Author: Md Tauhidul Islam, Junyan Liu, Lei Xing 👨🔬
Affiliation: Department of Radiation Oncology, Stanford University 🌍
Purpose: Radiation-induced lung injury (RILI) is a common complication in patients receiving radiotherapy for lung cancer, with approximately 16%–28% developing pulmonary fibrosis. The exact mechanisms why fibrosis occur are not fully understood, making it challenging to predict which patients are at risk after radiation treatment. To personalize treatment and identify patients at risk for pulmonary fibrosis, deep learning models can significantly aid biomarker discovery. To address these challenges, we developed and validated a deep learning framework that decompose bulk RNA sequencing data from lungs into single-cell-like cell components. Our aim is to utilize this deep learning framework to analyze the post-radiation lung microenvironment, identify novel biomarkers associated with pulmonary fibrosis, and improve the accuracy of risk prediction for this condition.
Methods: We obtained publicly available datasets of mouse lungs exposed to 65 Gy or 75 Gy of radiation, comprising both bulk RNA sequencing and scRNA-seq. We analyzed these data using our previously developed “genoMap” pipeline, which converts gene count matrices into image representations based on gene-gene interaction. The scRNA-seq data provided references for key cell types in the lung, i.e., endothelial cells, fibroblasts, myofibroblasts, and immune cells, enabling us to decompose the bulk RNA data into respective cell components. Finally, we compared these cell proportions and gene expression profiles among control, 65 Gy, and 75 Gy groups to identify meaningful differences.
Results: Our initial results demonstrate that genoMaps can effectively visualize and differentiate bulk RNA sequencing data in mouse lungs following radiation exposure. By quantifying the most influential genes in these altered patterns, we identified several fibrosis-related genes, including ADCY8, CHIL3, CSF3R, and TSPAN17.
Conclusion: These findings underscore the promise of our genoMap approach for pinpointing key biomarkers associated with pulmonary fibrosis, paving the way for more accurate early prediction of high-risk patients.