TIGP (BIO)—Statistical applications for human genetic data: from GWAS to integrative analysis

Abstract

The expansion of human genomic datasets, particularly through large-scale biobanks, has created unprecedented opportunities for advancing precision medicine while presenting significant analytical challenges. This presentation provides an overview of computational methods for genetic data analysis, beginning with foundational concepts in genome-wide association studies (GWAS)—a cornerstone approach in biobank research for identifying disease-associated variants and enabling polygenic risk score development. In addition, we will examine current opportunities and challenges in biobank analysis. The second section presents recent research progress on a multimodal fusion machine learning framework that systematically integrates histopathological images and gene expression data for enhanced cancer prognosis prediction. This approach employs biologically-informed dimensionality reduction through pathway-based feature extraction, attention-based tensor product fusion to capture cross-modal interactions, and Cox proportional hazards modeling, achieving a concordance index of 0.88 in brain tumor survival prediction. The integrative analysis presented in the second section is joint research with Ms. Yun-Ya Liu.