# NIH grant: Novel Methods for Identifying Genetic Interactions for Cancer Prognosis (2R01CA204120-09A1) — retained record

> Retained us-gov-api record from us-gov-api:nih-reporter.

Canonical: https://highsignal.app/data/records/fae332a96fb62b6c

[← sources](https://highsignal.app/data)


source: [us-gov-api](https://highsignal.app/data/us-gov-api) (us-gov-api:nih-reporter)  published 2026-08-21 00:00 ingested 2026-08-24 06:48

[original source ↗](https://reporter.nih.gov/project-details/2R01CA204120-09A1)

## retained text



```
NIH grant 2R01CA204120-09A1: Novel Methods for Identifying Genetic Interactions for Cancer Prognosis
Institute: {'code': 'CA', 'abbreviation': 'NCI', 'name': 'National Cancer Institute', 'admin_org_id': '12500008', 'admin_funding_url': ''}. PI: Shuangge  Ma. Award: $424882.
Start: 2026-08-21T00:00:00.
Project Summary
Gene-environment (G-E) interactions have critical implications for better understanding and modeling cancer
prognosis. For over a decade, our group has been continuously supported by NIH and other funding agencies
and has taken a leading role in developing cutting-edge methods for G-E interaction analysis. In the first funding
period, we developed a series of regularized regression methods. In the current funding period, we have been
the first to develop statistically principled deep learning (SDL) techniques tailored to G-E interaction analysis.
We have extensively analyzed groundbreaking prognosis studies on lung cancer, skin cancer, lymphoma, thyroid
cancer, and others. Our research has led to the identification of important G-E interactions otherwise missed,
advanced understanding of disease etiology and prognosis, suggested new targets for therapeutic development,
and assisted in constructing more accurate disease models.
In a series of recent studies, we have shown that deep neural network (DNN)-based G-E interaction analysis
significantly outperforms regression-based, more flexibly accommodating unspecified nonlinear relationships
and more accurately modeling/predicting prognosis. However, such advantages are often accompanied by
inferior stability and interpretability, which critically hinders practical applications.
To directly tackle such limitations and facilitate broader utilization, in this renewal, we will be the first to develop
a comprehensive set of novel G-E interaction analysis strategies and methods, integrate outside information to
assist local data analysis, and fundamentally improve stability and interpretability. Specifically, we will: (Aim 1)
Develop SDL techniques to integrate information in pathological imaging data; (Aim 2) Develop SDL techniques
to integrate information in published literature; and (Aim 3) Develop SDL techniques to integrate information in
raw data of independent studies. These three sources of information are likely the most effective and cost-
effective and complement each other. The proposed DNN techniques will be tailored to the unique characteristics
of each type of information, while enjoying methodological coherence. We will develop effective computational
algorithms and portable software, conduct extensive simulations and benchmarking, and establish statistical
properties, which will lead to foundational contributions to cancer analytics. Equally importantly, we will (Aim 4)
Analyze the Yale SPORE, Veteran Affairs (VA) Million Veteran Program (MVP), and TCGA data on lung and
skin cancers. Such data are among the most powerful. Analysis will be comprehensively conducted on multiple
prognosis outcomes and molecular measurements. Results will be thoroughly and rigorously evaluated,
benchmarked, and validated. This effort will lead to a high and direct translational and clinical impact.
```



document fetched 2026-08-30 02:46