Abstract 6283: Performance evaluation of tumor microenvironment deconvolution and gene prediction methods in lung cancer

Abstract The tumor microenvironment (TME) is complex and plays a key role in cancer development, progression and metastasis. Two layers of information are critical for understanding tumor compositions (1) the proportion of each cell type (2) the level of gene expression in each cell type. Many single-cell profile-assisted algorithms have been developed to infer cell type proportions from bulk RNA-seq data, including CIBERSORTx, MuSic and Scaden, and there are several benchmark studies published to compare the methods. However, most of the deconvolution methods do not support prediction of gene expression in a heterogenous population of tumor cells, and there’s no published study systemically comparing the gene expression prediction performance of the methods. To overcome the limitations, we simulate large-scale of pseudobulk data using scRNA-seq profiling from lung tumors, representing various levels of tumor purities and cell lineages, to compare the cell fraction and gene expression prediction performance of recently published TME deconvolution methods. Our results indicate BayesPrism is the best at predicting gene expression in each of the cell types, and differential gene expression between two groups across cell types, especially in tumor epithelial cells, with higher balanced accuracy and correlation with ground truth. The inclusion of normal epithelial cells in the reference could increase additional risks for tumor deconvolution. Altogether we evaluate both cell type proportion and gene expression prediction performance of recently published methods and offer a guideline for future benchmarking work. Currently we're working on the downstream application of inferring cellular composition and gene expression in large cohorts of bulk RNA-seq data, allowing for more insights into tumor-intrinsic expression and tumor-microenvironment interactions. Citation Format: Fanying Tang, Dean Lee, Lexiang Ji, Jincheng Wu. Performance evaluation of tumor microenvironment deconvolution and gene prediction methods in lung cancer [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 6283.