A 13-gene prognostic model developed using machine learning to predict the response to neoadjuvant chemoradiotherapy in rectal carcinoma.

backgroundsNeoadjuvant chemoradiation (nCRT) is a standard treatment for rectal carcinoma that reduces tumor size and local recurrence while improving the rate of sphincter preservation. However, many patients remain insensitive to nCRT, with some exhibiting tumor progression. To date, there is still a lack of clinically available prognostic models to differentiate the sensitivity of patients with rectal carcinoma to nCRT. This study aimed to develop a genetic predictive model that predicts the effectiveness of nCRT in patients with rectal carcinoma, offering guidance for future treatments and studies on the underlying mechanisms.

methodsBased on the NCBI GEO database datasets, key hub genes affecting the efficacy of neoadjuvant chemoradiotherapy in rectal carcinoma were identified using WCGNA. Subsequently, a consistency analysis of 101 model combinations constructed using ten different machine learning algorithms was performed in two independent cohorts, and a prognostic model (chemoradiation resistance score, CRTR score) was developed and validated. Moreover, the clinical applicability of CRTR in immunotherapy and drug selection was investigated using multi-omics analysis and public databases. Finally, the effect of KIF14 on the radiosensitivity of rectal carcinoma cells was studied using in vitro experiments.

resultsThe CRTR model was composed of 13 genes impacting nCRT sensitivity, whereas six genes (KIF4, DBF4, UBL4A, SLC10A3, PRRG4, and PAPSS2) acted as protective factors and seven (BMS1, DSC2, PROM2, MNAT1, PPID, SMPDL3B, and TNFRSF14) served as risk factors. The CRTR model showed a significant negative correlation with the prognosis of patients with rectal carcinoma undergoing nCRT. Furthermore, patients with higher CRTR values displayed an increased potential to benefit from immunotherapy. Drug sensitivity analysis indicated that aurora kinase inhibitors, telomerase inhibitors, JAK1 inhibitors, and others may enhance the efficacy of nCRT. Finally, we identified KIF14 as the gene that contributed the most to the model and performed preliminary validation. Radiation significantly upregulated the expression of KIF14, and overexpression of KIF14 increased the radiosensitivity of rectal carcinoma cells.

conclusionThe CRTR score, based on 13 genes, was able to predict the prognosis of patients with rectal carcinoma undergoing neoadjuvant chemoradiotherapy and demonstrated immense potential in providing personalized risk assessments and recommendations for targeted immunotherapy. The core gene, KIF14, in the CRTR model may serve as a potential predictive biomarker of radiosensitivity in rectal carcinioma.