LUADnet: a deep learning model for prediction of clinical outcomes in lung adenocarcinoma based on gene expression
Cheng Cheng1,2, Zhanlue Liang1,2, Renjie Xu1,2
1Department of Pulmonary and Critical Care Medicine, West China Hospital, State Key Laboratory of Respiratory Health and Multimorbidity, Sichuan University, Chengdu, China.
Background:
Lung adenocarcinoma (LUAD) is the most prevalent subtype of non-small cell lung cancer (NSCLC), and immunochemotherapy is widely utilized in its treatment. However, there are several drawbacks, including immune escape, immune-related adverse events (irAEs), a significant economic burden, and unfavorable outcomes. Therefore, it is crucial to identify patients who are likely to respond to non-immunotherapy. This study aimed to identify key transcriptomic features that distinguish responders from non-responders to non-immunotherapy in LUAD and to develop a deep learning model for effective patient stratification.
Methods:
Differentially expressed gene (DEG) analysis and functional enrichment analysis were performed on the transcript profile of responders and non-responders from The Cancer Genome Atlas (TCGA). LUADnet model was developed using gene expression data, which integrates the global feature extraction module (GFEM), the local feature extraction module (LFEM), and the channel selection module (CSM), followed by the classifier module.
Results:
We identified 814 upregulated and 492 downregulated DEGs. Functional enrichment analysis revealed that the upregulated DEGs were enriched in pathways related to immune activation, transmembrane transport, and G protein-coupled receptor (GPCR) signaling, while the downregulated DEGs were primarily associated with the cell cycle pathway. The LUADnet demonstrated superior performance compared to state-of-the-art models, achieving an overall F1-score of 0.9193 and an accuracy of 0.9206. Additionally, ablation experiments indicate that the combination of LFEM, GFEM, and CSM enhances overall classification performance.
Conclusions:
This study explored the tumor microenvironment of responding and non-responding LUAD patients following non-immunotherapy and developed a robust deep learning model to stratify these patients, which will facilitate therapeutic strategies and alleviate economic burdens.
