Integrating proteogenomics to predict key targets for recurrence and metastasis in locally advanced colorectal cancer and clinical validation

Patients with locally advanced colorectal cancer (LACRC) often experience recurrence within five years post-surgery, highlighting the need to identify novel targets that elucidate the underlying regulatory mechanisms. This study investigated recurrence in LACRC through integrated proteogenomics analysis, identifying COL10A1 and THBS2 as key targets. This single-center retrospective study included 78 patients with LACRC who underwent radical surgery and standardized chemotherapy from March 2022 to September 2025. We analyzed CPTAC data alongside clinical whole-exome sequencing (WES) data from this cohort using Fisher’s exact test, the log-rank test, and Cox regression. Logistic regression and random forest models were developed using gene mutation and clinicopathological variables and evaluated by repeated stratified 10-fold cross-validation. Proteogenomic analysis identified COL10A1 and THBS2 as overexpressed in LACRC. COL10A1 expression correlated with DFS, whereas THBS2 was associated with both OS and DFS ( P < 0.05). In the recurrence group, the COL10A1 and THBS2 mutation frequencies were 63.16% and 71.05%, respectively (both P < 0.05). COL10A1 mutation was independently associated with poorer DFS (HR = 2.350, 95% CI = 1.038–5.321, P = 0.040), whereas THBS2 mutation was associated with improved OS (HR = 0.607, 95% CI = 0.369–0.997, P = 0.049). The logistic regression model achieved a mean internally cross-validated AUC of 0.843, whereas the random forest model achieved an AUC of 0.724. Mutations in COL10A1 and THBS2 were associated with recurrence-related outcomes in this cohort. Bioinformatic analyses suggest potential involvement in ECM-related pathways. A multivariable model incorporating mutation and clinicopathological features showed potential for recurrence-risk stratification, pending external validation.