Letter to the Editor: on the stability and internal consistency of component-wise sparse mixture regression-based clustering

Abstract

Understanding the relationship between molecular markers and a phenotype of interest is often obfuscated by patient-level heterogeneity. To address this challenge, Chang et al. recently published a novel method called Component-wise Sparse Mixture Regression (CSMR), a regression-based clustering method that promises to detect heterogeneous relationships between molecular markers and a phenotype of interest under high-dimensional settings. In this Letter to the Editor, we raise awareness to several issues concerning the assessment of CSMR in Chang et al., particularly its assessment in settings where the number of features, P, exceeds the study sample size, N, and advocate for additional metrics/approaches when assessing the performance of regression-based clustering methodologies.

Keywords: disease heterogeneity; mixture modeling; supervised learning.

Figure 1

Performance of CSMR on the CCLE data set. (A) Boxplot of clustering IC, calculated as the ARI of cluster memberships between each pair of runs, from 100 separate applications of CSMR to the CCLE data set using the same tuning parameters for each run. (B) Bar plot of frequency of specific features being selected by CSMR from 100 separate applications of CSMR to the CCLE data set. Out of the 500 considered features, 425 features were selected by CSMR at least once. On average, CSMR selected 43.6 features each iteration (standard deviation = 11.7).

Figure 2

CSMR model performance in various simulation settings. Boxplots showing performance metric values when . The x-axis shows at 100, 200 and 300, and the colors indicate . As and P increase, both accuracy and IC decline.