sequence-to-function modeling

Human-specific CCG-repeat insertion in the RBFOX1 promoter implicated in autism and speech enhances EGR1-responsive activity

A human-specific CCG insertion in the RBFOX1 promoter boosts EGR1-driven reporter activity in HEK293T cells, as does an autism-linked CCG-gaining duplication in PTCHD1.

A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics

A sparse mixture-of-experts DNA model conditioned on hyperbolic taxonomy embeddings learns better microbial representations than its taxonomy-free twin and comes close to Evo2-7B in probing and fine-tuning at lower compute.

Accessing Enzyme Kinetic Data and Prediction Methods at Scale

OpenKinetics Predictor unifies thirteen enzyme kinetic predictors behind one interface and API; on E. coli iML1515, eight kcat predictors correlate pairwise at r 0.08 to 0.67.

Coevolutionary mining of prokaryotic non-coding elements with a genome language model

Minerva extracts RNA base-pairing, protein-contact and repeat maps from a genome language model without alignments, flagging unannotated structured loci and a phage ncRNA array reverse-transcribed into short cDNAs.

BOTANIC-1: a series of long-context plant genomic foundation models in the agentic era

Bidirectional Mamba2 plant genomic language models (318M to 3.2B parameters) trained on 320 land-plant genomes top a 22-task frozen-model aggregate; a small variant reaches 131,072 bp context.

A multi-tissue epigenomic atlas links the sheep non-coding genome to domestication and complex traits

SheepEpimap combines 516 histone CUT&Tag, ATAC-seq and RNA-seq datasets from 43 sheep tissues to annotate 2.93 million cis-regulatory elements and 145,407 allele-specific variants for trait interpretation.

GFMBench-API: A Standardized Interface for Benchmarking Genomic Foundation Models

Wrap a genomic foundation model's inference methods once to score it on supervised and zero-shot GUE, VariantBenchmarks, BEND, TraitGym and ClinVar tasks; code is promised upon acceptance.

FLIP2: Expanding Protein Fitness Landscape Benchmarks for Real-World Machine Learning Applications

FLIP2 adds seven protein fitness datasets and 16 splits for enzymes, light-sensitive proteins, and interactions; simple ridge and zero-shot baselines often rival fine-tuned protein language models.

Functional pocketomes and complete allosteric maps of five protein kinases

Energy-model maps of five human kinases show conserved distance-dependent decay of inhibitory allostery, but distal allosteric pockets and activating sites diverge between kinases.

BioReason-Pro: Advancing Protein Function Prediction with Multimodal Biological Reasoning

A reasoning LLM combining ESM3 embeddings with GO-GPT term predictions produces protein function annotations that experts rated tie-or-better than UniProt in 79% of cases (SFT model).

Machine learning-based prediction of human structural variation and characterization of associated sequence determinants

A sequence-only convolutional neural network and an annotation-based random forest predict human structural variant breakpoints from 400 bp windows (AUROC 0.903, 0.904); their ensemble reaches 0.916.

ARSENAL: Learning Transferable Regulatory DNA Representations with Targeted Short-Context Language Models

A 350 bp masked DNA language model trained on ENCODE cCREs with a frequency-domain regularizer recovers TF motifs zero-shot, scores regulatory variants, improves ChromBPNet inputs, and guides sequence design.

Replaying germinal center evolution on a quantified affinity landscape

Across 119 monoclonal germinal centers, B cell affinity rises reliably despite divergent phylogenies. Selection is noisy but persistent; mutability limits exploration. A DMS-based landscape shows saturating fitness and survivorship biases.

PARAS: high-accuracy machine-learning of substrate specificities in nonribosomal peptide synthetases

Random forest predictors PARAS and PARASECT, trained on 3,653 curated adenylation domains, outperform earlier NRPS substrate predictors and correctly assigned L-Trp to an unusual tryptopeptin synthetase domain.

DefensePredictor: A Machine Learning Model to Discover Novel Prokaryotic Immune Systems

DefensePredictor combines ESM2 protein language model embeddings with genomic context features in a gradient boosting classifier to identify prokaryotic anti-phage defense genes, validating 45 previously unknown Escherichia coli systems.

Protein Structure Informed Bacteriophage Genome Annotation with Phold

Phold pairs ProstT5-predicted 3Di tokens with Foldseek searches against 1,363,704 curated phage structures, annotating 49.4% of CDS in 1,419 INPHARED genomes versus 34.7% for Pharokka MMseqs2.

Helixer--de novo Prediction of Primary Eukaryotic Gene Models Combining Deep Learning and a Hidden Markov Model

Helixer, a hybrid convolutional/bLSTM network with a hidden Markov model post-processor, predicts eukaryotic primary gene models from DNA alone, outscoring GenemarkES and AUGUSTUS on most species.

Synteny-aware functional annotation of bacteriophage genomes with Phynteny

Phynteny combines ESM2 protein embeddings, a BiLSTM, and circular-attention transformer layers to exploit phage gene order, assigning one of nine PHROG functional categories to unannotated genes.

A contextualised protein language model reveals the functional syntax of bacterial evolution

Trained on over 1.3 million bacterial genomes represented as ordered protein sequences, the transformer foundation model Bacformer predicts protein interactions, operons, essentiality, and phenotypes while generating trait-conditioned synthetic proteomes.

Scalable emulation of protein equilibrium ensembles with generative deep learning

BioEmu, a diffusion model trained on structures, molecular dynamics trajectories, and stability assays, samples protein equilibrium ensembles and folding free energies within ~1 kcal/mol in GPU-hours.

Medium-sized protein language models perform well at transfer learning on realistic datasets

Medium-sized protein language models (e.g. ESM C 600M) with mean embeddings nearly match ESM-2 15B and ESM C 6B in transfer learning on mutational scanning and PISCES data.

Inference and visualization of complex genotype-phenotype maps with gpmap-tools

gpmap-tools uses Gaussian processes to infer and visualize genotype-phenotype maps with higher-order epistasis; across 262,144 Shine-Dalgarno variants it reveals register-specific peaks linked by three-nucleotide-offset ridges.

Genome modeling and design across all domains of life with Evo 2

A 40B-parameter DNA model trained on 9.3 trillion tokens scores coding and noncoding variants zero-shot, exposes interpretable sequence features, and writes chromosome-scale sequences.

Simulating 500 million years of evolution with a language model

ESM3, a 98-billion-parameter multimodal protein language model tokenized across sequence, structure, and function, designs a bright green fluorescent protein at 58% sequence identity to known fluorescent proteins.

Prediction of protein biophysical traits from limited data: a case study on nanobody thermostability through NanoMelt

NanoMelt, a ridge-stacked ensemble of regressors on four diverse sequence embeddings, predicts nanobody melting temperature from 640 curated sequences, reaching Spearman 0.832 under nested cross-validation.

ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis-regulatory sequence syntax, transcription factor footprints and regulatory variants

Learning and removing Tn5/DNase-I sequence bias lets a small base-resolution CNN (ChromBPNet) recover TF motifs and footprints from accessibility data and rival Enformer at predicting chromatin QTL effects.

Sequence modeling and design from molecular to genome scale with Evo

7B-parameter convolution-attention DNA model with 131 kb context, trained on prokaryotic genomes, performs zero-shot fitness and essentiality prediction and generates Cas loci, transposons, and 650 kb sequences.

The simplicity of protein sequence-function relationships

Reference-free analysis with a sigmoid link shows additive plus pairwise effects explain a median 96% of phenotypic variance across 20 combinatorial protein mutagenesis datasets.

ProteinGym: Large-Scale Benchmarks for Protein Design and Fitness Prediction

ProteinGym unifies over 250 deep mutational scans and ~65k clinical variants to benchmark over 70 models in zero-shot and supervised settings; TranceptEVE ranks best overall.

Cell-free biosynthesis combined with deep learning accelerates de novo-development of antimicrobial peptides

Autoencoder design plus a roughly 24-hour cell-free screen tested 500 candidates over five rounds, finding 30 functional AMPs; six were broad-spectrum with no resistance emerging in E. coli.