Machine learning based intratumor heterogeneity signature for predicting prognosis and immunotherapy benefit in stomach adenocarcinoma

Abstract

Abstract Stomach adenocarcinoma (STAD) is a prevalent malignancy that is highly aggressive and heterogeneous. Intratumor heterogeneity (ITH) showed strong link to tumor progression and metastasis. High ITH may promote tumor evolution. An ITH-related signature (IRS) was created using as integrative technique including 10 machine learning methods based on TCGA, GSE15459, GSE26253, GSE62254 and GSE84437 datasets. The relevance of IRS in predicting the advantages of immunotherapy was assessed using a number of prediction scores and three immunotherapy datasets (GSE78220, IMvigor210 and GSE91061). Vitro experiments were performed to verify the biological functions of AKR1B1. The RSF + Enet (alpha = 0.1) projected model was proposed as the ideal IRS because it had the highest average C-index. The IRS demonstrated a strong performance in serving as an independent risk factor for the clinical outcome of STAD patients. It performed exceptionally well in predicting the overall survival rate of STAD patients, as seen by the TCGA cohort’s AUC of 1-, 3-, and 5-year ROC curves, which were 0.689, 0.683, and 0.669, respectively. A low IRS score demonstrated a superior response to immunotherapy, as seen by a lower TIDE score, lower immune escape score, greater TMB score, higher PD1&CTLA4 immunophenoscore, higher response rate, and improved prognosis. Common chemotherapeutic and targeted treatment regimens had lower IC50 values in the group with higher IRS scores. Vitro experiment showed that AKR1B1 was upregulated in STAD and knockdown of AKR1B1 obviously suppressed tumor cell proliferation and migration. The present investigation produced the best IRS for STAD, which may be applied to prognostication, risk stratification, and therapy planning for STAD patients.

Keywords

• Stomach adenocarcinoma
• Intratumor heterogeneity
• Prognostic signature
• Machine learning
• Immunotherapy