Advancing prognostic and therapeutic prediction in lung squamous cell carcinoma through integrated multi-omics...

Lung squamous cell carcinoma (LUSC) is a highly malignant cancer with a poor prognosis. This study aimed to develop a precision treatment strategy for LUSC by leveraging 117 machine learning (ML) combinations and integrating multicenter, multi-omics data.

We integrated four-dimensional data from The Cancer Genome Atlas (TCGA), including messenger RNA (mRNA), long non-coding RNA (lncRNA), somatic mutations, and DNA methylation. Using 10 clustering algorithms, stable multi-omics consensus clusters (CCs) were identified. The optimal prognosis signature (OPS) was constructed based on the CC-related genes and 117 ML algorithms in early- and advanced-stage LUSC, with data sourced from TCGA, GSE157011, GSE73403, GSE37745, and GSE14814 cohorts. The efficacy of the OPS was assessed from multiple perspectives, including enrichment pathway, tumor microenvironment (TME), drug sensitivity, and single-cell analyses.

We comprehensively identified two CCs of LUSC, each exhibiting distinct molecular characteristics. CC1 was primarily associated with a poor prognosis, active tumor proliferation, and aggressive features. In early-stage LUSC, the OPS was developed based on 14 up-regulated genes in CS1. The prognostic prediction power of the OPS surpassed that of most prognostic signatures. The OPS was associated with cancer progression and variation in the TME. AZD6482_2169 and Selumetinib_1736 were identified as potential therapeutic agents for the high OPS group. The single-cell analysis suggested that the OPS-related genes were predominantly expressed in progenitor cells. In advanced-stage LUSC, the OPS was developed based on four up-regulated genes in CS1. The OPS also outperformed most prognostic signatures in terms of its prognostic accuracy. Additionally, sepantronium bromide_1941 was identified as a potential therapeutic agent for the high OPS group. The single-cell analysis suggested that the OPS-related genes were predominantly expressed in epithelial cells.

This study developed a novel approach for the precise and individualized treatment of LUSC patients using advanced ML and multi-omics integration technology.