Identification and exploration of the pyroptosis-related molecular subtypes of breast cancer by bioinformatics and machine learning

Objectives: To classify breast cancer (BRCA) according to the expression of pyroptosis-related genes and explore their molecular characteristics. Methods: Nonnegative matrix factorization (NMF) was used for subtype classification based on 21 pyroptosis-related genes in the TCGA database. Survival analysis and t-distributed stochastic neighbor embedding (t-SNE) analysis were conducted to assess the NMF results' performance. XGBoost, CatBoost, logistic regression, neural network, random forest, and support vector machine were utilized to perform supervised machine learning and construct prediction models. Genetic mutations, tumor mutational burden, immune infiltration, methylation, and drug sensitivity were analyzed to explore the molecular signatures of different subtypes. Lasso, RF, and Cox regression were operated to construct a prognostic model based on differentially expressed genes. Results: BRCA patients were divided into two subtypes (named Cluster1 and Cluster2). Survival analysis (P = 0.02) and t-SNE analysis demonstrated that Cluster1 and Cluster2 were well classified. The XGBoost model achieved reliable predictions on both training and validation sets. Regarding molecular characteristics, Cluster1 had higher TMB, immune cell infiltration, and m(6)A methylation-related gene expression than Cluster2. There was also a statistically significant difference between the two subtypes concerning drug susceptibility. Finally, a 5-gene prognostic model was constructed using Lasso, RF, and Cox regression and validated in the GEO database. Conclusion: Our study may provide new insights from bioinformatics and machine learning for exploring pyroptosis-related subtypes and their respective molecular signatures in BRCA. In addition, our models may be helpful for the treatment and prognosis of BRCA.