Fig. 1

Explainable AI and machine learning models for enhancer identification. A A graphical representation of our rule based and machine learning analysis. ChIP-seq data for histone modifications and STARR-seq enhancer annotations are combined and tiled into bins covering the Drosophila genome. Using these bins, traditional machine learning models (ML) and explainable AI models (XAI) can be trained to predict enhancer locations. B Confusion matrix statistics from individual bin predictions. TP— true positives (detected by XAI/ML and STARR-seq), TN—true negatives (not detected by XAI/ML or STARR-seq), FP—false positives (detected only by XAI/ML), and FN—false negatives (detected only by STARR-seq). Accuracy ((TP + TN)/(TP + TN + FP + FN)), precision (TP/(TP + FP)), and recall (TP/(TP + FN)) were computed and plotted for the best performing explainable AI and neural network (NN) models. All models were trained in BG3 cells and then applied to S2 data. C Overlap between XAI and STARR-seq predicted enhancers in BG3 and S2 cells respectively