JHU Deep Learning in Genomics Journal Club
When Genomics Sequences Meet Deep Learning

Kuan-Hao Chao - "Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation" (Linder et al.)

Overview

In this talk, we will explore Borzoi, a novel machine-learning model designed to predict RNA-seq coverage profiles directly from genomic DNA sequence. Borzoi captures cell-type-specific and tissue-specific expression patterns by learning complex cis-regulatory relationships, providing a unified framework that encompasses transcriptional, splicing, and polyadenylation signals. We will review how Borzoi leverages sequence-based deep learning to score variant effects, outperforming traditional models on quantitative trait loci prediction, and demonstrate how interpretability methods extract biologically relevant motifs driving RNA regulation.

Paper Information

  • Title: Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation
  • Authors: Johannes Linder, Divyanshi Srivastava, Han Yuan, Vikram Agarwal, David R. Kelley, et al.
  • Journal: Nature Genetics, Volume 57, pages 949–961 (2025)
  • DOI & Link: https://www.nature.com/articles/s41588-024-02053-6

Slides

Below are the slides for this talk, embedded directly from Google Drive:

Contact

For questions or further discussion, please reach out to:
Kuan-Hao Chao
Computational Genomics PhD Candidate
Johns Hopkins University, Center for Computational Biology
Email: kchao10@jhu.edu