Tissue-specific transfer learning improves functional variant and therapeutic target discoveries in breast and prostate cancer

DNA foundation models trained on large-scale genomic and epigenomic datasets have shown promise for regulatory variant interpretation, yet their application to tissue-specific contexts remain limited. Here, we present a transfer learning (TL) framework to adapt Enformer, a deep neural network trained on 5,313 multi-omics tracks, to breast and prostate cancer using 275 and 357 tissue-specific transcription factor (TF) ChIP–seq tracks, respectively. We computed tissue-specific cis-regulatory activity (tCRA) scores for millions of single-nucleotide variants (SNVs) in genome-wide association study (GWAS) datasets and prioritized high-impact SNV subsets (1M, 1.5M, and 2M). These TL-prioritized variants demonstrated consistently greater enrichment in tissue-specific enhancers, cancer GWAS risk variants, and ClinVar pathogenic variants compared to the original Enformer model. Transcriptome-wide association studies (TWAS) using TL-based SNVs identified more cancer-relevant genes, many of which exhibited functional essentiality (DepMap), therapeutic tractability (drug databases), and disease relevance (DisGeNET). Notably, TL models outperformed the base model in identifying genes enriched for drug targets and clinically relevant disease associations. Our results show that TL-derived tCRA scores enhance regulatory variant prioritization and improve susceptibility gene discovery in a tissue-specific manner. Our study provides a generalizable framework for tailoring foundation models to disease-relevant contexts, with implications for variant interpretation, therapeutic target discovery, and precision medicine.Author summary: Understanding how genetic changes contribute to cancer remains a central challenge in human genetics. While powerful deep learning models like Enformer can predict how DNA variants might affect gene regulation, they are often trained on very broad data and may not capture the tissue-specific mechanisms relevant to specific cancers. In this study, we developed a transfer learning (TL) approach to adapt Enformer for breast and prostate cancer by retraining it on datasets specific to each cancer type. This allowed us to compute regulatory scores for millions of genetic variants and identify those most likely to affect cancer risk. We found that our TL-enhanced models perform better at highlighting genetic variants located in tissue-specific regulatory regions. Using these high-priority variants, we linked genes to cancer risk through transcriptome-wide association studies (TWAS) and showed that many of the identified genes are important for cancer cell growth and are potential drug targets. Our findings demonstrate how adapting existing models to more disease-relevant data can significantly improve our ability to uncover genes and variants involved in cancer. This work provides a new tool for researchers aiming to understand genetic risk and discover future therapies.