Identifying a Minimal Multi-Kingdom Gut Microbiome Panel for Colorectal Cancer Detection Using LASSO Logistic Regression

Abstract

Colorectal cancer (CRC) is one of the most prevalent and lethal cancers worldwide. Incidence is rising in low- and middle-income countries and among adults under 50, compounding an already substantial global burden. Current screening tools, including colonoscopy, fecal immunochemical testing, and multi-target stool DNA testing, are limited by invasiveness, cost, and suboptimal sensitivity and specificity. The gut microbiome has emerged as a promising source of CRC biomarkers, with specific microbial signatures consistently associated with CRC across independent cohorts. However, existing microbiome-based prediction panels are large, bacteria-only, and have not undergone formal parsimony evaluation. To characterize bacterial and fungal community alterations in CRC, I analyzed the Lin et al. (2022) multi-cohort fecal metagenomic dataset spanning 979 samples across 8 geographically diverse cohorts. Differential abundance analysis identified 123 bacterial and 46 fungal taxa significantly altered in CRC, with established signatures including Fusobacterium nucleatum and Aspergillus rambellii replicated across cohorts. Building on these findings, I applied a two-pass LOCO framework with LASSO logistic regression to identify and validate a minimal multi- kingdom CRC prediction panel. A panel of just 6 taxa, comprising four bacterial and two fungal taxa, achieved a pooled out-of-sample AUROC of 0.776 and was statistically equivalent to a model comprising 76 taxa. These findings demonstrate that a parsimonious, multi-kingdom microbiome panel can achieve cross-cohort consistent CRC prediction accuracy and establish a principled framework for minimal diagnostic panel development that prioritizes clinical translation.