Abstract 1095: Machine learning models predict tumor-specific total mRNA expression associated with clinical outcomes across 14 cancer types

Abstract Understanding unique cancer tumor phenotypes is critical in determining disease progression, drug response, and clinical outcomes. Tumor cell-specific total mRNA expression (TmS) which quantifies total mRNA content within tumor cells, has been shown as a robust biomarker to unravel genetic makeup of tumors and their surrounding environments. Currently, quantifying TmS from transcriptomic data is time consuming and resource intensive. Moreover, the unavailability of matched DNA/RNAseq data with clinical annotations limits the translation of TmS to clinical settings. Machine learning (ML) is powerful to create blackbox models that utilize raw high-dimensional omic data to predict clinical outcomes. In this work, we propose TmSNet, TmS prediction by ensemble Network. It is an ML framework that extracts cancer-type-specific gene expression, methylation, and miRNA profiles from 14 cancer types in TCGA cohort (Train =70%, test =30%) to predict TmS. For each cancer type, we extracted relevant features with an ensemble feature extractor block, using random forest (RF), gradient boosting (GB), lasso, and elastic net (EN) models. The extracted features from each model were used to train an ensemble of ML models consisting of variational auto encoder (VAE), model agnostic meta learning (MAML), support vector machine (SVM), and multi-layer perceptron (MLP). TmSNet achieved strong predictive performance across all cancer types (Table 1). Our findings suggest that TmSNet effectively predicts TmS from limited data, with strong performance and limited overfitting. The predicted TmS values from TmSNet successfully stratifies patient survival within early and late-stage cancer. This can serve as a valuable prognostic tool for multi-omic analyses in cancer, with implications for tumor phenotyping, cancer screening, and therapeutic intervention. Citation Format: Ankita Paul, Shuangxi Ji, Jessica C. Lal, Carissa Fong, Suraksha Vinod, Huiying Hu, Kevin Chen, Matthew Montierth, Quang Tran, Wenyi Wang. Machine learning models predict tumor-specific total mRNA expression associated with clinical outcomes across 14 cancer types [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1095.