Geometry based gene expression signatures detect cancer treatment responders in clinical trials

AO_SCPLOWBSTRACTC_SCPLOWO_ST_ABSAimC_ST_ABSThe overall aim of this project is to determine if gene expression signatures of tumors, constructed from geometrical attributes of data, can be used to both create a definitive classification of responders and non-responders, and to predict patient treatment response in an unbiased manner. This is tested in an open-sourced Pfizer clinical trial data on avelumab plus axitnib in advanced renal cell carcinoma (n = 726). ResultsGeometrical gene expression signatures were able to be used to create standardized classification of responders to the intervention, as demonstrated by dramatically different Kaplan-Meier (KM) estimators based on responder category assigned in the Pfizer trial. Furthermore, unbiased prediction based on leave one out methodology was able to correctly predict the responder classification with 82.0% accuracy. Biomarkers of response generated indicated that the strongest predictive gene was PODXL (podocalyxin), with an inconsistent influence on responder class based on over- and underexpression. A KM estimator of the out-of-sample predictions showed nearly four times the average effect in samples predicted to be responders against those predicted to not be responders, and accounted for 79.2% of the treatment effect. ConclusionsGene expression based geometrical signatures are able to create "gold standard" classification of responders and non-responders in clinical trial data, and are highly accurate at predicting these labels in an out-of-sample, unbiased test. These methods can be used to find more stable biomarkers of response, as well as increase the chances of a clinical trial being approved.