Integrating multi-omics analysis and machine learning to refine molecular subtypes and prognostic assessment of lower-grade glioma.

Lower-grade glioma (LGG) is a highly heterogeneous disease, making accurate prognosis prediction and the development of precise, personalized treatment plans for patients challenging. The multi-omics data from patients with LGG were analyzed using 10 clustering algorithms to identify subgroups at very high resolution. Ten machine learning algorithms were subsequently integrated to develop a robust artificial intelligence model (AIM). We identified two cancer subtypes (CSs) linked to prognosis through multi-omics clustering. After screening 46 hub genes, we integrated 10 machine learning algorithms into 117 combinations to select the AIM with the highest average C-index. The gradient boosting machine (GBM)-based AIM outperformed previous prognostic signatures in nearly all cohorts. Patients in the low-AIM group had a better prognosis and greater sensitivity to immunotherapy, whereas those in the high-AIM group had a poorer prognosis and lower immunotherapy sensitivity. However, MG132 showed promise as a potential therapeutic agent. In vitro studies confirmed the oncogenic role of the hub gene CSDC2 in LGG cells; its knockdown reduced the proliferation, invasion, and migration of these cells. In summary, the AIM we developed is highly valuable for predicting the prognosis of patients with LGG and identifying those who are sensitive to immunotherapy.