HL38: Kyoungjae Won - Applying histone modification information to genome-wide prediction of transcription factor binding sites
Started with 8 histone modifications as training data, and 13 transcription factors (ChIP-seq binding data; evaluation set) and another evaluation set. All in mouse stem cells. - Barb Bryant
Shows example Oct4, with enhancer specific binding properties - low H3K4me3 and high H3K4me1. - Barb Bryant
To train a model for Klf4: looked for high H3K4me3 with the Klf4 PSSM; at teh enhancer looked for high H3K4me1/2 and Klf4 PSSM. - Barb Bryant
His software/method is called Chromia. - Barb Bryant
Train with HMM; 3 HMMs, one for promoter binding Klf4, one for enhancer binding Klf4, and one for background. - Barb Bryant
Conservation information seems to worsen the predictions! - Barb Bryant
Used RNAi against the TF - Barb Bryant
Used a 3-state HMM, which captures the spatial pattern of histone marks - Barb Bryant
Chromia predicts locations of enhancers and promoters. Chromia predicts nucleosome free regions. - Barb Bryant
Publication: BMC Bioinformatics - Barb Bryant
Colleagues Wei Wang, Bing Ren, Robert Shoemaker, Iouri Chevelev - Barb Bryant
Reference for this work: - Barb Bryant
Question about relative weight of PSSMs and histone marks in the models. - Barb Bryant