Dualmarker: a flexible toolset for exploratory analysis of combinatorial dual biomarkers for clinical efficacy

Xiaopeng Ma,Pei Zhang,Xikun Wu,Ruiqi Huang

doi:10.1186/s12859-021-04050-6

Abstract

BackgroundAn increasing number of clinical trials require biomarker-driven patient stratification, especially for revolutionary immune checkpoint blockade therapy. Due to the complicated interaction between a tumor and its microenvironment, single biomarkers, such as PDL1 protein level, tumor mutational burden (TMB), single gene mutation and expression, are far from satisfactory for response prediction or patient stratification. Recently, combinatorial biomarkers were reported to be more precise and powerful for predicting therapy response and identifying potential target populations with superior survival. However, there is a lack of dedicated tools for such combinatorial biomarker analysis.ResultsHere, we present dualmarker, an R package designed to facilitate the data exploration for dual biomarker combinations. Given two biomarkers, dualmarker comprehensively visualizes their association with drug response and patient survival through 14 types of plots, such as boxplots, scatterplots, ROCs, and Kaplan–Meier plots. Using logistic regression and Cox regression models, dualmarker evaluated the superiority of dual markers over single markers by comparing the data fitness of dual-marker versus single-marker models, which was utilized for de novo searching for new biomarker pairs. We demonstrated this straightforward workflow and comprehensive capability by using public biomarker data from one bladder cancer patient cohort (IMvigor210 study); we confirmed the previously reported biomarker pair TMB/TGF-beta signature and CXCL13 expression/ARID1A mutation for response and survival analyses, respectively. In addition, dualmarker de novo identified new biomarker partners, for example, in overall survival modelling, the model with combination of HMGB1 expression and ARID1A mutation had statistically better goodness-of-fit than the model with either HMGB1 or ARID1A as single marker.ConclusionsThe dualmarker package is an open-source tool for the visualization and identification of combinatorial dual biomarkers. It streamlines the dual marker analysis flow into user-friendly functions and can be used for data exploration and hypothesis generation. Its code is freely available at GitHub at https://github.com/maxiaopeng/dualmarker under MIT license.

Highlights

An increasing number of clinical trials require biomarker-driven patient stratification, especially for revolutionary immune checkpoint blockade therapy
For gene expression profiling (GEP), we focused on the tumor microenvironment-related genes (1392) listed in the HTG Precision immune-oncology panel [12]
Using the population median as the cutoff point, patients were stratified into 4 groups/quadrants, and a comparison of the 4 group sizes showed weak dependence between tumor mutation burden (TMB) and the TGF-beta signature (p value = 0.06, Fisher exact test, Fig. 2d)

Summary

Introduction

An increasing number of clinical trials require biomarker-driven patient stratification, especially for revolutionary immune checkpoint blockade therapy. Due to the complicated interaction between a tumor and its microenvironment, single biomarkers, such as PDL1 protein level, tumor mutational burden (TMB), single gene mutation and expression, are far from satisfactory for response prediction or patient stratification. In the revolutionary field of immune checkpoint blockade (ICB) [1,2,3], such biomarkers include PDL1 protein level, gene expression profiling (GEP), gene mutation, and tumor mutation burden (TMB). A third study using the same dataset found that AIRD1A mutation in combination with CXCL13 gene expression predicted overall survival (OS) [6] In all these studies, combinatorial dual markers outperformed single markers in the prediction of response and survival

Results

Discussion

Conclusion