AI technique improves cancer gene discovery for breast and prostate cancer
AI technique improves cancer gene discovery for breast and prostate cancer
Publish Date: 2026-05-29 11:41:00
Source Domain: news.vumc.org
-
Genome-wide association studies (GWAS) have identified numerous genetic “spots” linked to diseases like cancer, but connecting these genetic changes with their specific roles in disease is still complex.
-
Artificial intelligence deep-learning models, like Enformer, can predict the impact of DNA changes on gene regulation but lack specificity to certain tissues.
-
A research team led by Qing Li, Xingyi Guo, and Quan Long developed an AI transfer learning approach to improve Enformer’s predictions for breast and prostate cancer.
-
The team retrained the Enformer model using tissue-specific datasets consisting of transcription factor chromatin immunoprecipitation sequencing for breast (275) and prostate (357) cancers.
-
The new transfer learning models computed regulatory scores for millions of GWAS variants, identified critical genes likely to affect cancer risk, and pointed to them as potential drug targets important for cancer cell growth.
-
The study in PLOS Genetics demonstrated that the transfer learning models were superior to the base model in pinpointing clinically relevant genes associated with cancer, offering a generalizable framework for adapting AI models to disease contexts.
-
The research results highlight the potential of adapting existing AI models with more disease-specific data to enhance cancer gene and variant discovery.
