PRADclass: Multi-pronged Gleason grade-informed computational strategy identifies consensus biomarker features of prostate adenocarcinoma that predict aggressive cancer

Abstract

ABSTRACTBackgroundProstate adenocarcinoma (PRAD) is the most common cancer in men worldwide, yet gaps in our knowledge persist with respect to molecular bases of PRAD progression and aggression. It is largely an indolent cancer, asymptomatic at early stage, and slow-growing in most cases, but aggressive prostate cancers cause significant morbidity and mortality within five years. Automated methods to type the aggressiveness of PRAD are necessary and urgent for informed treatment management.MethodsBased on TCGA transcriptomic data pertaining to PRAD and the associated clinical metadata, we used the grading guidelines of the International Society of Urological Pathology (ISUP), and converted the clinical information of a cancer sample to its Gleason grade. To model the distinction between aggressive prostate cancers (Gleason grade IV or V) and indolent prostate cancers (Gleason grade I or II), we performed: (i) Gleason-grade wise linear modeling, followed by five contrasts against controls and ten contrasts between grades; and (ii) Gleason-grade wise network modeling using weighted gene correlation network analysis (WGCNA). Consensus between the grade-salient genes from the statistical modeling and the trait-specific key genes from network modeling were used as features for learning a ternary classification: benign, indolent or aggressive malignancy.ResultsThe statistical modeling yielded 77 Gleason grade-salient genes, viz. ten genes in grade-1, two genes in grade-II, one gene in grade-III, 34 genes in grade-IV, and 30 genes in grade-V. Using the WGCNA method, we reconstructed grade-specific networks, and defined trait-specific key genes in grade-wise significant modules. Consensus analysis yielded two genes in Grade 1 (SLC43A1, PHGR1), 26 genes in Grade 4 (LOC100128675, PPP1R3C, NECAB1, UBXN10, SERPINA5, CLU, RASL12, DGKG, FHL1, NCAM1), and seven genes in Grade 5 (CBX2, DPYS, FAM72B, SHCBP1, TMEM132A, TPX2, UBE2C). PRADclass, a RandomForest model trained on these 35 consensus biomarkers, yielded 100% cross-validation accuracy on the ternary classification problem.ConclusionsConsensus of orthogonal computational strategies has yielded Gleason grade-specific biomarkers that are useful in pre-screening (cancer vs normal) as well as typing the aggressiveness of cancer. PRADclass has been deployed at:https://apalania.shinyapps.io/pradclass/for scientific and non-commercial use.