Biology subjects

Cornman, A.

Publications and source records attributed to Cornman, A..

5 recordsLinked to original sources

bioRxiv2026

NPannotator: a genome- and chemistry- constrained automation for type I polyketide synthase pathway elucidation

Natural products (NPs) are structurally diverse bioactive compounds whose biosynthesis is encoded within biosynthetic gene clusters (BGCs). Although databases such as the Minimum Information about a Biosynthetic Gene Cluster (MiBIG) repository now catalog thousands of experimentally validated NP structures, the full biosynthetic pathway connecting individual domain sequences to specific chemical features on final NP structures remains largely unannotated. This gap is especially pronounced for type I polyketide synthases (PKSs). These are modular assembly lines in which multiple enzymatic domains work in concert to condense acyl-CoA building blocks into complex polyketide scaffolds. Within these systems, acyltransferase (AT) domains govern which starter and extender units are incorporated at each elongation step, yet the substrate specificities of AT domains are known for only a fraction of cataloged clusters. Moreover, the catalytic order of genes encoding PKS modules is not immediately apparent from existing database entries, leaving the correct module ordering for observed product structures uncertain. Here, we present NPannotator, an automated, genomic context-aware cheminformatics pipeline that infers both the catalytic ordering of PKS domains and the substrate specificities of a given PKSs AT domains. NPannotator loads a precomputed database of synthetically generated polyketide backbones, iteratively replaces default malonyl-CoA substrates with candidate starter and extender units via SMARTS-based substructure matching against the target NP, and selects the arrangement that maximizes chemical similarity. When benchmarked on the type I PKSs annotated within the expert-reviewed ClusterCAD dataset, NPannotator recovered 62.0% of both correct gene orderings and AT substrate annotations, and achieved 80.0% accuracy on gene ordering alone. By bridging gene-level architecture with chemical outcomes, NPannotator represents a step toward systematically decoding how protein sequence and genomic organization encode chemical structure in the world of natural products.

biochemistry↗

bioRxiv2026

Linear-time prediction of proteome-scale microbial protein interactions

Protein-protein interactions (PPIs) underpin biological function, yet proteome-scale interaction prediction remains bottlenecked by the quadratic computational complexity of all-vs-all pairwise comparisons. Here, we present FlashPPI, a contrastive learning framework, grounded in residue-level interactions, that enables linear-time prediction of physical protein interfaces across a microbial proteome. By leveraging a genomic language model that captures cross-protein co-evolutionary signals from metagenomic sequences, FlashPPI aligns interacting partners in a shared latent space. We demonstrate a four-fold performance increase over existing sequence-based methods, while reducing proteome-wide screening time from days to minutes. Crucially, FlashPPI achieves comparable screening performance to state-of-the-art structure-folding models at a fraction of the computational cost. Finally, we integrate FlashPPI into seqhub.org, an interactive web platform that combines predicted networks with functional annotations and genomic context, making proteome-wide network analysis rapid and accessible for microbial discovery.

molecular biology↗

bioRxiv2024

Gaia: A Context-Aware Sequence Search and Discovery Tool for Microbial Proteins

Protein sequence similarity search is fundamental to genomics research, but current methods are typically not able to consider crucial genomic context information that can be indicative of protein function, especially in microbial systems. Here we present Gaia (Genomic AI Annotator), a sequence annotation platform that enables rapid, context-aware protein sequence search across genomic datasets. Gaia leverages gLM2, a mixed-modality genomic language model trained on both amino acid sequences and their genomic neighborhoods to generate embeddings that integrate sequence-structure-context information. This approach allows for the identification of functionally related genes that are found in conserved genomic contexts, which may be missed by traditional sequence- or structure-based search alone. Gaia enables real-time search of a curated database comprising over 85M protein clusters (defined at 90% sequence identity) from 131,744 microbial genomes. We compare the sequence, structure and context sensitivity of gLM2 embedding-based search against existing tools like MMseqs2 and Foldseek. We showcase Gaia-enabled discoveries of phage tail proteins and siderophore synthesis loci that were previously difficult to annotate with traditional tools. Gaia search is freely available at https://gaia.tatta.bio.

bioinformatics↗

bioRxiv2024

The OMG dataset: An Open MetaGenomic corpus for mixed-modality genomic language modeling

Biological language model performance depends heavily on pretraining data quality, diversity, and size. While metagenomic datasets feature enormous biological diversity, their utilization as pretraining data has been limited due to challenges in data accessibility, quality filtering and deduplication. Here, we present the Open MetaGenomic (OMG) corpus, a genomic pretraining dataset totalling 3.1T base pairs and 3.3B protein coding sequences, obtained by combining two largest metagenomic dataset repositories (JGIs IMG and EMBLs MGnify). We first document the composition of the dataset and describe the quality filtering steps taken to remove poor quality data. We make the OMG corpus available as a mixed-modality genomic sequence dataset that represents multi-gene encoding genomic sequences with translated amino acids for protein coding sequences, and nucleic acids for intergenic sequences. We train the first mixed-modality genomic language model (gLM2) that leverages genomic context information to learn robust functional representations, as well as coevolutionary signals in protein-protein interfaces and genomic regulatory syntax. Furthermore, we show that deduplication in embedding space can be used to balance the corpus, demonstrating improved performance on downstream tasks. The OMG dataset is publicly hosted on the Hugging Face Hub at https://huggingface.co/datasets/tattabio/OMG and gLM2 is available at https://huggingface.co/tattabio/gLM2_650M.

systems biology↗

bioRxiv2024

Diverse Genomic Embedding Benchmark for functional evaluation across the tree of life

Biological foundation models hold significant promise for deciphering complex biological functions. However, evaluating their performance on functional tasks remains challenging due to the lack of standardized benchmarks encompassing diverse sequences and functions. Existing functional annotations are often scarce, biased, and susceptible to train-test leakage, hindering robust evaluation. Furthermore, biological functions manifest at multiple scales, from individual residues to large genomic segments. To address these limitations, we introduce the Diverse Genomic Embedding Benchmark (DGEB), inspired by natural language embedding benchmarks. DGEB comprises six embedding tasks across 18 expert curated datasets, spanning sequences from all domains of life and encompassing both nucleic acid and amino acid modalities. Notably, four datasets enable direct comparison between models trained on different modalities. Benchmarking protein and genomic language models (pLMs and gLMs) on DGEB reveals performance saturation with model scaling on numerous tasks, especially on those with underrepresented sequences (e.g. Archaea). This highlights the limitations of existing modeling objectives and training data distributions for capturing diverse biological functions. DGEB is available as an open-source package with a public leaderboard at https://github.com/TattaBio/DGEB.

bioinformatics↗