IBCN 2020: An Evaluation of Single-Sample Tumor Subtype Classification Methods

(UroToday.com) Given the recent large efforts to molecularly subclassify bladder cancer, there has been a growing emphasis to ensure external validation and appropriate classification. Researchers from Lund University, Sweden, used bioinformatics tools including rule-based and centroid based algorithms in an attempt to identify methods that can reproducibly classify a newly profiled bladder tumor samples in a manner that is independent of preprocessing, sample purity, and mRNA assay technology.


Using external datasets, they noted that rule-based methods (for example, Gene A>Gene B indicates Subtype X), particularly using iterative random forest algorithms, had more distinct subtype calls, and outperformed other traditional methods (Figure). On the other hand, single sample centroid classifiers based on raw expression values suffered from poor separation between predictions that performed progressively worse as more genes were included.

Figure: Performance of single sample predictor.


IBCN_mRNA_assay.png



Overall, the random forest algorithm emerged as a promising new method for single sample classification, being robust even when using few genes, which allowed for evaluation of class stability and putative incorrect reference labels during training.

A challenge noted by the authors was deciding the selection of specific gene-rules to use in the classifier. During the discussion, others also highlighted that another potential shortcoming is the current inability to adjust for tumor heterogeneity when applying such classifiers.

Presented by: Pontus Eriksson, PhD, Lund Bladder Cancer Group, Lund University, Sweden.

Written by: Anirban P. Mitra, MD, PhD, Urologic Oncology Fellow, The University of Texas MD Anderson Cancer Center, Houston, TX, USA, Twitter: @APMitra, with Ashish M. Kamat, MD, MBBS, President of IBCN and IBCG, Endowed Professor, The University of Texas MD Anderson Cancer Center, Houston, TX, USA, Twitter: @UroDocAsh  at the International Bladder Cancer Network (IBCN) Annual Meeting, #IBCN2020, October 17, 2020.
email news signup