A new study published in Nature Methods by Jacob Schreiber, of the Research Institute of Molecular Pathology at the Vienna BioCenter and UMass Chan Medical School, aims to close that gap. The paper introduces tangermeme, a free, open-source Python package described as an everything-but-the-model toolkit for genomic deep learning. The name of the design philosophy is deliberate. Rather than competing with the many frameworks that build, train and host neural networks, tangermeme concentrates exclusively on what happens after a model has been trained: making predictions efficiently, perturbing sequences, attributing importance to individual nucleotides, calling regulatory patterns and designing new DNA. The result, according to the paper, is a comprehensive, flexible and efficient Swiss Army knife for cis-regulatory pattern identification and analysis.

The rationale for this division of labor is grounded in how the field has actually evolved. Model architectures and training strategies churn constantly, with convolutions giving way to transformers and new optimizers arriving every year. Downstream analyses, by contrast, are remarkably stable and largely agnostic to architecture. Estimating the effect of a noncoding variant on a model’s prediction works essentially the same way whether the underlying network uses convolutional filters or self-attention, and whether it was trained with one optimizer or another. This modularity means that a well-engineered library of analysis operations can serve any model, old or new. Yet until now, no optimized repository of such methods existed that generalized across model types. Instead, researchers typically ship bespoke analysis code bundled with each released model, and existing packages either implement only a few related algorithms or focus on training and fine-tuning with limited post-training support.

Tangermeme’s central technical trick is the strict separation of sequence manipulations from model operations, which can then be freely stacked to create analyses. In silico marginalization compares predictions before and after substituting a short sequence into a template, quickly identifying which motifs from a database the model responds to. Ablation is the conceptual opposite, altering a region the model may be using and measuring the drop in prediction. Variant effect estimation compares predictions before and after one or a few noncontiguous substitutions, allowing fine-mapping of candidate regulatory variants. Crucially, the operation performed after a sequence manipulation need not be a simple prediction; it can be DeepLIFT/SHAP attribution, in silico saturation mutagenesis, or any custom operation a researcher writes. This composability lets scientists ask questions such as how surrounding sequence context influences transcription factor binding, whether a regulatory element is suitable for a synthetic construct, or how two variants at the same locus interact, all in a handful of commands.

The package also embraces DNA design, implementing several methods catalogued in a recent review of enhancer modeling. The simplest is screening: generate random sequences, score them against a user-defined objective, and keep the best. Greedy substitution, sometimes called directed evolution, iteratively improves a sequence by testing every possible single-nucleotide change and keeping the best one at each step. Motif implantation works similarly but swaps in entire motifs from a database rather than single bases. Most novel is a construct marginalization method that merges design with marginalization, creating short DNA inserts that reliably shift model predictions, such as patterns of predicted chromatin accessibility, averaged across many background sequences, without requiring prior knowledge of what the construct should contain. Such libraries of designed inserts could prove valuable for synthetic biology and therapeutic applications.

Engineering choices under the hood are where tangermeme distinguishes itself from prior efforts. Operations come with built-in batching, support every data type and device available in PyTorch, and work out of the box on models with multiple inputs and outputs, arbitrary internal units including convolutions, long short-term memory blocks and transformers, and any output format from a single binary value to a full base-pair-resolution profile. Speed benchmarks illustrate the payoff: tangermeme one-hot encodes the entirety of human chromosome 1 in under two seconds, roughly three times faster than the next fastest implementation tested. That may sound trivial, but fast encoding enables on-the-fly batch generation during analysis, and the same optimization philosophy runs throughout the codebase. Batching also solves a practical bottleneck in DeepLIFT/SHAP, which in many implementations requires all background sequences to fit in GPU memory simultaneously, a serious constraint for massive models.

Perhaps the most consequential finding to emerge from building the toolkit carefully is a subtle failure mode in widely used attribution code. DeepLIFT/SHAP works by overriding the backward pass of nonlinear operations to incorporate a reference sequence, and identifying which operations are nonlinear requires checking each computational layer against an internal dictionary. Existing implementations can silently fail when they encounter a nonlinear operation outside that dictionary or when an activation is reused, often without documentation or warnings. The telltale sign is convergence deltas, which should be zero in theory and near machine precision in practice, ballooning to indicate that the attributions are simply wrong. Tangermeme never fails silently: it raises warnings when convergence deltas are too high, can return those deltas for monitoring, and allows users to register custom nonlinear functions to avoid the problem entirely.

Beyond wrapping established algorithms, tangermeme introduces genuinely new methods for distilling learned cis-regulatory logic, centered on objects called seqlets, the contiguous spans of high-attribution nucleotides that attribution methods such as DeepLIFT/SHAP, in silico saturation mutagenesis, PISA or saliency produce. The package’s novel recursive seqlet caller is, according to the paper, the first to call variable-length seqlets directly, using a principled statistical definition: a seqlet is a span whose attribution sum is statistically significant against a null distribution, with the requirement that all subspans within it also pass the test. The algorithm bins attribution values, builds null distributions for each candidate length, converts them to P values, and decodes the longest non-overlapping significant spans. Compared against a reimplementation of TF-MoDISco’s caller on synthetic sequences with an inserted MYC motif, the recursive approach produced shorter, more precise calls and ran faster on modest-to-large numbers of sequences, while TF-MoDISco’s wider, more permissive calls reflect design choices made for its original role inside a larger pipeline.

The payoff of this machinery is demonstrated on real models. Applying the automatic seqlet-calling and annotation pipeline, which maps called seqlets to motifs from the JASPAR database and then counts motif occurrences, pairwise co-occurrences and spacing relationships, Schreiber compared two models ostensibly trained to predict the same thing: MYC binding. At the PLD6 promoter, attribution analysis revealed that BPNet’s predictions are driven almost exclusively by a MYC motif, whereas the much larger Beluga model relies on a broader lexicon of motifs also associated with chromatin accessibility and transcription initiation. Counting annotated seqlets across all MYC peaks and their pairwise occurrences made these differences in learned regulatory logic immediately visible, despite both models achieving similar predictive goals.

Tangermeme is released under the MIT license, installable with a single pip command and accompanied by extensive documentation and tutorials that reproduce the paper’s figures. Although the demonstrations focus on transcription factor binding and chromatin accessibility, the authors emphasize that the toolkit applies to models of any genomic modality, including alternative splicing, transcription, enhancer activity, RNA stability and RNA binding. Schreiber also points toward the future: as large language model-based coding agents become common in research, well-documented packages with tested, composable building blocks will let such systems reuse reliable code rather than reinventing analyses, reducing the burden of auditing generated code. Future development will focus on expanding functionality and deepening documentation for exactly that purpose, positioning tangermeme as infrastructure for an era in which interpreting the genome’s regulatory code may depend as much on trustworthy software as on the models themselves.

Subject of Research: A deep learning interpretation toolkit for cis-regulatory genomics

Article Title: Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models

Article References: Schreiber, J. (2026). Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models. Nature Methods. https://doi.org/10.1038/s41592-026-03254-z

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03254-z

Keywords: deep learning, genomics, cis-regulatory logic, software toolkit, feature attribution, seqlets, transcription factor binding, variant effect prediction, DNA design, interpretability, Nature Methods, open source

The motivation behind the work lies in what geneticists call the missing heritability problem. Genome-wide association studies have catalogued thousands of common genetic variants linked to common diseases, yet together these variants explain only a portion of the heritability estimated from family studies. Rare variants, defined here as those carried by fewer than one percent of the population, have long been suspected of hiding much of the missing signal. Population-scale sequencing efforts, including the exome sequencing of nearly half a million UK Biobank participants, have uncovered millions of these rare missense variants. The difficulty is that the overwhelming majority of them are evolutionarily young, present in only one or a few individuals, and statistically almost impossible to test for association one at a time. Distinguishing the functional few from the neutral many has remained one of the field’s central challenges.

RovHer attacks the problem from an unusual angle. Instead of asking whether a variant is deleterious to a protein in the abstract, the method asks whether a variant contributes to the variance of a measurable trait in a real population. The researchers trained their model using exome-wide association statistics covering 4,927,334 rare variants, treating the statistical evidence of association as a noisy label of functionality. On top of that foundation, they used a machine-learning technique known as multivariate adaptive regression splines, or MARS, to integrate 75 different features describing each variant and the gene it sits in. These features ranged from conservation measures and protein-domain information to gene-level annotations such as tolerance to loss-of-function mutations. Crucially, the MARS framework imposes no a priori constraints on how these features should be combined, allowing the data to reveal nonlinear interactions, for example cases where a variant’s importance depends on the properties of the gene it inhabits.

The benchmark results are dramatic. Across 21 quantitative traits measured in up to 357,086 individuals of European ancestry in the UK Biobank, the top one percent of variants prioritized by RovHer, a set of just 13,410 missense variants, explained on average 16.1 percent of the total heritability attributable to rare missense variation. Seven widely used alternative methods, including established pathogenicity predictors such as CADD, REVEL and AlphaMissense, managed an average of only 1.5 percent, with the best performer reaching 2.3 percent. In other words, RovHer achieved roughly a tenfold gain over the existing state of the art. Because the evaluation metric was heritability explained rather than agreement with laboratory assays or clinical classifications, the comparison directly measured what matters most for complex-trait genetics: how much inherited variation a prioritized variant set actually captures.

The authors were careful to rule out trivial explanations for this performance. RovHer’s predictions proved independent of allele frequency, meaning the method is not simply rewarding variants that happen to be slightly more common and therefore easier to detect statistically. The scores also generalized beyond the population in which they were trained. In a replication analysis using whole-genome sequencing data from the National Institutes of Health’s All of Us Research Program, RovHer outperformed competing methods across European, African and Admixed American ancestry groups, with sample sizes of 172,108, 61,620 and 52,645 respectively. This multi-ancestry validation matters because rare variant landscapes differ substantially between populations, and tools trained exclusively on European data have historically underperformed elsewhere.

Beyond population-level statistics, the team tested whether RovHer could identify individual carriers at elevated risk of severe clinical outcomes. Across 17 monogenic gene-trait pairs, where a single gene is known to exert a strong effect on a specific condition, high-scoring variants effectively flagged carriers with increased disease risk. Examples included variants in LDLR, the low-density lipoprotein receptor gene whose pathogenic mutations cause familial hypercholesterolemia, and variants in SLC34A3, a sodium-dependent phosphate transporter gene linked to kidney stone disease and altered estimated glomerular filtration rates. The analysis showed that carriers of top-ranked missense variants in these genes had measurably worse kidney function and higher chronic kidney disease risk than non-carriers, suggesting the method can bridge the gap between statistical genetics and clinical risk stratification.

Technically, the choice of MARS proved central to the method’s success. Unlike deep neural networks, which can be difficult to interpret, multivariate adaptive regression splines fit piecewise linear functions with hinge points, producing a model whose behavior can be inspected through partial dependence plots and variable importance measures. This allowed the researchers to see which of the 75 input features drove predictions and where thresholds lay. The team also demonstrated robustness to training choices: models trained on height, body mass index, creatinine or apolipoprotein B each performed well when tested on other traits, and adding more training traits provided incremental gains rather than overfitting. This suggests the method captures general principles of variant functionality rather than quirks of any single phenotype.

Perhaps the most practical aspect of the study is its accessibility. The researchers have released RovHer scores for all 4,927,334 rare variants observed in the UK Biobank, along with precomputed scores for approximately 80 million potential single-nucleotide rare variants across the human exome, hosted on Zenodo. The software package runs offline on all major operating systems, requires only a simple tab-delimited file of variant identifiers as input, and is free for noncommercial use, with a version stripped of commercially licensed prediction methods also provided. Tutorials and accompanying code for rare variant heritability estimation are available on GitHub. For biobank-scale analyses the authors recommend Unix-based hardware with at least 300 gigabytes of memory, but individual variant scoring is far lighter.

The broader significance of the work lies in its reframing of rare variation’s role in common disease. Rather than treating rare missense variants as an undifferentiated mass too sparse to analyze, RovHer shows that a small, computationally identifiable subset carries a disproportionate share of trait-relevant signal. This opens the door to incorporating rare variant information into polygenic risk scores, which currently rely almost entirely on common variants, and to prioritizing which variants of uncertain significance in clinical exomes deserve the closest scrutiny. It also complements recent work showing that rare penetrant mutations confer severe risks for common diseases, and that the genetic architecture of complex traits differs in the tails of their distributions.

Limitations remain, as with any predictive model. RovHer was trained on European-ancestry UK Biobank data, and while its multi-ancestry replication is encouraging, further calibration across diverse populations will be needed as adoption spreads. The method also inherits the limitations of its input features, since variants in poorly characterized genes or genomic regions with sparse annotations may be scored less reliably. Nonetheless, the tenfold improvement in heritability capture, the clinical validation across monogenic gene-trait pairs and the open availability of scores for essentially every possible exomic variant make RovHer a landmark addition to the geneticist’s toolkit. As sequencing continues to expand across global populations, heritability-optimized approaches of this kind are likely to become standard instruments for separating the functional signal from the neutral noise in the rare variant universe.

Subject of Research: Heritability-optimized functional prioritization of rare coding variants in complex traits

Article Title: A heritability-optimized method for functional prioritization of rare coding variants in complex traits

Article References: Pang, K., Pathan, N., Le, A., Man, A., Graf, J., Li, Y., da Rocha, G. L., Shemesh, E., Grafodatskaya, D., Chong, M. R., & Paré, G. (2026). A heritability-optimized method for functional prioritization of rare coding variants in complex traits. Nature Genetics. https://doi.org/10.1038/s41588-026-02766-w

Image Credits: AI Generated

DOI: 10.1038/s41588-026-02766-w

Keywords: rare variants, missense variants, heritability, complex traits, UK Biobank, RovHer, variant effect prediction, machine learning, genetic risk, population genetics, monogenic disease, exome sequencing

The analogy is more than a metaphor. DNA is written in an alphabet of four nucleotide letters, proteins in an alphabet of twenty amino acids, and RNA in a four-letter code of its own. Just as words derive meaning from their context in a sentence, the function of a nucleotide or amino acid depends on its position within a sequence and on the sequences that surround it. The transformer architecture, introduced in 2017 in the landmark paper Attention Is All You Need, excels precisely at capturing such long-range contextual dependencies. By applying self-attention mechanisms across millions of biological sequences, models can learn statistical patterns that reflect evolutionary constraints, structural motifs, and functional sites without any explicit programming.

The review organizes the field into biological foundation models, general-purpose systems pretrained on massive unlabeled sequence databases, and specialized models tailored to particular problems. On the protein side, models such as ProtTrans and the Evolutionary Scale Modeling family demonstrated that unsupervised learning on hundreds of millions of sequences yields internal representations from which structure and function can be read off. This line of work culminated in AlphaFold and its successors, which achieved atomic-level accuracy in protein structure prediction and, with AlphaFold 3, extended to predicting interactions between proteins, nucleic acids, and small molecules. Generative protein LLMs such as ProGen and ProGen2 have gone further, producing entirely novel amino acid sequences that fold into functional proteins, a capability the review highlights as one of the most consequential advances of the past decade.

DNA language models have followed a parallel trajectory. The Nucleotide Transformer, built and evaluated on human and multi-species genomes, showed that robust foundation models could be trained at nucleotide resolution. Subsequent work demonstrated that such models are powerful predictors of genome-wide variant effects, meaning they can estimate whether a mutation in a patient’s genome is likely to be harmful, a task central to clinical genetics. The review also points to Evo and its successor Evo 2, genome-scale models capable of modeling and designing sequences across all domains of life, from bacteria to humans. These systems operate at a scale that was unthinkable for genomics only a few years ago, processing contexts long enough to encompass entire genes and their regulatory neighborhoods.

RNA has proven a particularly fertile ground because its structure and function are notoriously difficult to predict from sequence alone. Classical approaches relied on thermodynamic folding algorithms, but deep learning models such as UFold and RNAformer have shown that learned representations can match or exceed physics-based methods for secondary structure prediction. More recently, language model-based deep learning approaches have achieved accurate RNA 3D structure prediction, and motif-aware pretraining strategies have produced multi-purpose RNA models that handle splicing prediction, binding site identification, and function annotation within a single framework. The review notes that self-supervised learning on millions of primary RNA sequences from dozens of vertebrate species has substantially improved sequence-based splicing prediction, with direct implications for interpreting disease-causing mutations in untranslated regions and splice sites.

Beyond the three canonical sequence types, the review surveys a rapidly growing ecosystem of specialized applications. Interaction prediction models forecast which proteins bind which RNAs, which RNAs pair with other RNAs, and how mutations perturb these contacts. Structure prediction systems now cover proteins, RNA, and biomolecular complexes at near-experimental accuracy. Perhaps most striking is the single-cell revolution: foundation models such as scGPT, scBERT, and Geneformer treat gene expression profiles across thousands of individual cells as a kind of language, learning universal representations that support cell type annotation, perturbation response prediction, and the integration of massive transcriptomic datasets. Studies assessing GPT-4 for cell type annotation suggest that general-purpose chatbots can also contribute, though specialized models currently hold the edge on most benchmarks.

The authors are candid about the obstacles. Hallucination, the tendency of generative models to produce plausible but false outputs, is a documented failure mode of natural language generation that carries serious risks when outputs inform drug design or clinical interpretation. Benchmarking remains immature: evaluation suites such as DART-Eval for regulatory DNA and PertEval-scFM for perturbation prediction reveal that performance gains do not always transfer across datasets and cell types. Computational cost is another barrier, since genome-scale attention models demand enormous memory, motivating architectural innovations such as the Hyena hierarchy and Mamba-based state space models that the review identifies as promising directions for long-context biological modeling. Data quality, batch effects in single-cell experiments, and the interpretability of learned representations round out the open challenges.

What distinguishes this review from many surveys of artificial intelligence in biology is its educational dimension. Because it appears in a journal focused on digital education, the authors devote substantial attention to integrating LLMs into genomics teaching and learning. They propose practical projects in which students fine-tune pretrained models on small datasets, use conversational agents to simplify bioinformatics workflows, and learn to critically evaluate model outputs. Tools such as BioMANIA, which allows researchers to interrogate bioinformatics pipelines through natural language conversation, and BioCoder, a benchmark for bioinformatics code generation, illustrate how LLMs can lower the technical barriers that traditionally separate biologists from computational analysis. The authors argue that fluency in these tools is becoming as essential to the modern life scientist as statistical literacy.

The trajectory the review describes is remarkably compressed. In roughly four years, language models have progressed from proof-of-concept embeddings to systems that simulate 500 million years of evolutionary history, design functional proteins, predict the pathogenicity of genetic variants, and annotate cell atlases containing millions of cells. The Wuhan University team frames the current moment as an inflection point: the same scaling laws and reasoning enhancements that drive progress in general-purpose AI, including reinforcement learning approaches that sharpen chain-of-thought reasoning, are poised to accelerate biological discovery further. If the challenges of reliability, evaluation, and interpretability can be met, the authors suggest, the fusion of language models and genomics may redefine both how biology is researched and how the next generation of scientists is trained, turning the genome from an inscrutable code into a text that machines can read, annotate, and even rewrite.

Subject of Research: Applications of large language models in genomics, including biological foundation models for protein, DNA, and RNA sequence analysis

Article Title: AI-Empowered Genome Decoding: Applications of Large Language Models in Genomics

Article References: Li, S., Fan, W., & Zhou, Y. (2025). AI-Empowered Genome Decoding: Applications of Large Language Models in Genomics. Frontiers of Digital Education, 2(1), Article 14. https://doi.org/10.1007/s44366-025-0051-1

Image Credits: AI Generated

DOI: 10.1007/s44366-025-0051-1

Keywords: large language models, genomics, deep learning, protein structure prediction, DNA language models, RNA secondary structure, single-cell analysis, foundation models, AlphaFold, transformer architecture, bioinformatics education, variant effect prediction

Developed by Polina Polunina and Wolfgang Maier of the University of Freiburg together with Alan Rubin of the Walter and Eliza Hall Institute of Medical Research, VEFill is built on LightGBM, a gradient boosting framework well suited to structured, high-dimensional tabular data. Rather than relying on a single source of information, the model integrates a rich set of biologically informed features: evolutionary conservation scores from the EVE generative model, amino acid substitution matrices such as BLOSUM62 and PAM250, physicochemical descriptors of the wild-type and variant residues, and contextual sequence embeddings from ESM-1v, a transformer-based protein language model trained on millions of natural sequences. The authors trained and optimized the model using Bayesian hyperparameter search and stored all underlying data in a structured PostgreSQL database to ensure reproducibility.

The training ground for VEFill was the Human Domainome 1 dataset, a standardized collection of stability measurements generated with an abundance-based protein fragment complementation assay across 522 human protein domains, using site-saturation mutagenesis to introduce every possible amino acid substitution. From this resource, the team assembled a feature-complete subset of 140 domains comprising 136,854 mutations for the full model, and used the broader set of 521 domains, totaling 562,208 mutations, to train reduced-feature versions that depend only on ESM-1v embeddings and positional mean scores.

The performance figures are striking. In a leave-protein-out evaluation, where the model must predict variant effects for protein domains entirely absent from training, the best configuration achieved a coefficient of determination of 0.64 and a Pearson correlation of 0.80. Feature ablation experiments revealed a clear hierarchy of information: models relying solely on evolutionary scores, substitution matrices, or biochemical descriptors performed poorly, with R-squared values below 0.15. Adding ESM-1v embeddings lifted performance substantially, and combining the embeddings with the mean experimental score per position proved even more powerful, indicating that learned sequence representations and data-driven positional priors capture complementary aspects of mutational tolerance.

Perhaps the most practically important finding concerns data scarcity. In controlled subsampling experiments on 28 high-quality datasets, VEFill consistently outperformed a battery of existing imputation methods, including Envision, FactorizeDMS, AALasso, and nearest-neighbor approaches, once at least 20 percent of variants had been experimentally measured. Below that threshold, predictions became markedly harder, and the authors are candid that true zero-shot prediction without any positional context remains challenging, particularly for functionally complex proteins. This matters because a survey of 841 public score sets from the MaveDB repository showed that fully complete datasets are vanishingly rare, with only 44 out of 841 achieving complete coverage of all possible substitutions. The vast majority of real-world experiments therefore fall squarely within the regime where VEFill delivers its greatest benefit.

The team also probed how few measurements are actually needed per position. When per-protein models were trained with a restricted number of substitutions per site, accuracy rose rapidly and began to plateau at roughly four mutations per position. Intriguingly, models trained exclusively on five carefully chosen amino acids, histidine, glutamic acid, asparagine, isoleucine, and glycine, each representing a distinct physicochemical class, matched the performance of models trained on randomly selected substitutions at equivalent coverage. This suggests that sparse, information-efficient mutational libraries could be designed deliberately, prioritizing a small representative set of substitutions per site without sacrificing predictive power.

Noise ceiling analyses added an important dose of realism. Because experimental variability imposes a fundamental upper bound on achievable accuracy, the researchers simulated replicate experiments using reported measurement uncertainties and found that estimated ceilings ranged from 0.68 to 0.99 across domains. VEFill’s per-protein correlations, between 0.52 and 0.91 under an 80/20 split, sat consistently below but generally close to these limits, indicating that much of the remaining discrepancy between predicted and observed scores reflects measurement noise rather than model failure. Error profiles were lowest for near-neutral variants, where data are densest, and predictions at the extreme deleterious and high-activity tails should be read as conservative approximations rather than precise estimates.

Generalization beyond the training data showed both promise and limits. When the cross-protein model was applied to eight full-length proteins from independent MaveDB datasets, including Parkin, PTEN, aspartoacylase, calmodulin, TPK1, and TP53, it performed substantially better on stability-based assays than on activity-based ones, a mismatch the authors attribute to differences between training and test assay modalities and the complexity of cellular phenotypes. A lightweight two-feature version using only ESM-1v embeddings and positional mean scores performed comparably to the full model in most settings, offering a practical alternative when evolutionary scores or other annotations are unavailable. Analysis of protein family composition revealed that a few Pfam families dominated the training data, but retraining on a Pfam-unique subset increased error only modestly, suggesting the findings are robust rather than inflated by family-level redundancy.

Fine-grained evaluations exposed the model’s blind spots in biologically meaningful ways. Proline substitutions consistently produced elevated errors, likely because their unusual conformational constraints disrupt secondary structure in ways that sequence-based features do not fully capture. In the TRIM44 zinc finger domain, histidine-to-cysteine mutations were poorly predicted when entire positions were held out but accurately recovered when single variants were withheld, hinting at context-dependent chemistry, such as zinc coordination, that only fine-grained positional information can resolve. The authors suggest that future versions could incorporate explicit structural features, cross-species transfer learning, or systematic assay harmonization to close these gaps.

Overall, VEFill arrives at a moment when the variant effect field is scaling rapidly, with MaveDB now listing more than 2,600 public datasets and over 1,100 for human proteins. By providing an interpretable, scalable tool that turns partially complete mutational maps into denser, more usable resources, the study lowers the experimental barrier for variant prioritization, protein engineering, and the clinical interpretation of missense mutations. The code and trained models have been released openly, inviting the community to apply sparse-library design strategies and, ultimately, to build the comprehensive Atlas of Variant Effects that the field has long envisioned.

Subject of Research: Machine learning-based imputation of missing scores in deep mutational scanning datasets across human protein domains

Article Title: VEFill: accurate and generalizable deep mutational scanning score imputation across protein domains

Article References: Polunina, P. V., Maier, W., & Rubin, A. F. (2026). VEFill: accurate and generalizable deep mutational scanning score imputation across protein domains. Molecular Systems Biology, 22(6), 979-1002. https://doi.org/10.1038/s44320-026-00203-y

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00203-y

Keywords: deep mutational scanning, variant effect prediction, machine learning, gradient boosting, protein stability, ESM-1v, protein language models, MaveDB, Human Domainome, imputation, missense variants, bioinformatics

Why does this matter? Proteins almost never work alone. They form intricate interaction networks that govern everything from immune responses to reproduction, growth, and product quality. Understanding these networks at a systems level is essential for improving traits such as disease resistance, fertility, and meat or egg quality—the cornerstones of food security and sustainable farming. Yet existing interaction databases like STRING, BioGRID, IntAct, and MINT offer only sparse coverage of agricultural species, and they typically lack the structural annotations needed to identify binding interfaces, functional hotspots, or the molecular consequences of mutations. Without such detail, linking a genetic variant to a measurable trait—an essential step in molecular breeding—remains a matter of guesswork.

The new database was built through an ambitious integrative pipeline that combines three complementary prediction strategies. The first, interolog mapping, exploits evolutionary conservation: if two proteins in a model organism are known to interact, their homologs in a farm animal are likely to interact as well. The team assembled a rigorously filtered library of experimentally supported interactions from seven model organisms, including human, mouse, fly, worm, yeast, E. coli, and Arabidopsis, retaining only high-confidence interactions scored by the HIPPIE scheme. The second strategy, domain–domain interaction inference, predicts interactions based on conserved protein domain pairings drawn from the 3did catalog and expanded with an expectation–maximization algorithm applied to curated interaction data.

The third and most modern component is a deep learning model. The researchers retrained PPITrans, a transformer-based architecture built on the ProtT5 protein language model, using a curated set of 44,035 high-confidence human interactions as positive samples and 430,370 negative pairs. Training on an NVIDIA A30 GPU with the AdamW optimizer, the model learned sequence-derived interaction patterns that generalize across species. The scores from all three methods were then fused through a logistic regression model trained on a mouse benchmark dataset, producing a single interaction probability for every protein pair. On independent testing, the integrated model achieved an area under the ROC curve of 0.914 and an area under the precision–recall curve of 0.679, outperforming each individual component. Interactions were classified as high-confidence only above a stringent threshold calibrated at a 0.1 percent false positive rate.

The resulting species-specific networks are striking in scale. Cattle lead the pack with 173,331 predicted interactions covering 34.0 percent of the proteome, followed by pig with 90,987 interactions, chicken with 61,593, duck with 43,886, and sheep with 40,624. In total, these interactomes span proteomes at coverage rates ranging from 19.5 to 34.0 percent—a dramatic expansion of molecular data for organisms that have historically been interaction-data deserts. Cross-validation against independent resources lent credibility to the predictions: the cattle network shared 8,127 interactions with a D-SCRIPT-based dataset and captured 207 of 632 experimentally solved cattle complex structures in Interactome3D, both overlaps highly significant by Fisher’s exact test.

What truly distinguishes AASIA, however, is its structural dimension. The team used ESMFold, a cutting-edge structure prediction system powered by the ESM-2 protein language model, to generate three-dimensional models for every protein monomer across the five proteomes and, in multimer mode, for all 410,000-plus interacting pairs. Unlike earlier approaches requiring multiple sequence alignments or structural templates, ESMFold predicts structures directly from amino acid sequences, enabling proteome-scale modeling at unprecedented speed. Most monomer models showed high confidence, with pLDDT scores predominantly above 70. Each complex was then annotated with interface residues defined by geometric contact and hotspot residues identified through computational alanine scanning with the FoldX force field, which flags positions whose mutation would destabilize the interface by at least 0.5 kilocalories per mole.

To demonstrate the database’s power, the researchers dove deep into the chicken interactome. Its network topology follows classic small-world, scale-free architecture, with an average of 24 interactions per protein, a mean shortest path length of 3.3, and a clustering coefficient of 0.63—far higher than randomized networks of the same size. Fifty predicted chicken interactions overlapped significantly with experimentally supported ones, and predicted pairs showed Gene Ontology semantic similarity comparable to known interactions and far above random pairs. A flagship case study examined the actin-capping protein assembly: the predicted CAPZA1–CAPZB heterodimer aligned with the experimentally solved crystal structure with a root-mean-square deviation of just 1.6 angstroms, and the predicted binding interfaces for both subunits with ACTIN matched mutagenesis-verified actin-binding motifs at the C-termini of the CapZ subunits.

Perhaps the most compelling illustration of AASIA’s practical value involves a predicted interaction between XRCC5, a DNA repair factor, and H2AC39, a histone variant, in chicken. The interaction was independently supported by all three prediction methods and scored 0.94 overall. Remarkably, the predicted interface overlaps a known quantitative trait locus associated with eggshell strength. A residue at that locus, arginine 10 of XRCC5, sits directly at the interaction interface and was flagged as a putative hotspot: FoldX predicted that mutating it to alanine would destabilize the complex by 1.09 kilocalories per mole, and the database’s variant effect module predicted a corresponding drop in the interaction score. This offers a testable, structure-based hypothesis that genetic variation at this position could alter chromatin dynamics during eggshell mineralization—precisely the kind of genotype-to-phenotype bridge molecular breeders have long sought.

The platform itself is designed for broad usability, built on a Vue.js frontend, Express.js backend, and MySQL metadata management, with Cytoscape.js for network visualization and Mol* and NGL Viewer for interactive 3D structure exploration. Users can search by UniProt, Ensembl, or Entrez identifiers, gene names, or keywords; browse protein and interaction pages with functional annotations from Gene Ontology and KEGG; and, crucially, submit their own protein sequences to the Predict module for de novo interaction inference or quantitative assessment of mutation effects on interaction scores. This last capability turns AASIA from a static repository into an active hypothesis-generating engine.

The authors acknowledge limitations: public databases mix direct binary interactions with association-level evidence, and ESMFold does not explicitly model intrinsically disordered regions that mediate many interactions. Still, performance on a stricter binary-only test subset was even stronger, with an AUROC of 0.928. Looking forward, the team plans to expand AASIA with host–pathogen interaction data—potentially illuminating conserved interfaces exploited by avian influenza and African swine fever viruses—and to integrate transcriptomic, epigenomic, and genome-wide variant data. By transforming raw omics information into spatially resolved, residue-level insight, AASIA positions itself as a foundational tool for the next generation of livestock genomics and biotech breeding.

Subject of Research: A structure-informed protein–protein interactome database for five agricultural animal species

Article Title: AASIA: A comprehensive protein structural interactome database for agricultural animals

Article References: Li, J., Yuan, M., Jiang, L., Li, D., Yin, Z., Zhu, F., Shi, W., Hou, Z., & Zhang, Z. (2026). AASIA: A comprehensive protein structural interactome database for agricultural animals. Journal of Advanced Research. https://doi.org/10.1016/j.jare.2026.09.015

Image Credits: AI Generated

DOI: 10.1016/j.jare.2026.09.015

Keywords: AASIA, protein-protein interactions, agricultural animals, structural biology, ESMFold, deep learning, interolog mapping, molecular breeding, livestock genomics, variant effect prediction, disease resistance, food security