Predicting Cancer Prognostics from Tumour Transcriptomics Using an Auto Machine Learning Approach

Abstract

Cancer prognostics using tumour transcriptomics is a promising precision medicine approach for helping decisions during cancer treatment. However, currently used cancer prognostic biomarkers still have low predictive power. This work tested the potential of applying machine learning (ML) algorithms for generating patients’ survival prognostics on lung, breast, and kidney tumour transcriptomics datasets. We evaluated the performance of models generated by ML and reported their optimal sensitivity, specificity, accuracy, and computed ROC-AUC. The results support the potential for applying auto ML approaches for the future development of cancer prognostics tools based on transcriptomics data.

Keywords

• bioinformatics
• cancer prognostics
• machine learning
• transcriptomics