Protein language models (PLMs) are often assumed to capture evolutionary information by training on large protein sequence datasets. Yet it remains unclear whether PLMs can reason about evolution—that is, infer evolutionary relationships between sequences. We test this capability by evaluating whether standard PLM usage, frozen or fine-tuned embeddings with distance-based comparison, supports evolutionary reasoning. Existing PLMs consistently fail to recover phylogenetic structure, despite strong performance on sequence-level tasks such as masked-token and contact prediction. We present Phyla, a hybrid state-space and transformer model that jointly processes multiple sequences and is trained using a tree-based objective across 3,000 phylogenies spanning diverse protein families. Phyla outperforms the next-best PLM by 9\% on tree reconstruction and 23\% on taxonomic clustering while remaining alignment- and guide-tree-free. Although classical alignment pipelines achieve higher absolute accuracy, Phyla narrows the gap and achieves markedly lower end-to-end runtime. Applied to real data, Phyla reconstructs biologically accurate clades in the tree of life and resolves genome-scale relationships among Mycobacterium tuberculosis isolates. These findings suggest that, under standard usage, evolutionary reasoning does not reliably emerge from large-scale sequence modeling. Instead, Phyla shows that models trained with phylogenetic supervision can reason about evolution more effectively, offering a biologically grounded path toward evolutionary foundation models.

SUMMARY: Pan-genome analysis is a fundamental tool for studying bacterial genome evolution; however, the variety in methods used to define and measure the pan-genome poses challenges to the interpretation and reliability of results. Using Mycobacterium tuberculosis, a clonally evolving bacterium with a small accessory genome, as a model system, we systematically evaluated sources of variability in pan-genome estimates. Our analysis revealed that differences in assembly type (short-read versus hybrid), annotation pipeline, and pan-genome software, significantly impact predictions of core and accessory genome size. Extending our analysis to two additional bacterial species, Escherichia coli and Staphylococcus aureus, we observed consistent tool-dependent biases but species-specific patterns in pan-genome variability. Our findings highlight the importance of integrating nucleotide- and protein-level analyses to improve the reliability and reproducibility of pan-genome studies across diverse bacterial populations. AVAILABILITY AND IMPLEMENTATION: Panqc is freely available under an MIT license at https://github.com/maxgmarin/panqc.

MOTIVATION: The gene content regulates the biology of an organism. It varies between species and between individuals of the same species. Although tools have been developed to identify gene content changes in bacterial genomes, none is applicable to collections of large eukaryotic genomes such as the human pangenome. RESULTS: We developed pangene, a computational tool to identify gene orientation, gene order, and gene copy-number changes in a collection of genomes. Pangene aligns a set of input protein sequences to the genomes, resolves redundancies between protein sequences and constructs a gene graph with each genome represented as a walk in the graph. It additionally finds subgraphs, which we call bibubbles, that capture gene content changes. Applied to the human pangenome, pangene identifies known gene-level variations and reveals complex haplotypes that are not well studied before. Pangene also works with high-quality bacterial pangenome and reports similar numbers of core and accessory genes in comparison to existing tools. AVAILABILITY AND IMPLEMENTATION: Source code at https://github.com/lh3/pangene; prebuilt pangene graphs can be downloaded from https://zenodo.org/records/8118576 and visualized at https://pangene.bioinweb.org.

MOTIVATION: Short-read whole-genome sequencing (WGS) is a vital tool for clinical applications and basic research. Genetic divergence from the reference genome, repetitive sequences and sequencing bias reduces the performance of variant calling using short-read alignment, but the loss in recall and specificity has not been adequately characterized. To benchmark short-read variant calling, we used 36 diverse clinical Mycobacterium tuberculosis (Mtb) isolates dually sequenced with Illumina short-reads and PacBio long-reads. We systematically studied the short-read variant calling accuracy and the influence of sequence uniqueness, reference bias and GC content. RESULTS: Reference-based Illumina variant calling demonstrated a maximum recall of 89.0% and minimum precision of 98.5% across parameters evaluated. The approach that maximized variant recall while still maintaining high precision (<99%) was tuning the mapping quality filtering threshold, i.e. confidence of the read mapping (recall = 85.8%, precision = 99.1%, MQ ≥ 40). Additional masking of repetitive sequence content is an alternative conservative approach to variant calling that increases precision at cost to recall (recall = 70.2%, precision = 99.6%, MQ ≥ 40). Of the genomic positions typically excluded for Mtb, 68% are accurately called using Illumina WGS including 52/168 PE/PPE genes (34.5%). From these results, we present a refined list of low confidence regions across the Mtb genome, which we found to frequently overlap with regions with structural variation, low sequence uniqueness and low sequencing coverage. Our benchmarking results have broad implications for the use of WGS in the study of Mtb biology, inference of transmission in public health surveillance systems and more generally for WGS applications in other organisms. AVAILABILITY AND IMPLEMENTATION: All relevant code is available at https://github.com/farhat-lab/mtb-illumina-wgs-evaluation. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Metabolism is central to cell physiology, and metabolic disturbances play a role in numerous disease states. Despite its importance, the ability to study metabolism at a global scale using genomic technologies is limited. In principle, complete genome sequences describe the range of metabolic reactions that are possible for an organism, but cannot quantitatively describe the behaviour of these reactions. We present a novel method for modeling metabolic states using whole cell measurements of gene expression. Our method, which we call E-Flux (as a combination of flux and expression), extends the technique of Flux Balance Analysis by modeling maximum flux constraints as a function of measured gene expression. In contrast to previous methods for metabolically interpreting gene expression data, E-Flux utilizes a model of the underlying metabolic network to directly predict changes in metabolic flux capacity. We applied E-Flux to Mycobacterium tuberculosis, the bacterium that causes tuberculosis (TB). Key components of mycobacterial cell walls are mycolic acids which are targets for several first-line TB drugs. We used E-Flux to predict the impact of 75 different drugs, drug combinations, and nutrient conditions on mycolic acid biosynthesis capacity in M. tuberculosis, using a public compendium of over 400 expression arrays. We tested our method using a model of mycolic acid biosynthesis as well as on a genome-scale model of M. tuberculosis metabolism. Our method correctly predicts seven of the eight known fatty acid inhibitors in this compendium and makes accurate predictions regarding the specificity of these compounds for fatty acid biosynthesis. Our method also predicts a number of additional potential modulators of TB mycolic acid biosynthesis. E-Flux thus provides a promising new approach for algorithmically predicting metabolic state from gene expression data.