DGM Is Dead; Long Live DGM-Star! A Quantization-Theoretic Approach to Gamma Rate Discretization
- 1European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI)
Abstract
Modeling substitution rate heterogeneity as a discretized Gamma distribution is standard in phylogenetic inference. Ferretti et al. (2026) have recently demonstrated that the most commonly used equiprobable Gamma rate discretization, known simply as the discrete Gamma model (DGM), substantially overestimates average branch lengths in reconstructed phylogenies when the true rates are drawn from a continuous Gamma distribution. We propose an alternative method to discretize the Gamma rate distribution that considerably reduces the bias in estimated branch lengths and can be used as a drop-in replacement for the standard DGM. The method works especially well on large trees with short branches, yielding up to a 20-fold reduction in bias on simulated phylogenies. It is developed by applying quantization-theoretic tools to the analysis of a simplified star-phylogeny abstraction introduced by Ferretti et al. (2026), where the new method is unbiased by construction. We explain Ferretti et al.'s counterintuitive observation that increasing the number of rate categories can exacerbate the bias of the standard DGM, even though this should in principle make DGM a better approximation to the continuous case, and offer our method as an immediately available alternative.
Related
- Dead-end filtration of sponge-like colloids : the case of casein micelle
- Homozygous kinase-dead Csf1r mutation in outbred mice reveals essential and redundant functions of tissue resident macrophages
- Cuttlefish use the negation of dead prey to judge food value in foraging decision.
- Amino Acid Deficiency Secondary to Continuous Venovenous Hemofiltration in Acute Decompensation of Organic Acidemias: An Anabolic Dead End?
- DEAD-box RNA helicase 21 interacts with the nucleocapsid protein and inhibits porcine deltacoronavirus infection through upregulation of interferon responses
BibTeX
@misc{sokolovskii2026dead,
title = {{DGM Is Dead; Long Live DGM-Star! A Quantization-Theoretic Approach to Gamma Rate Discretization}},
author = {R. Sokolovskii and C. Stelz and Y. Wang and B. Bettisworth and V. N. Hoang and S. Martin and T. Heinis and O. M. Kozlov and B. Q. Minh and A. Stamatakis and Z. Yang and N. Goldman},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.30.763052},
url = {https://www.biorxiv.org/content/10.64898/2026.10.30.763052v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - DGM Is Dead; Long Live DGM-Star! A Quantization-Theoretic Approach to Gamma Rate Discretization
AU - R. Sokolovskii
AU - C. Stelz
AU - Y. Wang
AU - B. Bettisworth
AU - V. N. Hoang
AU - S. Martin
AU - T. Heinis
AU - O. M. Kozlov
AU - B. Q. Minh
AU - A. Stamatakis
AU - Z. Yang
AU - N. Goldman
PY - 2026
DA - 2026/10/30
PB - bioRxiv
DO - 10.64898/2026.10.30.763052
UR - https://www.biorxiv.org/content/10.64898/2026.10.30.763052v1
AB - Modeling substitution rate heterogeneity as a discretized Gamma distribution is standard in phylogenetic inference. Ferretti et al. (2026) have recently demonstrated that the most commonly used equiprobable Gamma rate discretization, known simply as the discrete Gamma model (DGM), substantially overestimates average branch lengths in reconstructed phylogenies when the true rates are drawn from a continuous Gamma distribution. We propose an alternative method to discretize the Gamma rate distribution that considerably reduces the bias in estimated branch lengths and can be used as a drop-in replacement for the standard DGM. The method works especially well on large trees with short branches, yielding up to a 20-fold reduction in bias on simulated phylogenies. It is developed by applying quantization-theoretic tools to the analysis of a simplified star-phylogeny abstraction introduced by Ferretti et al. (2026), where the new method is unbiased by construction. We explain Ferretti et al.'s counterintuitive observation that increasing the number of rate categories can exacerbate the bias of the standard DGM, even though this should in principle make DGM a better approximation to the continuous case, and offer our method as an immediately available alternative.
ER -