Microbiome functions are encoded within the genes of the community-wide metagenome. A natural question is whether properties of a microbial community can be predicted just from knowing the raw DNA sequences of its members. In this work, we employ set-aggregated genome embeddings (SAGE) to predict community-level abundance profiles, exploiting the few-shot learning capabilities of genomic language models (GLMs). We benchmark this approach to show improved generalization on novel genomes compared to classical bioinformatics approaches. Model ablation shows that community-level latent representations directly result in improved performance. Lastly, we demonstrate the benefits of intermediate transformations between latent representations and demonstrate the differences between GLM embedding choices.
Tabular foundation models (TFMs) achieve strong performance on microbiome abundance data, yet their robustness under realistic distribution shift remains poorly characterized. We introduce a benchmark that evaluates the robustness of TFMs to biologically inspired perturbations across six gut microbiome datasets spanning four disease contexts. In this in-context learning setting, models receive unperturbed support sets as context and are evaluated on perturbed query samples. To isolate robustness beyond "shortcut" features, we preserve the most discriminative taxa and apply three controlled perturbation strategies: (i) removal of high-abundance (uninformative) taxa, (ii) sparsification via increased zero-inflation, and (iii) zero-imputation via spurious non-zero injections. Our results show that protecting discriminative features is insufficient to guarantee stability under support-query shift: across datasets, all perturbations degrade model performance, with zero-imputation consistently the most harmful, indicating that corrupting global feature structure can break generalization even when key taxa are retained. Sparsification disproportionately affects TFMs relative to a classical random forest baseline, suggesting greater sensitivity to zero-inflation-type shifts. The code is publicly available at: https://github.com/UMMISCO/metagenomics-fm/.
Giulia Perciballi, Ahmad Fall, Federica Granese +2
Genomic foundation models such as Evo 2 learn rich sequence representations, but their value for biosecurity screening is largely unexplored. We ask how much biosecurity-relevant signal is linearly accessible in these representations by training minimal linear and attention probes on frozen Evo 2 layer-26 activations, without fine-tuning the underlying model. Across held-out metagenomic test sets, the probes detect antimicrobial resistance (AMR) with strong discrimination: a linear probe reaches a region-level ROC-AUC of 0.888 (mean-pool), rising to 0.977 with a single-head attention probe. The probes resolve finer-grained AMR drug-class subcategories and separate them from unrelated functional genes, providing additional evidence that the learned signal is not explained solely by generic functional-gene status. Bacterial virulence is also decodable, though more weakly (region-level ROC-AUC 0.833). The AMR probe retains comparable ranking performance on simulated short reads without retraining, enabling evaluation before assembly in settings where assembly is computationally costly or unreliable. It achieves a read-level ROC-AUC of 0.898 (mean-pool), comparable to the mean-pooled full-region result. Within SynGenome, AMR-associated prompt labels are only weakly recoverable from Evo 1.5-generated sequences; these prompt-derived labels do not establish the function of the generated response sequences. A complementary sparse-autoencoder analysis recovers interpretable resistance-associated features but proves less consistent than the supervised probes. Together, these results position lightweight embedding-based probes as a fast, inexpensive first-pass detection layer for metagenomic biosurveillance and map both strengths and current limits of the approach. This work was conducted as part of the AIxBio Hackathon 2026 hosted by BlueDot Impact, Apart Research, and Cambridge Biosecurity Hub.
Identifying operons is a fundamental step in understanding prokaryotic gene regulation, as classifying genes into operons supports the reconstruction of regulatory networks, functional annotation of unannotated genes, and drug candidate development. Experimental approaches such as RT-PCR and RNA-seq provide precise evidence of operon structure, but are laborious and largely limited to well-studied model organisms, making scalable computational methods essential for genome-wide operon identification. Prior computational approaches have employed traditional classifiers such as logistic regression and decision trees, motivating our use of these as physicochemical baselines. The DGEB benchmark evaluates operonic pair classification by embedding each sequence independently with a pre-trained protein language model and computing pairwise cosine similarity. In contrast, our Siamese MLP learns a classifier over the fused embedding space, which is theoretically better motivated for binary classification, as cosine similarity can yield meaningless scores depending on the regularization of the embedding model. While protein language model embeddings substantially outperform physicochemical features in ROC-AUC, a learned Siamese MLP head does not significantly improve over unsupervised cosine similarity in Average Precision, suggesting that the geometry of the embedding space already captures the functional relationships needed for this task. Nonetheless, our Siamese MLP achieves a ROC-AUC of 0.71, competitive with state-of-the-art models on the DGEB leaderboard. These findings indicate that protein language model embeddings are a viable, scalable foundation for operonic pair classification across diverse microbial genomes, with implications for automated genome annotation, regulatory network reconstruction, and characterization of organisms lacking experimental operon annotations.