Context-aware sequence-to-function model of human gene regulation

Sequence-to-function models have been very successful in predicting gene expression, chromatin accessibility, and epigenetic marks from DNA sequences alone. However, current state-of-the-art models have a fundamental limitation: they cannot extrapolate beyond the cell types and conditions included in their training dataset. Here, we introduce Corgi, a context-aware sequence-to-function model that overcomes this limitation by integrating DNA sequence and trans-regulator expression to predict chromatin accessibility, histone modifications, and gene expression coverage, even in held-out cell types. Trained on a diverse set of bulk and single-cell sequencing datasets, Corgi achieves top performance in joint cross-sequence and cross-cell-type epigenetic track prediction. Additionally, we present an advanced model version, Corgi+, which is state-of-the-art in imputation of epigenetic tracks using only RNA-seq data. We further show that Corgi learns key cell type-specific trans-regulators in a zero-shot manner, and it can predict genomic variant effects in held-out cell types.