deep learning

Kidzoi Enables Cell-Type-Specific Regulatory Activity and Variant Effect Prediction in Kidney

Kidzoi is a kidney-specific sequence-to-function model that improves variant effect prediction and chromatin accessibility prediction in proximal tubule cells.

Lorentz Tangent Diffusion Adapter for Multi-label Prediction of Protein-Protein Interaction Types

The paper introduces LTDA, a post-embedding module for PPI type prediction, showing improvements over existing methods on benchmark datasets.

AmyloDeep: pLM-based ensemble model for predicting amyloid propensity from the amino acid sequence

AmyloDeep predicts amyloid propensity from amino acid sequences using an ensemble of deep learning models, achieving high accuracy and robustness.

Loop engineering and activity improvement of TEV protease by a phagemid-based selection system

This study improves TEV protease activity through loop engineering using a phagemid-based selection system combined with deep learning.

Mitigating Class Imbalance in Graph-Structured Data via Hierarchical Learning: Insights from Protein Binding Site Prediction

The paper introduces CLARA, a hierarchical learning framework to mitigate class imbalance in graph data, achieving high F1-score in protein binding site prediction.

Distinct sequence grammars of nucleosomal and linker DNA shape mutational and methylation landscapes in cancer

The paper develops DeepND to analyze nucleosomal and linker DNA sequence patterns, identifying motifs linked to cancer mutations and methylation.

The mammalian genome is punctuated by regularly spaced nucleosome islands

The paper trains a deep-learning model to predict nucleosome occupancy from DNA sequence in mouse embryonic stem cells and identifies short DNA motifs that organize nucleosome arrays.

G4SCOPE identifies sequence features and epigenetic contexts of chromatin G-quadruplexes

G4SCOPE predicts chromatin G-quadruplex occupancy from DNA sequence, revealing epigenetic differences in BRCA1 and BRCA2 deficient cells.

Antibody-Antigen Affinity Prediction with Chain-Aware Protein Language Modeling

AbAffinity predicts antibody-antigen affinity using a chain-aware sequence model, achieving high correlation on benchmark datasets and maintaining performance without structural data.

Structure-Informed Stress-Testing of the Standard Numbering Scheme for Class B β-Lactamases Using AlphaFold2

The paper uses AlphaFold2 to model class B β-lactamases without experimental structures, tests the BBL numbering scheme's validity, and identifies structural exceptions.

Near-atomistic simulations reveal the molecular principles that control chromatin structure and phase separation

This study introduces OpenCGChromatin, a model that simulates chromatin structure and phase separation, revealing how linker-DNA length and histone modifications affect chromatin organization.

Deep generative embeddings of gene expression and splicing reposition the interpretation of single-cell transcriptomic signatures

Crecerelle integrates gene expression and alternative splicing to improve cell-type annotation and uncover regulatory pathways in single-cell transcriptomics.

Rust fungi secretomes contain structurally diverse effector families including cysteine-rich metal-binding proteins

This study uses AlphaFold2 to analyze the structural diversity of effector proteins in rust fungi secretomes, identifying structurally related clusters and metal-binding effector candidates.

Tracking gene flow in the gut microbiome reveals phage activation and host-range expansion in response to diet

The paper tracks gene flow in the gut microbiome, showing diet and community ecology influence phage activation and horizontal gene transfer.

InsectDCT: A generalized pipeline for detection, taxonomic classification, and tracking of insects in camera-trap recordings

InsectDCT detects, classifies, and tracks insects in camera-trap footage using deep learning, achieving high accuracy in complex environments.

Deep learning of fossil pollen morphology reveals 25,000 years of ecological change in East African grasslands

This paper uses deep learning to analyze fossil grass pollen morphology and reconstruct 25,000 years of ecological change in East African grasslands, showing reduced diversity and shifts in C3:C4 ratios during the last ice age.

GGE: General-purpose deep meta-learning for classification of human transcriptomes with limited data

GGE is a deep meta-learning framework that improves transcriptome classification with limited data by leveraging patterns across diverse biomedical datasets.

AbRefine: Framework-Conditioned Refinement for Structure-Guided Antibody CDR Design

AbRefine improves antibody CDR design by fusing structural and sequence predictions using framework context, achieving better recovery and lower log-likelihood than static fusion.

Reliable Evaluation of Transductive Population Graph Neural Networks for Multisite Autism fMRI

Site-aware population GNN on ABIDE-I: cohort AUC about 0.94 (kept with site-only edges) was not reproduced by a site-prior classifier (0.506); a directed-edge support gain (+0.252) vanished when symmetrised (−0.003).

A Structure-Aware Generative Framework for Exploring Protein Sequence and Function Space

This paper introduces a structure-aware generative framework that integrates geometric encoding and coevolutionary constraints to map, cluster, and design protein sequences, improving homology detection and structure-guided sequence generation.

Embedding wear assessment in musculoskeletal simulation: A proof-of-concept application to total hip arthroplasty

This paper presents a computational framework for embedding wear estimation in musculoskeletal simulations, demonstrating its feasibility in total hip arthroplasty.

Neural networks learn forward dynamics when freed from numerical integration

The paper shows that a two-stage model (APE) outperforms a direct-mapping neural network in predicting limb dynamics under novel conditions by explicitly modeling equations of motion and numerical integration.

LOAM: A family of genomic language models trained on long-read soil metagenomes

LOAM is a family of genomic language models trained on long-read soil metagenomes, showing competitive performance despite a small training corpus and highlighting the importance of model scale and training data composition.

Dendritic Morphology Constrains Microtubule Dynamics and Organization

This study shows that dendritic morphology constrains microtubule dynamics and organization in Drosophila neurons, challenging the traditional view that microtubules shape cell morphology.

Design-Evolution Synergy Enables New-to-Nature Enzyme Chemistry

This study combines design and evolution to create new-to-nature enzymes capable of performing regio- and enantioselective C-H amination reactions for synthesizing chiral piperidines.

Multi-omic analysis reveals maturation programs in human pluripotent stem cell-derived cardiomyocytes during long-term culture

This study identifies molecular pathways and regulatory programs associated with hPSC-CM maturation through multi-omic profiling during long-term culture.

EnsPlex: Integrative Protein Complex Prediction through Multi-source Complementary Structural Sampling and Topology-Aware Candidate Selection

EnsPlex improves protein complex structure prediction by integrating multi-source sampling and topology-aware candidate selection, achieving a 27.8% relative improvement over AlphaFold3 in antigen-antibody systems.

Deep learning models detect neuromuscular disease and quantify functional impairment from video-derived biomechanics data

This paper develops deep learning models to detect neuromuscular diseases and quantify movement impairment using video-derived biomechanics data from smartphone recordings.

Cell type-resolved connectomic reconstruction of the spinal dorsal horn reveals sensory neuron synaptic organization and modality-specific inhibition

The paper presents tools to reconstruct spinal cord circuits and identifies modality-specific inhibitory synapses between sensory neurons and inhibitory neurons.

Accurate ab initio gene prediction in eukaryotes with Tiberius in multiple clades

Extended Tiberius models outscore Helixer and ANNEVO on all 33 genomes from the five clades all three tools cover, and approach BRAKER3 accuracy in four clades without extrinsic evidence.