Deep learning sequence-based ab initio prediction of variant effects on expression and disease risk

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Deciphering gene expression regulation and the impact of genomic variations is crucial for human genetics, precision medicine, and evolutionary biology.
• The vast scale of the noncoding genome presents a significant challenge in understanding mutation effects.

Purpose of the Study:

• To develop a deep learning framework, ExPecto, for accurate *ab initio* prediction of tissue-specific transcriptional effects of DNA mutations.
• To analyze the regulatory mutation space and its implications for gene expression and disease risk.

Main Methods:

• Developed ExPecto, a deep learning framework predicting mutation effects on gene expression from DNA sequence.
• Applied ExPecto to prioritize causal variants in genome-wide association studies (GWAS) loci.
• Conducted *in silico* saturation mutagenesis to characterize regulatory mutation space for over 140 million promoter-proximal mutations.

Main Results:

• ExPecto accurately predicts *ab initio* tissue-specific transcriptional effects of mutations, including rare and unobserved ones.
• Experimental validation confirmed predictions for four immune-related diseases.
• Characterized the regulatory mutation space for human RNA polymerase II-transcribed genes, enabling analysis of evolutionary constraints.

Conclusions:

• ExPecto provides an end-to-end computational framework for predicting gene expression and disease risk from DNA sequence.
• The framework facilitates the study of evolutionary constraints on gene expression and the prediction of mutation-induced disease effects.