This repository contains the analysis code, pipelines, and plotting scripts for the manuscript: "Sequence-to-function deep learning decodes human cis-regulatory evolution" Mangan et al., 2026
This repository provides the computational framework used to predict lineage-specific cis-regulatory elemenets (linCREs) across modern humans, archaic hominin, and great ape personalized genomes. It includes the pipelines for genome-wide chromatin accessibility prediction using Enformer, variant effect prediction using AlphaGenome, identifying differential accessibility, downstream functional and eovlutionary enrichments, and plotting and visualization of experimental validation.
This repository is organized by computational environment and analysis stage:
clusterAnalysisScripts/- Scripts designed to run on the MIT Luria high-performance compute cluster.
- Contains the heavy-compute pipelines (e.g., genome assembly, multiple alignment, genome-wide Enformer scanning, peak calling)
localAnalysisScripts/- Scripts designed for local machine execution.
- Contains R scripts for downstream statistical analyses, differential accessibility quantification, GREAT enrichments, and manuscript figure generation.
SupplementaryTables/- Contains the final formatted supplementary tables referenced in the manuscript
To facilitate reproducibility, we have provided the raw data required for figure generation wherever possible. We have generally not included intermediate files when these files can be regenerated from provided data and scripts after short computes.
Additional large data files, including the 35-way genome alignment, are indexed on Zenodo: Zenodo archive: https://doi.org/10.5281/zenodo.22047727
For questions regarding the code, data, or analyses, please open an issue in this repository or contact:
Riley J. Mangan
Massachusetts Institute of Technology / Broad Institute
Email: rimangan@mit.edu