Using Deep Learning Models of Gene Regulation to Guide Drug Prioritization

Area of Science:

• Genomics
• Computational Biology
• Pharmacology

Background:

• Most drug discovery computational methods overlook noncoding genetic variation.
• Over 90% of genome-wide association study (GWAS) risk variants are in noncoding regions, posing a challenge for therapeutic hypothesis generation.
• Linking regulatory variations to potential therapies is crucial for advancing drug repurposing.

Purpose of the Study:

• To develop an integrative deep learning framework for drug repurposing guided by noncoding genetic variation.
• To connect allele-specific enhancer predictions with candidate therapeutics.
• To establish a scalable strategy for generating pharmacologically relevant hypotheses from the noncoding genome.

Main Methods:

• Developed a deep learning framework integrating allele-specific enhancer prediction and drug prioritization.
• Employed transcription factor (TF)-based and gene-based prioritization strategies.
• Utilized MCF7-breast cancer cell line data as a proof-of-concept.

Main Results:

• GWAS heritability was significantly enriched in MCF7 enhancers.
• Identified 1537 breast cancer risk variants with predicted regulatory effects, enriched for FOXA1 motifs.
• Prioritization strategies identified 63 (TF-based) and 140 (gene-based) candidate compounds, including approved drugs like fulvestrant, toremifene, and raloxifene.
• 15 high-confidence repurposing candidates were identified through integrated evidence.

Conclusions:

• The framework successfully recovered approved breast cancer therapeutics, validating the biological relevance of deep learning-predicted regulatory variants.
• This regulatory variant-guided approach connects noncoding genetic variation to candidate therapeutics.
• The study provides a scalable method for drug repurposing by leveraging noncoding genomic data.