Discovery of genes positively modulating treatment effect using potential outcome framework and Bayesian update

Area of Science:

• Genomics
• Cancer Research
• Biostatistics

Background:

• Cancer genomics generates vast mutation and gene expression data, but clinical utility remains limited.
• Identifying genetic alterations that predict treatment response is a significant challenge.
• Conventional statistical methods lack power and causal inference capabilities in genomic datasets.

Purpose of the Study:

• To develop and evaluate a C-search algorithm for identifying causal genes that optimize cancer treatment efficacy.
• To address the limitations of conventional statistics in genomic data analysis for causal inference.

Main Methods:

• The C-search algorithm was developed using the potential outcome framework and Bayesian posterior updates.
• Algorithm precision was validated via simulation datasets.
• Implementation and external validation were performed using cBioPortal and CancerSCAN datasets.

Main Results:

• The C-search algorithm successfully identified 9 out of 10 causal genes in simulation data.
• Discovery rates increased significantly with data instances, outperforming the log-rank test.
• Identified causal genes were associated with improved patient survival in both cBioPortal and CancerSCAN datasets.

Conclusions:

• The C-search algorithm outperforms traditional methods like the log-rank test for identifying causal gene effects, particularly with limited sample sizes.
• This algorithm can effectively discover causal genes from diverse genetic datasets with numerous variables and limited samples.