Machine Learning Based Identification of an Amino Acid Metabolism Related Signature for Predicting Prognosis and Immune Microenvironment in Pancreatic Cancer

Background: Pancreatic cancer is a highly aggressive neoplasm characterized by poor diagnosis. Amino acids play a prominent role in the occurrence and progression of pancreatic cancer as essential building blocks for protein synthesis and key regulators of cellular metabolism. Understanding the interplay between pancreatic cancer and amino acid metabolism offers potential avenues for improving patient clinical outcomes.

Methods: A comprehensive analysis integrating 10 machine learning algorithms was executed to pinpoint amino acid metabolic signature. The signature was validated across both internal and external cohorts. Subsequent GSEA was employed to unveil the enriched gene sets and signaling pathways within high- and low-risk subgroups. TMB and drug sensitivity analyses were carried out via Maftools and oncoPredict R packages. CIBERSORT and ssGSEA were harnessed to delve into the immune landscape disparities. Single-cell transcriptomics, qPCR, and Immunohistochemistry were performed to corroborate the expression levels and prognostic significance of this signature.

Results: A four gene based amino acid metabolic signature with superior prognostic capabilities was identified by the combination of 10 machine learning methods. It showed that the novel prognostic model could effectively distinguish patients into high- and low-risk groups in both internal and external cohorts. Notably, the risk score from this novel signature showed significant correlations with TMB, drug resistance, as well as a heightened likelihood of immune evasion and suboptimal responses to immunotherapeutic interventions.

Conclusion: Our findings suggested that amino acid metabolism-related signature was closely related to the development, prognosis and immune microenvironment of pancreatic cancer.