Machine Learning and Network Analyses Reveal Disease Subtypes of Pancreatic Cancer and their Molecular Characteristics

Given that the biological processes governing the oncogenesis of pancreatic cancers could present useful therapeutic targets, there is a pressing need to molecularly distinguish between different clinically relevant pancreatic cancer subtypes. To address this challenge, we used targeted proteomics and other molecular data compiled by The Cancer Genome Atlas to reveal that pancreatic tumours can be broadly segregated into two distinct subtypes. Besides being associated with substantially different clinical outcomes, tumours belonging to each of these subtypes also display notable differences in diverse signalling pathways and biological processes. At the proteome level, we show that tumours belonging to the less severe subtype are characterised by aberrant mTOR signalling, whereas those belonging to the more severe subtype are characterised by disruptions in SMAD and cell cycle-related processes. We use machine learning algorithms to define sets of proteins, mRNAs, miRNAs and DNA methylation patterns that could serve as biomarkers to accurately differentiate between the two pancreatic cancer subtypes. Lastly, we confirm the biological relevance of the identified biomarkers by showing that these can be used together with pattern-recognition algorithms to accurately infer the drug sensitivity of pancreatic cancer cell lines. Our study shows that integrative profiling of multiple data types enables a biological and clinical representation of pancreatic cancer that is comprehensive enough to provide a foundation for future therapeutic strategies.