Ensemble Machine Learning Approaches for Proteogenomic Cancer Studies

Abstract Background: The identification of important proteins is critical for medical diagnosis and prognosis in common diseases. Diverse sets of computational tools were developed for omics data reductions and protein selections. However, standard statistical models with single feature selection involve the multi-testing burden of low power with the available limited samples. Furthermore, high correlations among proteins with high redundancy and moderate effects often lead to unstable selections and cause reproducibility issues. Ensemble feature selection in machine learning may identify a stable set of disease biomarkers that could improve the prediction performance of subsequent classification models, and thereby simplify their interpretability. In this study, we developed a three-stage homogeneous ensemble feature selection approach for both identifying proteins and improving prediction accuracy. This approach was implemented and applied to ovarian cancer proteogenomics data sets: 1) binary putative homologous recombination deficiency positive or negative; and 2) multiple mRNA classes (differentiated, proliferative, immunoreactive, mesenchymal, and unknown). We conducted and compared various machine learning approaches with homogeneous ensemble feature selection including random forest, support vector machine, and neural network for predicting both binary and multiple class outcomes. Various performance criteria including sensitivity, specificity, kappa statistics were used to assess the prediction consistency and accuracy. Results: With the proposed three-stage homogeneous ensemble feature selection approaches, prediction accuracy can be improved with the limited sample through continuously reducing errors and redundancy, i.e. Treebag provided 83% prediction accuracy (85% sensitivity and 81% specificity) for binary ovarian outcomes. For mRNA multi-classes classification, our approach provided even better accuracy with increased sample size. Conclusions: Despite the different prediction accuracies from various models, homogeneous ensemble feature selection proposed identified consistent sets of top ranked important markers out of 9606 proteins linked to the binary disease and multiple mRNA class outcomes.