Prediction of age using shotgun metagenomic sequencing and random forest algorithm based on cadaveric colon

Age estimation is important for the identification of unknown cadavers in forensic practice. Previous studies have shown that gut microbiota is associated with host age, but most evidence has been derived from fecal samples of living individuals. In this study, shotgun metagenomic sequencing was performed on mid-colon tissue samples from 76 cadavers to explore age-associated taxonomic and predicted functional patterns in cadaveric colon microbiota and to evaluate their potential value for forensic age estimation. After quality control and taxonomic annotation, 3980 microbial species were identified. Descriptive differences among age groups were observed in microbial composition, alpha diversity, species-enrichment patterns, co-occurrence network structure, and KEGG functional profiles; however, a multivariable PERMANOVA did not detect a statistically significant association between age group and overall species-level community composition after accounting for postmortem sampling interval, cause-of-death category, and sex. The complete random forest pipeline was re-evaluated using repeated nested five-fold cross-validation, with all data-dependent filtering, transformation, feature selection, and hyperparameter tuning restricted to the outer training data. The resulting out-of-fold performance was limited (R² = 0.005, MAE = 13.686 years, and RMSE = 17.741 years), and predictions showed regression toward the cohort mean. Overall, this study provides preliminary evidence that cadaveric colon microbiota contains age-associated microbial signals, but these findings should be interpreted as exploratory. Larger cohorts and independent external validation are needed before microbiome-based age prediction can be applied in forensic practice.