Machine learning-based integration develops a multiple programmed cell death signature for predicting the clinical outcome and drug sensitivity in colorectal cancer

Tumorigenesis and treatment are closely associated with various programmed cell death (PCD) patterns. However, the coregulatory role of multiple PCD patterns in colorectal cancer (CRC) remains unknown. In this study, we developed a multiple PCD index (MPCDI) based on 19 PCD patterns using two machine learning algorithms for risk stratification, prognostic prediction, construction of nomograms, immune cell infiltration analysis, and chemotherapeutic drug sensitivity analysis. As a result, in the TCGA-COAD, GSE17536, and GSE29621 cohorts, the MPCDI can effectively distinguished survival outcomes in CRC patients and served as an independent factor for CRC patients. We then explored the immune infiltration landscape in two groups using the nine algorithms and found more overall immune infiltration in the high-MPCDI group. TIDE scores suggested that the increased immune evasion potential and immune checkpoint inhibition therapy may be less effective in the high-MPCDI group. Immunophenoscores indicated that anti-PD1, anti-cytotoxic T-lymphocyte associated antigen 4 (anti-CTLA4), and anti-PD1-CTLA4 combination therapies are less effective in the high-MPCDI group. In addition, the high-MPCDI group was more sensitive to AZD1332, Foretinib, and IGF1R_3801, and insensitive to AZD3759, AZD5438, AZD6482, Erlotinib, GSK591, IAP_5620, and Picolinici-acid, which suggests that the MPCDI can guide drug selection for CRC patients. As a new clinical classifier, the MPCDI can more accurately distinguish CRC patients who benefit from immunotherapy and develop personalized treatment strategies for CRC patients.