Abstract 7591: Identification of a multi-protein signature to predict 5-year stomach cancer risk using high-throughput proteomics and machine learning.

Abstract Introduction: Stomach cancer is the fifth most common cancer worldwide, and the fourth leading cause of cancer death, with an average 5-year relative survival rate of only 20%-30%. This low survival rate is largely due to late diagnosis, which often occurs after the cancer has spread and is more difficult to treat. There is currently no guidance from the US Preventive Services Task Force around screening for stomach cancer or Helicobacter pylori (H pylori) infection, a primary risk factor for stomach cancer. This study investigated whether high-throughput proteomics could be used to develop a stomach cancer susceptibility model that may help stratify individual risk and direct screening procedures. Methods: Samples from the European Prospective Investigation into Cancer and Nutrition study were assayed using modified-aptamer proteomics technology (SomaScan™ 7K assay). Following data QC, 14,787 citrate plasma samples had clinical and proteomic data (totaling ∼103 million protein measurements), which included n=56 and n=219 stomach cancer diagnosis within 5 years and over 20-year follow-up period, respectively. Machine learning techniques were used to identify a model that predicts risk of stomach cancer diagnosis within 5 years of blood draw using an 80% training split of the data. Discrimination was assessed via C-index and 5-year AUC. A post-hoc subset analysis was performed in individuals with H pylori infection status (n=286). Results: An abundance of proteomic signal was detected with 922 proteins significantly associated with incident stomach cancer at FDR 0.1. An 8-protein accelerated failure time model was identified that accurately predicted stomach cancer risk with a 5-year AUC of 0.742 and C-Index of 0.697 in the hold-out validation dataset. Performance of the protein model was higher than the average performance of the Polygenic Risk Score-based models used to predict lifetime stomach cancer risk. The model included proteins known to be up or downregulated in stomach cancer (75%), although only a subset of these proteins (37.5%) are predictive of stomach cancer risk, and novel proteins (25%). In the subset of individuals positive for H pylori infection (n=214), the stomach cancer risk model had a 5-year AUC = 0.698 and C-Index of 0.599 suggesting the model can still accurately predict risk in individuals with H pylori infection. Conclusions: We successfully developed a multi-protein model to predict 5-year risk of stomach cancer in individuals with and without H pylori infection, highlighting the potential utility of high-throughput proteomics as a novel screening tool for assessing cancer risk. Citation Format: Jessica Chadwick, Clare Paterson, Sama Shrestha, Hannah Biegel, Jessica Kuzma, Stephen Williams, In collaboration with the EPIC – Proteomic investigators. Identification of a multi-protein signature to predict 5-year stomach cancer risk using high-throughput proteomics and machine learning abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7591.