Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation

Summary

This paper presents Borzoi, a sequence-based machine-learning model capable of predicting cell-type and tissue-specific RNA-seq coverage from DNA sequence alone. The model integrates multiple layers of gene regulation—transcription, splicing and polyadenylation—into a unified framework, enabling more comprehensive prediction of variant effects than existing tools that target individual regulatory functions. The authors demonstrate the model's potential to decipher the relationship between DNA sequence and regulatory function across diverse biological contexts.

Regional applicability

This is fundamental genomics research with no direct application to United Kingdom farming systems, soil health or agricultural practice. The methodology and tools may have indirect relevance for crop and livestock genomics research conducted in the UK, but the study itself is not agriculture or nutrition-focused.

Key measures

Predicted RNA-seq coverage; DNA variant effect scores; regulatory motif identification; performance benchmarking against quantitative trait loci

Outcomes reported

The study developed and validated Borzoi, a machine-learning model that predicts cell-type-specific and tissue-specific RNA-seq coverage directly from DNA sequence, and demonstrated its ability to score DNA variant effects across transcription, splicing and polyadenylation. The model's predictions were evaluated against quantitative trait loci data and compared to state-of-the-art models for individual regulatory functions.

Topic tags