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

Abstract

AbstractAimThe overall aim of this project is to determine if gene expression signatures of tumors, constructed from geometrical attributes of data, could be used to predict patient treatment response by detecting subgroups of responders. This is tested in Pfizer clinical trial data and compared with standard clustering methods (n= 726).ResultsGeometrical gene expression signature analysis demonstrated high utility to detect sub-groups with enhanced treatment response. In the Pfizer trial, gene expression signatures were able to detect three subgroups of responders (p= 0.012), containing 52.9% of patients and accounting for nearly all the observed treatment effect. Standard techniques following a similar methodology were able to partition a single subgroup containing 21.3% of patients.ConclusionsGene expression based geometrical signatures yielded vastly superior performance over standard clustering techniques, as demonstrated in Pfizer’s Phase III clinical trial data. These can be used to determine subgroups of enhanced treatment response in oncology clinical trials, and might lead to personalized treatment recommendations in the future.