A Computational Pipeline for Patient-Level Gene-Expression Analysis in Colorectal Cancer

Group-level differential expression does not show whether gene-expression changes are consistent across individual patients. I developed a Python-based computational pipeline to evaluate patient-level consistency using a publicly available colorectal cancer RNA-sequencing dataset containing 20 matched tumor-normal pairs. Paired differential expression analysis identified 8,300 of 14,911 genes as significant at FDR < 0.05. For each gene, I calculated a direction-agnostic patient-consistency score based on the proportion of non-tied patients showing the same direction of change and tested this against chance using an exact binomial test. Consistency varied independently of fold-change magnitude: 1,927 genes showed modest average changes but high patient-level consistency, whereas 567 showed large average changes with lower consistency. For 1,261 genes (8.5%), the majority patient direction differed from the group-level average. Results were reproduced in a held-out 14-patient cohort, with strong fold-change correlation (r = 0.858) and 95.7% directional replication among high-consistency genes. Hierarchical clustering and pathway-enrichment analyses further supported the distinction between consistent and heterogeneous expression patterns. The pipeline demonstrates a reproducible computational approach for analyzing patient-level variability in matched gene-expression data. Keywords computational pipeline; colorectal cancer; RNA sequencing; differential gene expression; patient-level consistency; tumor-normal pairs; cross-cohort validation; gene set enrichment; Python; bioinformatics