Enformer (Human) Model Predictor (Avsec et al. 2021) using the Genomic API for Model Evaluation (GAME) Framework

This record provides a Predictor container for the Enformer model (Avsec et al. 2021, Nature Methods). Enformer is a deep-learning architecure that substantially improves gene expression prediction from DNA sequences by integrating information from long-range interactions up to 100 kb away. Its key innovation is the use of novel transformer layers, which more effectively model the influence of distal regulatory elements like enhancers on gene expression and chromatin states in humans (and mice, but this Predictor is for humans only). The model predicts genomic tracks for the human genome, including CAGE for transcriptional activity, histone modifications, transcription factor binding, and DNA accessibility, all aggregated into 128-bp bins. It was trained in a multitask setting on a vast collection of human and mouse epigenomic datasets to study cis-regulatory evolution.