Chromatin occupancy dynamics are predictive of gene expression. (A) Comparison of the performance of each GP model using the coefficient of determination, R2. The full model incorporating all chromatin features and initial transcript level outperforms all other models after 15 min. Later time points rely less on initial transcript level for prediction, so the statistical benefits of chromatin features become more evident over time (compared to RNA-only baseline). (B) Comparison between true and predicted log2 transcript level for the full model after 7.5 min. Most genes are well predicted using 0 min transcript level. (C) Full model predictions at 30 min. Predictions remain well correlated, but less than at 7.5 min. (D) Full model predictions at 120 min. After two full hours have elapsed, transcript level predictions have become a bit less correlated, but still, R2 remains 0.44.
