Fig. 3
From: A Dirichlet-multinomial mixed model for determining differential abundance of mutational signatures
Comparison of coefficients that represent differential abundance for the models under consideration: diagREDM, HiLDA, and TCSM, for simulated dataset C1. Although diagREDM shows a marked increase in dispersion of softmax-transformed \(\widehat{\varvec{\beta }}_1\) across signatures (indicating differential abundance) and the coefficients of TCSM show an increased deviation from zero (also indicating differential abundance), the results from HiLDA do not show such trends unless the mixing proportion \(\pi\) is very high. In the case of diagREDM, the signatures represent the ground truth signatures used for simulation (SBS1, SBS5, SBS9, SBS40, SBS3, SBS13)