A Quantitative Model of Transcriptional Regulation Reveals the Influence of Binding Location on Expression
(A) Shown are representative scatter plots of predicted vs. observed expression differences for held out test genes in liver/cerebellum and liver/3T3-L1 cells. Predictions were made using a transcriptional model that takes into account the influence of both the genomic position and the particular proteins bound by a site. The median correlation from 11 separate trials was 0.65 and 0.74 for liver/cerebellum and liver/3T3-L1 respectively. (B) The prediction error of the full model that includes individual transcription factor influence weights is compared to a model that uses only position to predict influence. Modeling the influence of bound regulators improves predictive performance. Error bars indicate +/− s.e.m. (C) The expression influence for each protein is learned in our transcriptional model. Sites bound by proteins with known repressive activity (E2F4 and Sirt1) are predicted to have the smallest influence.