Abstract
Background
Lung adenocarcinoma (LUAD), a predominant subtype of non-small cell lung cancer, exhibits a high mortality rate. To date, no reliable or sensitive biomarkers or prognostic methods have been identified for its early detection or diagnosis. Gamma-aminobutyric acid (GABA), a critical inhibitory neurotransmitter in the central nervous system (CNS), primarily functions by interacting with GABA receptors (GABAR). Recent years have witnessed an increasing recognition of GABA's crucial role in mediating the onset or progression of numerous tumors outside the CNS. However, the research exploring the role of GABA in LUAD remains scant, and its specific molecular and cellular interaction mechanisms are yet to be fully elucidated.
Methods
We developed a new machine learning framework, based on the screening of GABA-related genes at the level of single cells and large transcriptomes. This framework comprises 10 algorithms and their 101 combination pairing patterns, which are used to construct consistent GABA-related features (GABARF). The performance of this framework is evaluated in the training set and external validation set. To provide a quantitative tool for predicting prognosis in clinical practice, we established a nomogram incorporating GABARF. Additionally, we conducted multi-omics analyses, including genomics, single-cell transcriptomics, and whole transcriptomics, to obtain and summarize more comprehensive prognostic features. Lastly, we assessed the response of the GABARF risk subgroups to immunotherapy and screened for personalized drugs for specific risk subgroups.
Results
Among the 124 GABA-related genes we investigated, 38 of these demonstrate a significant correlation with patient overall survival (OS). Our GABARF, which is based on machine learning, performed outstandingly in predicting prognosis and clinical interpretation, and also exhibits potential in predicting the occurrence and progression of LUAD. Multivariate analysis confirmed that GABARF is an independent prognostic factor for the OS of LUAD. Additionally, distinct GABARF risk subgroups exhibit significant differences in biological function, mutation status, and immune infiltration of tumors. Furthermore, significant differences exist in the Immune Phenotype Score (IPS) between the risk subgroups. Through integrating the sensitivity analysis of conventional LUAD drugs, it was found that patients in the low-risk group may benefit more from Immune Checkpoint Inhibitors (ICI) treatment, while patients in the high-risk group may be more sensitive to first-line chemotherapy drugs.
Conclusion
In the present study, a novel machine learning-based model for GABA-related features (GABARF) has been developed. This model serves as a robust tool for the prediction of prognosis, targeted prevention, and individualized treatment planning in lung adenocarcinoma (LUAD). A preliminary investigation into the interaction mechanism of GABARF at the molecular, cellular, and tumor immune microenvironment levels in LUAD has been initiated. This holds great potential to propel future basic research and advancements in the realm of neuro-tumor immunity crossover.