Recapitulation of patient-specific 3D chromatin conformation using machine learning.

Area of Science:

• Genomics
• Cancer Biology
• Computational Biology

Background:

• Cellular states are defined by regulatory networks linking enhancers to genes.
• Existing methods for mapping these networks require extensive data, limiting application to large patient cohorts.
• Chromatin immunoprecipitation sequencing (ChIP-seq) is infeasible for limited biopsy material.

Purpose of the Study:

• To develop a machine-learning approach for predicting enhancer-gene regulatory networks from limited biopsy samples.
• To overcome limitations of correlation-based methods in distinguishing target genes and regulatory states.
• To enable large-scale analysis of network rewiring in cancer.

Main Methods:

• Trained machine-learning models using chromatin interaction analysis with paired-end tag sequencing (ChIA-PET) and high-throughput chromosome conformation capture combined with chromatin immunoprecipitation (HiChIP) data.
• Utilized assay for transposase-accessible chromatin using sequencing (ATAC-seq) and RNA-seq data as input, suitable for limited biopsy material.
• Applied the model to 371 samples across 22 cancer types.

Main Results:

• Identified 1,780 enhancer-gene connections for 602 cancer genes.
• The model accurately predicts regulatory connections using only ATAC-seq and RNA-seq data.
• Validated predicted enhancers regulating ESR1 in breast cancer and A1CF in liver cancer using CRISPR interference (CRISPRi).

Conclusions:

• The developed method enables scalable prediction of enhancer-gene networks from limited patient biopsies.
• This approach facilitates the study of network rewiring in cancer and identification of novel cancer gene regulators.
• The findings have implications for understanding cancer biology and developing targeted therapies.