WCLC 2025 - Posters & ePosters-P2.06.43 Estimating Tumour Microenvironment Cellular States From Bulk RNAseq Produces Biomarkers of Clinical Outcome Across Stages

P2.06.43 Estimating Tumour Microenvironment Cellular States From Bulk RNAseq Produces Biomarkers of Clinical Outcome Across Stages

Back to course

Pdf Summary

This study addresses the urgent need for predictive biomarkers in lung adenocarcinoma (LUAD), a heterogeneous cancer, by enhancing understanding of the tumor microenvironment (TME). The authors developed CellTFusion, an integrated computational approach combining cell type deconvolution and transcription factor (TF) activity estimation from bulk RNA sequencing data, coupled with advanced machine learning models to extract clinically relevant features. <br /><br />The analysis was performed on a cohort of 62 primary LUAD samples from the IUCT consortium, spanning different disease stages, with validation using an independent early-stage LUAD cohort from Vanderbilt University comprising both bulk and single-cell RNA-seq samples. The study identified distinct cell groups predictive of immunotherapy response in late-stage non-small cell lung cancer (NSCLC) and bladder cancer cohorts.<br /><br />Key findings include: identification of transcriptomic profiles associated with patient survival through clustering of LUAD samples based on TF module, pathway scores, and immune deconvolution features. Significant correlations were found between these features and survival outcomes, supported by multivariate Cox proportional hazards modeling. Additionally, CellTFusion-derived cell groups, particularly involving CD4 T cells, B cells, and M2 macrophages, demonstrated high predictive power in machine learning models (including GLM, RF, SVM, XGBoost) for immunotherapy response in a cohort of 130 advanced NSCLC patients and in a bladder cancer cohort of 192 patients. Performance metrics such as AUC-ROC above 0.7 and feature importance rankings based on SHAP values underscored the robustness of the approach.<br /><br />This research leverages transcriptional regulatory networks and precise immune cell profiling from bulk RNA-seq, improving prediction of therapy responses and survival. Supported by the EUR CARe grant, this methodology has potential to guide personalized immunotherapy treatments in lung cancer by better characterizing patient immune profiles through computational integration of TME features.

Asset Subtitle

Marcelo Hurtado

Meta Tag

Speaker Marcelo Hurtado

Topic Pathology and Biomarkers

Keywords

lung adenocarcinoma

tumor microenvironment

CellTFusion

transcription factor activity

bulk RNA sequencing

machine learning models

immunotherapy response prediction

non-small cell lung cancer

immune cell profiling

transcriptional regulatory networks