Authors
Hanshi Xu, Guangquan Zhang, Hua Lin, Mark Grosser, Jie Lu
Published in
Briefings in bioinformatics. Volume 27. Issue 4. Jul 03, 2026.
Abstract
Breast cancer is one of the most distressing cancers affecting women, and early detection is considered the most effective way to reduce breast cancer mortality. However, the benefits of early detection vary among different risk groups. Therefore, using a combination of genetic information, family history, and other factors to stratify populations by risk can help more people benefit from early detection. Traditional polygenic risk score (PRS) is essentially a weighted sum calculation method that has achieved some success, but it neglects the interactions between genes-genes, genes-environment, and their potential impact on breast cancer risk. In this context, we developed a new deep learning-based method called wide, deep, and cross network (WDCN). Experimental results show that our algorithm outperforms PRS and other machine learning baseline methods and achieves an area under the receiver operating characteristic curve (AUROC) of 0.6439 when using 286 single nucleotide polymorphism (SNP) features and 0.8865 when incorporating environmental features with genetic data. Increasing the SNP set to 317 further raised the performance to 0.6464 and 0.8872, both with and without non-genetic factors. Risk stratification shows that individuals in the top 30% have a relative risk of 7.85 (95% CI: 6.98-8.83) compared with those in the bottom 30%. We also identified an interaction between rs2588809 and age. This novel approach has shown promise for initial risk stratification of populations, potentially providing better decision-making support for individuals and clinicians.
PMID:
42561150
Bibliographic data and abstract were imported from PubMed on 07 Aug 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 8
- Comments 0