Machine learning models predict the immunotherapy response in tumor based on DNA methylation

Abstract

Background The epigenetic control of immune responses plays a crucial role in the development and progression of cancer. The need to identify biomarkers and create new predictive models is crucial in order to reliably estimate response rates in tumour immunotherapy, which are currently low. Methods We conducted a screening to identify loci that had variable methylation patterns in response to immunotherapy. We next focused on pathways that are relevant to this response and increased their representation.We investigated the expression of methylation loci associated with immunotherapy in tissues.We have also provided a concise overview of the Qtl features associated with several CpG loci.We examined the relationship between the levels of TMB, NeoAg, and PD-L1 and the effectiveness of immunotherapy.Identification of base preferences in DNA sequences by motif analysis allows for the demonstration of unique sequence patterns linked with DNA methylation.We created a total of seven machine learning models, namely Lasso regression, Xgboost, SVM, random forest, KNN, Naive Bayes, and Decision Tree. We then compared their respective functions and choose the best model.. Result The five CpG loci that exhibited the most significant response to tumour immunotherapy were cg00045061, cg00107488, cg00056433, cg00090974, and cg00072957.We identified the immunotherapy-associated pathway, the ubiquitination-proteasome system, by screening differentially methylated sites.Upon analysis, we observed that the majority of the CpG loci that exhibited differential methylation were situated on the N Shore region of the CpG island.The GO enrichment analysis identified the top two pathways as modulation of microvillus length and CXCR4 chemokine receptor binding.On the whole the Random Forest model is considered the optimal choice for machine learning((Precision: 0.859,F1score: 0.907.Recalling: 0.941,ROC: 0.654). Conclusion Tumour methylation sites have the potential to be used as biomarkers for predicting the effectiveness of tumour immunotherapy and for future clinical applications.The Random Forest model is the most optimal choice among many machine learning algorithms for predicting methylation sites in immunotherapy.