deep learning

Single-cell multiomics reveals archetypal regulatory programs shared across CD4 and CD8 T cell subsets in viral infection

NeCS-Flow: Necrosis-Guided Contrast-Enhanced MRI Synthesis with Rectified-Flow Refinement

NeCS-Flow synthesizes gadolinium-free post-contrast T1w brain MRI using necrotic scaling and PMRF refinement, achieving perceptual realism and diagnostic equivalence.

Improved cryo-EM reconstruction of sub-50 kDa complexes using 2D template matching

This paper improves cryo-EM reconstruction of small complexes using 2D template matching and high-resolution structures as priors, enhancing density of ligand-binding sites in a 43 kDa protein kinase.

Rapid and Reliable Structural Modeling of Adaptive Immune Receptors Using an Optimized AlphaFold3 workflow

This paper optimizes AlphaFold3 for rapid and reliable structural modeling of antibodies and T-cell receptors, achieving 45-fold faster MSA phase and 3.6-fold faster inference.

Can Folding Models Tell Binders from Bluffers? Evidence from POISK: The Patent-Derived Antibody Dataset

The paper benchmarks structure prediction models on a new antibody dataset to assess their ability to distinguish binders from decoys, finding moderate success but limitations in epitope localization.

Design of full-length agonist and antagonist antibodies for hard targets

This paper designs full-length agonist and antagonist antibodies for hard targets using a computational pipeline, achieving nanomolar affinities and functional activity.

CBAM-enhanced MaxViT-Tiny for gastrointestinal endoscopy image classification: Robustness, explainability, and external validation

This paper proposes a CBAM-enhanced MaxViT-Tiny model for GI endoscopy image classification, achieving high accuracy but showing vulnerability to adversarial attacks and unreliable explanations.

flodl-bio: First Rust data, transformers and model layer for entire deep learning on biological sequences

flodl-bio introduces a Rust framework for deep learning on biological sequences, achieving state-of-the-art performance on transcription-factor-binding datasets.

Building from the ground up: augmentation strategies to improve protein language model performance on protein sets

This study investigates augmentation strategies to improve protein language model performance by reducing dataset size and enhancing generalization through active learning and generative models.

ALHENA, an H&E encoder distilled from transcriptomic foundation models for multi-scale slide analysis

ALHENA improves transcriptomic signature prediction and cell-type separability using H&E images alone, without requiring spatial transcriptomics data.

RiboCouple links RNA backbone generation to sequence design through a learned interaction topology

The paper introduces RiboCouple, a method that links RNA backbone geometry to sequence design using learned interaction topologies, improving refolding validity in RNA structure prediction.

ScienceRSI: Recursive Scientific Self-Improvement

ScienceRSI improves scientific AI through recursive self-improvement by transferring capabilities across model scales, achieving top performance on multiple benchmarks.

A Real-Time Automated Deep Learning Workflow for Non-invasive High-Magnification Imaging of C. elegans

This paper develops a real-time deep learning workflow for non-invasive high-magnification imaging of C. elegans, enabling stable imaging across magnifications without immobilization.

Structural probing of RNA hairpins quantifies protein occupancy on RNA and links it to function in human cells

This paper introduces a method to measure RNA structure and protein binding in human cells, showing how RNA folding and sequence determine protein occupancy and function.

Viral host assignment across the tree of life with VHAMSTeR

VHAMSTeR predicts viral host associations across five categories using genomic language models and biological features, outperforming existing methods and uncovering new RNA bacteriophages.

Sequence-only antigen context improves modeling of antibody hypervariable regions

This paper introduces LangAAI, a sequence-only model that improves antibody CDR design by conditioning on antigen sequences, showing better performance than existing methods without requiring 3D structures.

An integrative cardio-informatics approach identifies positionally conserved ancestors of human lncRNA genes transcribed across vertebrate hearts

The paper identifies conserved genomic positions of human heart-related lncRNA genes across vertebrates using integrative analysis and a neural network tool.

The Impact of Noise on Memorability in Minds and Machines

This study compares memorability estimates from deep neural networks and humans under noise, finding divergence and introducing GraphMem, a relational model that better aligns with human performance.

Multi-Peptide Prompting Enables In-Context Learning in Protein Language

This paper introduces multi-peptide example prompts (MPEPs) to enable in-context learning in protein language models for peptide classification without training.

RNASTOP: A Deep Learning Framework for mRNA Chemical Stability Prediction and Optimization

RNASTOP predicts and optimizes mRNA chemical stability using deep learning, improving accuracy by 13% and reducing free energy in a vaccine sequence.

Physics-driven self-supervised learning for quantitative high-fidelity structured illumination microscopy

This paper introduces qHiFi-SIM, a physics-driven self-supervised learning framework for quantitative high-fidelity structured illumination microscopy, achieving excellent intensity linearity and structural fidelity without ground-truth data.

Overcoming segmentation bottlenecks in brain tissue for spatial proteomic inference at population scale

The paper introduces SPInfer, a workflow for spatial proteomic inference that improves cell segmentation in human brain tissue, enabling better biological discovery at population scale.

Vision Transformers as a Framework for Imaging Fluorescence Correlation Spectroscopy

The paper introduces ImFCSViT, a Vision Transformer framework that improves fluorescence correlation spectroscopy by directly analyzing raw image data and extracting biophysical parameters more efficiently than traditional methods.

A physics-informed neural network framework for inferring infinitesimal phase response curves from finite-pulse circadian recordings

This paper introduces PINN-iPRC, a neural network framework that infers infinitesimal phase response curves from finite-pulse circadian recordings, achieving accurate results comparable to specialized experiments.

Alignment-free discovery of conserved RNA structures in the human genome

This paper introduces RLMHelix, an alignment-free method to discover conserved RNA structures in the human genome, identifying new candidates that correlate with RNA-binding protein interactions.

A clinically annotated proteomic and histological resource of 1,000 colorectal cancers

Deep proteomics of 997 treatment-naïve colorectal cancers defines six subgroups, splitting chromosomally unstable tumors into three and revealing a rare adverse subgroup, with H&E slide benchmarks.

OCTRON - a general purpose segmentation and tracking pipeline for behavioral experiments

OCTRON provides a segmentation and tracking pipeline for behavioral experiments, validated across species and setups.

AI-driven phenotypic profiling of organoids

Morphi is an AI framework that quantitatively profiles organoid phenotypes using label-free microscopy, enabling drug screening by capturing viability and mechanistic responses.

Multi-Angle Lateral-View Breast Tissue Classification in Thermographic Images Using Deep Reinforcement Learning and Statistical Shape Modelling

This paper develops a hybrid method combining deep reinforcement learning and statistical shape modeling to improve breast tissue classification in thermographic images, achieving 74.1% accuracy.

Projected Gradient Descent for Hallucination-Based Protein Binder Design

This paper evaluates projected gradient descent (PGD) for hallucination-based protein binder design, finding it improves design efficiency and diversity compared to softmax-based methods.