Abstract
AT-rich interactive domain-containing protein 1A (ARID1A) is frequently mutated in endometrial cancers. Although patient stratification based on mutations or mRNA expression is commonly performed, this approach may not accurately reflect the functional state of ARID1A. This functional state is not only directly reflected in upstream events such as gene expression but also influenced by various regulatory including protein expression and the presence and type of mutations. Although protein expression is more directly correlated with phenotypic outcomes, integrating different omics data remains challenging due to disparities in data availability. To address this challenge, we developed a novel patient stratification method that integrates proteomics and transcriptomics to assess the functional state of ARID1A in patients with uterine corpus endometrial carcinoma. Initially, missing protein expression data were imputed using machine learning, and the patients were labelled based on their ARID1A protein expression. We then labelled the patients according to ARID1A activity, inferred by analysing the transcriptional regulation of genes directly controlled by ARID1A. Finally, patients were stratified by ARID1A functional state, considering both protein expression and the inferred activity label. This approach identified different gene expression patterns that are undetectable using conventional methods based on mRNA expression and mutation. Gene set enrichment and over-representation analyses confirmed that the proposed method revealed immune-related differences in patients with ARID1A-deficient uterine corpus endometrial carcinoma. These results highlight its potential to identify novel therapeutic targets and immune alterations that are undetected by conventional techniques.