, as these were designed to optimize likelihood in some way. The HMM universe performs well overall; however, it is worse than empty universe for CTCF ChIP large and TF ChIP collections. A more detailed look reveals that the low HMM likelihood scores in these scenarios are driven by region coverage tracks, not boundaries, suggesting that for these collections, our current HMM parameterization may yield a universe with too much noise (see Materials and Methods; Supplementary Figure S5 ).
Results of universe likelihood and comparison between fixed and flexible scores. ( A ) Likelihood of each universe given collection. ( B ) Change of RBD score when we account for flexibility. ( C ) Change of likelihood, when we account for flexibility.
So far, our assessments have not taken into account that some universes can be flexible. We believe that flexible universes provide several advantages, and sought to assess them. Flexible intervals don’t quite fit into the standard 3-column BED format; however, they can be stored using optional fields of an extended BED format. In this approach, sequence name, start, and end of the flexible region are represented by the first three columns, and thickStart and thickEnd columns hold the information about end of the flexible start and start of the flexible end. To assess this, we applied our flexible-aware version of the RBD score. We observed that RBD score improves for all flexible universes (Figure 7B ). The change is less significant for CCF universe; however for LH and HMM universes, the new score is close to one, with the HMM universe performing slightly better.
We computed a version of the likelihood score that considers universe flexibility. This score shows a significant improvement; for all flexible universes, all collection scores change by an order of magnitude (Figure 7C ). Likelihood values that consider flexibility are similar for all universes; however, the HMM universe performs slightly worse for large ChIP-Seq collections ( CTCF ChIP large , TF ChIP ). This reflects that, unlike the CCF and LH universes, the HMM universe does not explicitly optimize likelihood. A more detailed look into likelihood showed that although the HMM universe has the best likelihood of cores of the regions, it performs less well for boundary positions (see Methods; Supplementary Figure S6 ).
Since each metric assesses different aspects of universe fit, considering them independently limits the scope of assessment. For a holistic view, we summarized the scores of each metric into normalized heatmaps, allowing comparison within and across metrics (Figure 8 ). We observed several informative cross-score patterns: First, the F 10 -score is the most consistent metric across universes, indicating that all these universes cover the collections to similar extent (Figure 8A ). In contrast, the RBD indicates more variation in how well universes represent region boundaries (Figure 8B ). This is consistent with our intuition that matching boundaries is a more difficult task, since it requires ensuring large regions are split well. The disparity between F 10 -scores and RBD score demonstrates their combined utility. For example, the union universe has perhaps the overall best F 10 -scores across universes, but has low RBD score. Inversely, the SCREEN universe has the best RBD score for the B-LCL ATAC collection, but is significantly worse than any data-driven universe for F 10 -score. In likelihood scores, the CC, CCF, LH universes outperformed other universes to similar extent, while tiles, RB and union were universally poor fits for all universes (Figure 8C ). The HMM has much better RBD score than CC, but a worse likelihood, reflecting that likelihood considers more than boundary positions. Additionally, the likelihood is stricter in boundary assessments: while for RBD score we take a median of actual values, for likelihood we use probability of a given position being a boundary. Comparison of flexible and fixed versions of RBD score and likelihood highlights the value of flexible regions (Figure 8D , E ); RBD score for LH and HMM improved significantly after accounting for flexibility.
Results of universe comparison using different scores. ( A ) Row normalized F 10 -score of each universe given collection. ( B ) Row normalized RBD score of each universe given collection. ( C ) Row normalized likelihood of each universe given collection. ( D ) Row normalized flexible version of RBD score of each universe given collection. ( E ) Row normalized flexible version of likelihood of each universe given collection.
To demonstrate how universe affects downstream analyses and how our universe building and evaluation methods can be applied, we performed a region set enrichment analysis using LOLA, a tool for statistical region enrichment analysis ( 15 ). The goal is to take some demo region sets and then use them to search a database of region sets to find similar regions, and explore how the choice of universe affects the results. We constructed two different experiments. For our first experiment, we used the Random ATAC collection as a database. We used three predefined universes – tiles, RB and SCREEN – as well as data-driven universes built from the Random ATAC and B-LCL ATAC collections. To see how the universes based on rare cell types perform, we also added data-driven universes built from fifteen Glia ATAC-Seq files. We queried the database with fifteen files that were not present in the database and were not used for universe construction, representing three different cell types: five files from A549, five from B-LCL, and five from Glia samples. For our second experiment, we used 623 ChIP-Seq files representing different TFs to build a database and data-driven universes. We queried the database with thirty files representing different TFs: 10 files from EZH2, 10 files from POLR2A, 10 files from YY1. For both experiments, we assessed performance of the region enrichment analysis with R -precision (rPPV), which measures the precision based on the top R results, where R is equal to the number of correct results in the whole database.
Our results demonstrate clear impact of the universe on analysis performance (Figure 9A ). In the first experiment, the data-driven universes were the best performers for the specific questions; Glia ATAC queries (gray dots) performed best under the tailored Glia data-driven universes, followed by the Random ATAC data-driven universes, and performed poorly with any pre-defined universes or with the B-LCL data-driven universes. The B-LCL queries (yellow dots) also performed best with the tailored B-LCL-driven universes, and performed reasonably well with predifined or Random ATAC data-driven universes, but poorly with the Glia data-driven universe. Finally, A549 samples performed equally well with predefined or Random ATAC data-driven universes, and poorly with the universes built on the other data types. This shows that using universes based on a specific cell type increase the performance of downstream analysis for this type. Our second experiment, based on TF data, shows a clear difference between the union universe and other more complex data driven universes. We also see an imbalance among the predefined universes, with the SCREEN universe outperforming tiles and RB in general. In conclusion, for ATAC-Seq data, rPPV is higher for data driven universes than predefined ones; for TF data rPPV is higher for more complex data-driven universes (CC, CCF, ML, HMM) than union universes. Overall, these results demonstrate that choosing the right universe is a complex task. In our experiments, the more complex data-driven universes (CC, CCF, LH and HMM) are always either comparable or superior to simpler data-driven or pre-defined universes, although the exact performance depends on the both the initial data and the downstream task. Most importantly, this analysis suggests that the assessment methods we present can be helpful in choosing the right universe.
Results of downstream enrichment analysis depends on universe. ( A ) R -precision (rPPV) of query files depending on universe. ( B ) Median of different universe, depending on the collection used for their construction.
Many integrative epigenome analyses require the data to be defined on a set of consensus regions, or universe. This universe is critical for analysis because it determines the precision of regions assessed. Our experiments highlight how different universes can have different levels of fit to collections and may therefore be useful for different tasks. Despite the importance of this choice, few approaches have been developed to aid analysts in building appropriate universes or assessing the fit of an existing universe. In this study, we have addressed these issues by presenting several novel methods to build universes from collections of region sets, as well as new ways to assess the fit of a universe to a collection of region sets. We also introduced the concept of flexible segments, and proposed several methods for constructing universes that can use either traditional fixed boundaries or flexible interval boundaries.
In general, data-driven universes outperformed predefined ones. However, the data-driven universes also have a weakness: by definition, they change with the underlying collection, and therefore cannot support an integrative analysis that spans collections. If results need to be compared across collection, then a shared universe is required. There are a few options for analysis in this case, all of which are facilitated by our work: First, a custom data-driven universe that spans all included collections can be built. Since we have described several ways to create well-performing data-driven universes, it would be easy to just design a bespoke universe for a given comparative analysis. However, this may not always be possible or convenient, and at some point, an external universe may be preferred. In this case, a trade-off is required: fit of the universe must be sacrificed to increase interoperability with other collections. Our assessment methods now provide a principled way to assess this trade-off and inform research decisions.
Among the non-data-driven universes we tested, SCREEN performed well for the collections we tested. However, there are almost certainly other collections or use cases for which the tiles, RB, or other external universes would be a better choice. Along these lines, we propose that our methods for building universes can be used in the future to create predefined universes from large collections, thereby creating even better global universes that can be re-used for integrative analysis. In the future, a centralized repository of universes, built using different methods and for different target use cases, could be a useful resource; a given collection of region sets could be represented into different universes based on the balance of fit and need for integration.
Based on our results, we propose that the union universe, though widely used, does not represent ChIP-Seq data well, particularly for large collections. Instead, we propose the HMM universe as a good all-around option that solves many of the issues with simpler universes. It has the highest sensitivity to boundary positions and good recall. It also provides adjustable parameters; by setting emission and transition probabilities, users may adjust model sensitivity and keep it consistent across collections. Still, the final choice of universe should consider the needs of downstream analysis. Our results show that assessing universe fit is a complex question, with many features to optimize. Even with our assessment metrics, it is difficult to claim an optimal universe for a given collection; instead, the answer depends on the downstream analysis priorities. For example, in general, the CCF and LH universe represent properties of a whole collection well, but they exclude infrequent regions; thus, they may be useful for NLP analysis of the genome but could lead to losing information about rare cell types in single cell analysis. In contrast, the union universe by definition covers all bases found in the region set collection, and therefore has a great F 10 -score; however, it also merges regions, which is reflected by poor RBD score and likelihood scores, indicating that it would not be a good fit for an application that requires high region resolution. Therefore, multiple perspectives must be considered for a holistic assessment of universe fit.
One advantage of our new universe construction methods is that they naturally create flexible universes. Flexible universes are a new way of summarizing information from large collections with less loss of information. We showed that proposed approaches of making flexible universes improve results over inflexible universes. We see flexible regions as a powerful new concept that can modify our current way of thinking about universes. Furthermore, we expect that using them for differential peak analysis, statistical region enrichment analysis and NLP approaches has potential to improve results. Flexible regions will become more useful as we and others develop the necessary tooling to work with them; for example, we will require tokenization methods that can project a traditional region set into a flexible universe quickly and accurately, while considering the universe flexibility.
In conclusion, this research provides new concepts, methods, and insight that will help researchers to determine the best analysis path for many types of genomic region analysis.
Software is available at https://github.com/databio/geniml .
Supplementary Data are available at NAR Online.
National Human Genome Research Institute [R01-HG012558]; National Institute of General Medical Sciences [R35-GM128636]. Funding for open access charge: NIH.
Conflict of interest statement . N.C.S. is a consultant for InVitro Cell Research, LLC.
Moore J.E. , Purcaro M.J. , Pratt H.E. , Epstein C.B. , Shoresh N. , Adrian J. , Kawli T. , Davis C.A. , Dobin A. , Kaul R. et al. . Expanded encyclopaedias of DNA elements in the human and mouse genomes . Nature . 2020 ; 583 : 699 – 710 .
Google Scholar
Barrett T. , Wilhite S.E. , Ledoux P. , Evangelista C. , Kim I.F. , Tomashevsky M. , Marshall K.A. , Phillippy K.H. , Sherman P.M. , Holko M. et al. . NCBI GEO: Archive for functional genomics data sets—update . Nucleic Acids Res. 2013 ; 41 : D991 – D995 .
Xue B. , Khoroshevskyi O. , Gomez R.A. , Sheffield N.C. Opportunities and challenges in sharing and reusing genomic interval data . Front. Genet. 2023 ; 14 : 1155809 .
Kruczyk M. , Umer H.M. , Enroth S. , Komorowski J. Peak finder metaserver - a novel application for finding peaks in ChIP-seq data . BMC Bioinformatics . 2013 ; 14 : 280 .
Lun A.T.L. , Smyth G.K. De novo detection of differentially bound regions for ChIP-seq data using peaks and windows: Controlling error rates correctly . Nucleic Acids Res. 2014 ; 42 : e95 .
Lun A.T.L. , Smyth G.K. Csaw: a bioconductor package for differential binding analysis of ChIP-seq data using sliding windows . Nucleic Acids Res. 2015 ; 44 : e45 .
Smith J.P. , Sheffield N.C. Analytical approaches for ATAC-seq data analysis . Curr. Protoc. Hum. Genet. 2020 ; 106 : e101 .
Fan H. , Atiya H.I. , Wang Y. , Pisanic T.R. , Wang T.-H. , Shih I.-M. , Foy K.K. , Frisbie L. , Buckanovich R.J. , Chomiak A.A. et al. . Epigenomic reprogramming toward mesenchymal-epithelial transition in ovarian-cancer-associated mesenchymal stem cells drives metastasis . Cell Rep. 2020 ; 33 : 108473 .
Smith J.P. , Corces M.R. , Xu J. , Reuter V.P. , Chang H.Y. , Sheffield N.C. PEPATAC: an optimized pipeline for ATAC-seq data analysis with serial alignments . NAR Genom. Bioinform. 2021 ; 3 : lqab101 .
Miller H.E. , Montemayor D. , Abdul J. , Vines A. , Levy S.A. , Hartono S.R. , Sharma K. , Frost B. , Chédin F. , Bishop A.J.R. Quality-controlled r-loop meta-analysis reveals the characteristics of r-loop consensus regions . Nucleic Acids Res. 2022 ; 50 : 7260 – 7286 .
Yan F. , Powell D.R. , Curtis D.J. , Wong N.C. From reads to insight: a hitchhiker’s guide to ATAC-seq data analysis . Genome Biol. 2020 ; 21 : 22 .
Chen H. , Lareau C. , Andreani T. , Vinyard M.E. , Garcia S.P. , Clement K. , Andrade-Navarro M.A. , Buenrostro J.D. , Pinello L. Assessment of computational methods for the analysis of single-cell ATAC-seq data . Genome Biol. 2019 ; 20 : 241 .
Simovski B. , Kanduri C. , Gundersen S. , Titov D. , Domanska D. , Bock C. , Bossini-Castillo L. , Chikina M. , Favorov A. , Layer R.M. et al. . Coloc-stats: a unified web interface to perform colocalization analysis of genomic features . Nucleic Acids Res. 2018 ; 46 : W186 – W193 .
Kanduri C. , Bock C. , Gundersen S. , Hovig E. , Sandve G.K. Colocalization analyses of genomic elements: approaches, recommendations and challenges . Bioinformatics . 2019 ; 35 : 1615 – 1624 .
Sheffield N.C. , Bock C. LOLA: Enrichment analysis for genomic region sets and regulatory elements in R and Bioconductor . Bioinformatics . 2016 ; 32 : 587 – 589 .
Nagraj V.P. , Magee N.E. , Sheffield N.C. LOLAweb: a containerized web server for interactive genomic locus overlap enrichment analysis . Nucleic Acids Res . 2018 ; 46 : W194 – W199 .
Gharavi E. , Gu A. , Zheng G. , Smith J.P. , Zhang A. , Brown D.E. , Sheffield N.C. Embeddings of genomic region sets capture rich biological associations in low dimensions . Bioinformatics . 2021 ; 37 : 4299 – 4306 .
Gharavi E. , LeRoy N.J. , Zheng G. , Zhang A. , Brown D.E. , Sheffield N.C. Joint representation learning for retrieval and annotation of genomic interval sets . Bioengineering . 2024 ; 11 : 263 .
Zheng G. , Rymuza J. , Gharavi E. , LeRoy N.J. , Zhang A. , Brown D.E. , Sheffield N.C. Methods for evaluating unsupervised vector representations of genomic regions . NAR Genom. Bioinform. 2024 ; 6 : lqae086 .
LeRoy N.J. , Smith J.P. , Zheng G. , Rymuza J. , Gharavi E. , Zhang A. , Brown D.E. , Sheffield N.C. Fast clustering and cell-type annotation of scATACdata with pre-trained embeddings . NAR Genom. Bioinform. 2024 ; 6 : lqae073 .
Samb R. , Khadraoui K. , Belleau P. , Deschênes A. , Lakhal-Chaieb L. , Droit A. Using informative multinomial-dirichlet prior in a t-mixture with reversible jump estimation of nucleosome positions for genome-wide profiling . Stat.Appl. Genet. Mol. Biol. 2015 ; 14 : 517 – 532 .
Granja J.M. , Corces M.R. , Pierce S.E. , Bagdatli S.T. , Choudhry H. , Chang H.Y. , Greenleaf W.J. ArchR is a scalable software package for integrative single-cell chromatin accessibility analysis . Nat. Genet. 2021 ; 53 : 403 – 411 .
Zerbino D.R. , Wilder S.P. , Johnson N. , Juettemann T. , Flicek P.R. The Ensembl Regulatory Build . Genome Biol. 2015 ; 16 : 56 .
Chikina M.D. , Troyanskaya O.G. An effective statistical evaluation of ChIPseq dataset similarity . Bioinformatics . 2012 ; 28 : 607 – 613 .
Zou Z. , Ohta T. , Miura F. , Oki S. ChIP-atlas 2021 update: a data-mining suite for exploring epigenomic landscapes by fully integrating ChIP-seq, ATAC-seq and bisulfite-seq data . Nucleic Acids Res. 2022 ; 50 : W175 – W182 .
Citing articles via.
Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide
Sign In or Create an Account
This PDF is available to Subscribers Only
For full access to this pdf, sign in to an existing account, or purchase an annual subscription.
COMMENTS
The 2024 Nucleic Acids Research Web Server issue is the 22nd in a series of annual issues dedicated to web-based software resources for analysis and visualization of molecular biology data. This issue includes 74 articles covering web servers that support research activities in a wide range of areas, ranging from software aimed at the wet lab ...
Nucleic Acids Research (NAR) is a peer-reviewed journal that publishes papers on various aspects of nucleic acids and proteins. It has a high impact factor and is covered by many indexing services and databases in biochemistry, molecular biology, genetics and related fields.
Browse the latest research articles published in Nucleic Acids Research, a leading journal in the field of molecular biology and biotechnology. Find out the latest developments in chromatin, epigenetics, genome editing, RNA, DNA and more.
Other research demonstrated that viruses could specifically infect and thus regress tumours 135,136 and that injecting certain tissues with nucleic acids could produce genes of interest 137 ...
Nucleic Acids Research is an open-access journal on nucleic acids and related topics, published by Oxford University Press since 1974. It has a high impact factor and publishes special issues on biological databases and web servers.
Nucleic acids function in encoding, transmitting and expressing genetic information in either the double-stranded form (mostly for DNA) or in single-stranded form (mostly for RNA). Latest Research ...
Here, Bou-Nader et al., define the nucleic acid-binding specificity of S9.6 and report its crystal structures free and bound to a hybrid, which reveal the asymmetric recognition of the RNA and DNA ...
In conclusion, this review provided insights into future trends and challenges in nucleic acid research, highlighting the enormous potential and impact of this field. Developments in these areas will not only deepen our understanding of fundamental biological processes but also pave the way for new diagnostic tools, personalized medicine, and ...
Abstract. Nucleic acids, deoxyribonucleic acid (DNA) and ribonucleic acid (RNA), carry genetic information which is read in cells to make the RNA and proteins by which living things function. The well-known structure of the DNA double helix allows this information to be copied and passed on to the next generation.
Nucleic Acid Research is a peer-reviewed journal that publishes papers on various aspects of nucleic acids and proteins. It covers topics such as chemistry, computational biology, gene regulation, genomics, molecular biology, RNA, structural biology and more.
The 2022 Nucleic Acids Research Database Issue contains 185 papers, including 87 papers reporting on new databases and 85 updates from resources previously published in the Issue. Thirteen additional manuscripts provide updates on databases most recently published elsewhere. Seven new databases focu …
NEW AND UPDATED DATABASES. In its 30th incarnation, the Nucleic Acids Research Database Issue once again ranges across biology with a total of 178 papers. Table Table1 1 lists the 90 new databases included, a recent record number, and there are 82 update papers from resources previously covered by NAR. Finally, six databases most recently published elsewhere contribute updates (Table (Table2). 2).
Nucleic Acids Research is a peer reviewed fully open access journal publishing 24 issues per year online and in print. All papers published in the Journal are made freely available online under open access publishing agreements, with applicable charges.
Read the latest Research articles in Nucleic acids from Nature. ... Four different XNAs — polymers with backbone chemistries not found in nature, namely, arabino nucleic acids, 2 ...
Nucleic Acids Research. Nucleic Acids Research is a peer-reviewed scientific journal published by Oxford University Press. It covers research on nucleic acids, such as DNA and RNA, and related ...
Learn about the journal profile, impact factor, ranking, and citation analysis of Nucleic Acids Research, a leading journal in the Web of Science Core Collection.
Molecular diagnostic technologies empower new clinical opportunities in precision medicine. However, existing approaches face limitations with respect to performance, operation and cost. Biological molecules including proteins and nucleic acids are being increasingly adopted as tools in the development of new molecular diagnostic technologies. In particular, leveraging their complementary ...
NEW AND UPDATED DATABASES. The 28th annual Nucleic Acids Research Database Issue contains 189 papers spanning, as usual, a wide range of biology. Unsurprisingly, COVID-19 casts a long shadow over the Issue. Seven new databases specifically address the pandemic and the SARS-CoV-2 virus responsible (Table (Table1) 1) but new and returning databases in all areas have rushed to support research ...
Browse the latest articles on nucleic acid research, including breakthrough, critical, and review articles, as well as data resources and analyses. Topics include gene regulation, chromatin, epigenetics, RNA, DNA, and more.
Nucleic acids have become important diagnostic markers for many diseases, enabled by breakthroughs in synthesis and sequencing technologies. ... The Collection primarily welcomes original research ...
The rapid transmission and high pathogenicity of respiratory viruses significantly impact the health of both children and adults. Extracting and detecting their nucleic acid is crucial for disease prevention and treatment strategies. However, current extraction methods are laborious and time-consuming and show significant variations in nucleic acid content and purity among different kits ...
Angewandte Chemie International Edition is one of the prime chemistry journals in the world, publishing research articles, highlights, communications and reviews across all areas of chemistry. Molecular spherical nucleic acids (m-SNAs) are a second generation of spherical nucleic acids (SNAs), which are of significance in potential application ...
The 29th annual Nucleic Acids Research Database Issue contains 185 papers covering topics from across biology and beyond. The ongoing COVID-19 pandemic continues to play a major role, inspiring the construction of seven new databases (Table (Table1). 1). The reader will also find its impact obvious in papers describing other new and returning ...
Publishes the results of leading edge research into physical, chemical, biochemical and biological aspects of nucleic acids and proteins involved in nucleic acid metabolism and/or interactions. Fully open access.
This innovative strategy has led researchers from Shinshu University, Japan, to develop a novel nucleic acid drug called iSN04. Associate Professor Tomohide Takaya, from the Faculty of Agriculture ...
Nucleic Acids Res; Nucleic Acids Research Vols. 1 to 52; 1974 to 2024; Vol. 52 2024: v.52(D1): D1-D1702 2024 Jan 5: v.52(1): 1-509 2024 Jan 11: v.52(2): 511-1003 ... Articles from Nucleic Acids Research are provided here courtesy of Oxford University Press. Follow NCBI. Connect with NLM National Library of Medicine 8600 Rockville Pike ...
Published online on Aug. 10 in Nucleic Acids Research, ... "This research is significant both for understanding how cells respond to DNA damage, to prevent them from becoming cancerous in the first place, how targeting this pathway can be used in new cancer treatments, and because it now opens up possibilities for new approaches to treating ...
Nucleic Acids Research | 51 | 2 | January 2023. Cover: Zalpha (Zα) domains bind to left-handed Z-DNA and Z-RNA. The Zα domain protein family includes cellular (ADAR1, ZBP1 and PKZ) and viral (vaccinia virus E3 and cyprinid herpesvirus 3 (CyHV-3) ORF112) proteins.
Here, S c is a sum of genome coverage by collection and g is the size of the genome. We realized that, in a sense, the CC universe is a point estimate of a more complex distribution of possible universes. We reasoned that we may gain some insight by modeling the boundaries of the consensus regions as intervals, rather than points.