Improved Variational Bayesian Phylogenetic Inference using Mixtures

Authors: Ricky Molén, Oskar Kviman, Jens Lagergren

TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Across eight real phylogenetic datasets and compared to the considered benchmarks, we show that VBPI-Mixtures result in lower-variance estimators of the marginal log-likelihood and smaller KL divergences to an MCMC-based approximation of the true tree-topology posterior.
Researcher Affiliation Academia Ricky Molén EMAIL KTH Royal School of Technology Science for Life Laboratory Oskar Kviman EMAIL KTH Royal School of Technology Science for Life Laboratory Jens Lagergren EMAIL KTH Royal School of Technology Science for Life Laboratory
Pseudocode No The paper describes the proposed method and derivations in text and mathematical equations, but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code for all experiments is provided at Github.
Open Datasets Yes We performed experiments on eight datasets (Hedges et al., 1990; Garey et al., 1996; Yang & Yoder, 2003; Henk et al., 2003; Lakner et al., 2008; Zhang & Blackwell, 2001; Yoder & Yang, 2004; Rossman et al., 2001) which we will refer to as DS1-8.
Dataset Splits No The paper refers to using datasets DS1-8 for experiments and mentions gathering candidate trees from ultrafast maximum likelihood bootstrap trees, but it does not specify explicit training, validation, or test splits for these datasets in the context of the VBPI models.
Hardware Specification Yes Most computations have been conducted on an AMD EPYC 7742 where two cores have been used per run. ... The Ufboot2 was run on i9-13900k
Software Dependencies No The paper mentions specific software and models like Mr Bayes, UFBoot, iqtree2, and the Jukes-Cantor 69 model, but it does not provide specific version numbers for these or any other general software dependencies (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes we trained all VBPI models during 400,000 iterations, using the same hyperparameter settings as Zhang & Matsen IV (2019); Zhang (2020). Based on the study in Zhang & Matsen IV (2024), we let K = 10 during training. ... All models are trained for 60k epochs, with decaying learning rates by 0.9 every 10k epochs.