Predicting chemotherapy responsiveness in gastric cancer through machine learning analysis of genome, immune, and neutrophil signatures.

Abstract

Background: Gastric cancer is a major oncological challenge, ranking highly among causes of cancer-related mortality worldwide. This study was initiated to address the variability in patient responses to combination chemotherapy, highlighting the need for personalized treatment strategies based on genomic data.

Methods: We analyzed whole-genome and RNA sequences from biopsy specimens of 65 advanced gastric cancer patients before their chemotherapy treatment. Using machine learning techniques, we developed a model with 123 omics features, such as immune signatures and copy number variations, to predict their chemotherapy outcomes.

Results: The model demonstrated a prediction accuracy of 70-80% in forecasting chemotherapy responses in both test and validation cohorts. Notably, tumor-associated neutrophils emerged as significant predictors of treatment efficacy. Further single-cell analyses from cancer tissues revealed different neutrophil subgroups with potential antitumor activities suggesting their usefulness as biomarkers for treatment decisions.

Conclusions: This study confirms the utility of machine learning in advancing personalized medicine for gastric cancer by identifying tumor-associated neutrophils and their subgroups as key indicators of chemotherapy response. These findings could lead to more tailored and effective treatment plans for patients.