protein sequence

This study combines machine learning and directed evolution to optimize the monooxygenase HpaBC for improved L-Dopa production activity.

TM-Vec 2s improves protein remote homology detection by being faster and more accurate than previous methods, enabling efficient large-scale database searches.

This paper introduces Boltz2ESI, a framework predicting enzyme-substrate interactions using a foundation model, outperforming existing methods by capturing active-site plasticity and evolutionary context.

TriFlow, discrete flow-matching sequence designer with global three-track structure attention, raises in silico binder success over ProteinMPNN in RFdiffusion and BindCraft-style pipelines on 500 targets and flags functional sites.

The paper shows that naive fusion of sequence and graph modalities in TCR-peptide binding prediction can degrade performance and introduces TRACE, a contrastive alignment method that stabilizes multimodal learning.

The paper examines the evolutionary divergence of the RAR1-SGT1 chaperone complex in leptosporangiate ferns, identifying key residues that dictate binding specificity and revealing a stepwise evolutionary pathway.

The paper introduces CORAL, a deep learning framework for RNA-protein interaction prediction that improves generalization and interpretability through cross-attention mechanisms.

LLPSense predicts and modulates condition-dependent protein phase separation using machine learning, revealing new insights into disease mechanisms and enabling programmable condensates.

The paper identifies a macrocyclic peptide that selectively binds NRAS with isoform and activation-state specificity, but lacks membrane permeability and functional inhibition.

This study evaluates protein language model embeddings for estimating structural similarity in the protein-sequence twilight zone and finds that ProstT5 performs best.

A parallelized BindCraft pipeline generated 306,146 in silico binder candidates against 8,296 human proteins, each with sequence, predicted complex and confidence metrics, freely accessible online.

This paper introduces EffieDes, a neuro-symbolic AI framework for protein sequence design that combines deep learning with automated reasoning to generate sequences with complex functional properties.

This paper introduces ILMC, a sampling strategy for protein language models that improves diversity while steering generation toward desired properties.

The paper introduces SPARK-ID, a method to map dynamic DNA damage response interactomes, revealing modular networks coordinated by connector proteins.

ViralMap predicts ten annotation classes for viral proteins from primary sequences, achieving high performance on seven classes and generalizing across viral strains and families.

The paper uses AlphaMissense and experimental assays to analyze 78 G6PC1 variants, identifying stability and allosteric modulation as key pathogenic mechanisms.

ProtSyntax is a protein language model that improves PTM site prediction and functional inference by integrating contextual and structural information.

The paper uses genome-informed machine learning to discover a new enzyme, PTC, that catalyzes phenazine thioconjugation from sparse data.

The paper develops a method to classify influenza A virus subtypes using protein sequences and identifies key amino acids associated with subtypes.

This study identifies surface secondary binding sites in β-glucosidase BglB that regulate glucose inhibition and catalytic activity through long-range structural and energetic changes.

The paper uses a protein language model to analyze the evolutionary history of bacterial aromatic polyketide ketosynthases and chain length factors, identifying a shared ancestral cluster and distinct evolutionary paths.

AINN-Express predicts VHH antibody expression using only sequence data, achieving high ROC-AUC scores in both known and new programs without requiring structural data or task-specific training.

The paper compares the distribution of young de novo proteins, old canonical proteins, and randomized sequences in sequence space using k-mer-based distances, finding distinct patterns and potential pathways between classes.

The paper measures protein-protein binding affinities in live cells by analyzing co-localization between fluorescently labeled proteins, revealing concentration-dependent interactions.

X-PAIR is a multitask deep learning framework that predicts protein-protein interactions and identifies interface residues from sequence data, outperforming existing methods and enabling proteome-scale analysis.

The paper introduces pHaseMD4AI, a molecular dynamics dataset with chemical and pH perturbations, to support biomolecular AI research and benchmarking.

This study analyzes how biological concepts are encoded in protein language models across their layers and finds that complexity increases with depth, with pretraining data and compute having a greater impact than parameter scaling.

TrioNsight is a meta-predictor that evaluates the clinical impact of TrioN-like Dbl-homology domain variants using structural, evolutionary, and physiochemical features.

fastCDS maps protein domains to genomic coordinates efficiently, enabling rapid analysis of exonic architecture and variant interpretation.

This study shows that evolutionary divergence in LRRK2 interaction domains leads to species-specific protein interaction networks in humans and mice.