Kareem, That is a very nice analogy you made with marbles in a jar! The white marbles represent peaks in your TF1 list and the black ones are non TF1 peaks. TF2 peaks would be the results sampled from the jar including both white and black marbles. The overlap would be white marbles with TF2 peaks. Is this what you are referring to too? Thanks! Best regards, Julie On 11/7/11 10:22 AM, "Carr, Kareem" <kareemcarr@fas.harvard.edu> wrote: Dear Dr. Zhu, I have been working with your ChIPpeakAnno package and I had a question about the p-value for makeVennDiagram. I have read the posts on stat.ethz.ch and gmane.org including the comments by Noah Dowell where he suggests picking one of the transcription factors and estimating it’s number of possible binding sites. When I relate the p-value computed by the hypergeometric distribution to the idea of having a jar of marbles which are both white and black and taking a sample. It seems to me that marbles are all the possible binding sites of my first transcription factor TF1. The black ones represent sites with no peak and the white ones represent sites with peaks. My second TF2 should represent taking a sample of marbles where some will be white and some will be black. My problem is the random variable represented by TF2 doesn’t only sample sites from all the binding sites of TF1. It can also be said to be sampling sites where TF2 could bind and TF1 could not. Therefore, in order to make this model of overlapping peaks work, we actually want the random variable represented by all binding sites of TF2 given that they are also binding sites of TF1. Do you agree with this analysis? I would appreciate any insight that you can give. Thanks. Kareem -------------------------------------------------------- Kareem Carr Research Fellow Department of Molecular and Cellular Biology Harvard University Website: http://www.people.fas.harvard.edu/~kareemcarr/ [[alternative HTML version deleted]]

Kareem, Many thanks for your positive feedback! Yes, TF2 could also have binding sites which are not possible binding sites for TF1, although the jar includes both binding sites of TF1 and non-binding sites of TF1. Best regards, Julie On 11/7/11 10:55 AM, "Carr, Kareem" <kareemcarr@fas.harvard.edu> wrote: Hi Julie, Thanks for the quick response. I would like to take this opportunity to say what great work ChIPpeakAnno is and how useful it has been in my work so far. Yes, your answer is what I was thinking, based on looking at your code. My confusion with the analogy is thinking of TF2 as only taking marbles from the jar (which represents binding sites of TF1). Couldn’t TF2 also have binding sites which are not possible binding sites for TF1? Thanks. Kareem From: Zhu, Lihua (Julie) [mailto:Julie.Zhu@umassmed.edu] Sent: Monday, November 07, 2011 10:50 AM To: Carr, Kareem Cc: bioconductor Subject: Re: ChIPpeakAnno, makeVennDiagram Kareem, That is a very nice analogy you made with marbles in a jar! The white marbles represent peaks in your TF1 list and the black ones are non TF1 peaks. TF2 peaks would be the results sampled from the jar including both white and black marbles. The overlap would be white marbles with TF2 peaks. Is this what you are referring to too? Thanks! Best regards, Julie On 11/7/11 10:22 AM, "Carr, Kareem" <kareemcarr@fas.harvard.edu> wrote: Dear Dr. Zhu, I have been working with your ChIPpeakAnno package and I had a question about the p-value for makeVennDiagram. I have read the posts on stat.ethz.ch and gmane.org including the comments by Noah Dowell where he suggests picking one of the transcription factors and estimating it’s number of possible binding sites. When I relate the p-value computed by the hypergeometric distribution to the idea of having a jar of marbles which are both white and black and taking a sample. It seems to me that marbles are all the possible binding sites of my first transcription factor TF1. The black ones represent sites with no peak and the white ones represent sites with peaks. My second TF2 should represent taking a sample of marbles where some will be white and some will be black. My problem is the random variable represented by TF2 doesn’t only sample sites from all the binding sites of TF1. It can also be said to be sampling sites where TF2 could bind and TF1 could not. Therefore, in order to make this model of overlapping peaks work, we actually want the random variable represented by all binding sites of TF2 given that they are also binding sites of TF1. Do you agree with this analysis? I would appreciate any insight that you can give. Thanks. Kareem -------------------------------------------------------- Kareem Carr Research Fellow Department of Molecular and Cellular Biology Harvard University Website: http://www.people.fas.harvard.edu/~kareemcarr/ [[alternative HTML version deleted]]