Integrating machine learning and multi-omics analysis to develop an immune-derived multiple programmed cell death signature for predicting clinical outcomes in gastric cancer

Abstract

Abstract Objectives Metastasis of tumor cells is the leading reason for mortality among patients diagnosed with gastric cancer (GC). Emerging evidence indicated a strong correlation between programmed cell death (PCD) and the invasion and metastasis of tumor cells. Therefore, we aimed to develop a programmed cell death signature to assess the prognosis and therapeutic efficacy in GC patients. Methods Here, we collected 1911 PCD-related genes from 19 different PCD patterns, and developed an immune-derived multiple programmed cell death index (MPCDI) using the integrating machine learning and multi-omics analysis, and systematically dissected heterogeneity in GC patients. Subsequently, we divided GC patients into two categories, namely high-MPCDI group and low-MPCDI group, using the median MPCDI as the threshold. We performed a comprehensive analysis of the clinical characteristics, somatic mutations, immune infiltration, drug sensitivity, and immunotherapeutic efficacy of the two groups. Results Survival and immunotherapy response analyses indicated that the high-MPCDI patients experienced a poorer overall survival (p=0.018) and were more resistant to commonly used chemotherapeutic drugs but benefited from immunotherapy compared to the low-MPCDI patients. In addition, MPCDI was confirmed as a standalone risk factor for overall survival, and nomograms can provide a precise tool for the clinical diagnosis of GC patients. Conclusions Taken together, the MPCDI can serve as a robust clinical diagnostic classifier to guide medication administration and improve outcomes in GC patients.