Application of Machine Learning Models Based on H&E Staining for Mitochondrial-Related Genes Classification and Prognosis of Lung Adenocarcinoma

Lung adenocarcinoma (LUAD) is a leading cause of cancer-related mortality worldwide, necessitating the identification of reliable prognostic biomarkers. This study innovatively combines mitochondrial-related genes classification with pathological histology. We developed machine learning models to predict LUAD prognosis with improved accuracy.We utilized RNA sequencing data from 443 LUAD patients and paired H&E-stained pathological images from 327 patients, both sourced from The Cancer Genome Atlas (TCGA) database. Using a non-negative matrix factorization (NMF) clustering algorithm, we clustered 213 mitochondrial-related genes into high-risk (Cluster 1) and low-risk (Cluster 2) categories.Survival analysis confirmed that the high-risk group is an independent risk factor for overall survival (OS) (HR = 1.463, p  = 0.031). In parallel, we constructed an eight-feature pathomics model using various machine learning techniques, achieving a strong predictive performance with an area under the curve (AUC) of 0.836. The pathomics score (PS) derived from this model was identified as an independent prognostic factor (HR = 1.686, p  = 0.013). Moreover, functional enrichment analysis revealed that the high-PS group is associated with several critical pathways and alterations, including activation of the G2M checkpoint pathway, upregulated lysine degradation, reduced resting dendritic cell infiltration, TP53/TTN mutations, and increased tumor mutation burden.This study represents a novel integration of mitochondrial metabolic genes classification with pathological histology, demonstrating that pathomics features indicative of mitochondrial subtypes serve as potential prognostic markers for LUAD and might suggest promising avenues for personalized treatment strategies.