deep learning

SHOT-CCR: Biologically guided adversarial training for test-time adaptation in cellular morphology

SHOT-CCR improves classification of genetic perturbations in Cell Painting datasets by reducing batch effects through biologically guided adversarial training.

Cell-Lens: Advancing LLM Reasoning at Single-Cell Resolution

Cell-Lens improves LLM reasoning on single-cell data by using a structured representation that captures more biological content than gene lists alone.

A Context-Aware 14-3-3 Binding Predictor Enabled by Data Augmentation, Multi-Scale Biological Features, and Protein Language Model Embeddings

The paper introduces CAMP-14-3-3, a framework that improves 14-3-3 binding prediction by integrating multi-scale biological features and protein language model embeddings, outperforming existing methods.

A comprehensive Arabidopsis transcription factor binding atlas reveals pervasive positional and syntactic organization of their DNA binding

This study creates a comprehensive Arabidopsis TF binding atlas, revealing family-specific binding patterns and the importance of DNA sequence syntax in TF-DNA interactions.

Ability of a Structural World Model to Detect Cryptic Pockets from Apo Structure

This paper introduces a method to detect cryptic pockets from apo protein structures using a structural world model, achieving high accuracy and outperforming existing approaches.

Integrating glycosylation in de novo protein design with ReGlyco Binder Design Filter

The paper integrates glycosylation filtering into a protein design pipeline to improve efficiency and reduce costs in binder design.

Speciesformer learns conserved cellular states for cross-species generative virtual cell modeling

Speciesformer learns conserved cellular states across species to enable cross-species generative modeling of virtual cell states and their responses to interventions.

RecallRefine: Coverage-Constrained Coarse-to-Fine Segmentationfor Small 3D Lesions

RECALLREFINE improves 3D lesion segmentation by addressing coarse-stage misses through coverage-constrained candidate selection, increasing sensitivity and Dice score for small lesions.

Scalable Computational Phenomics of Nuclear Morphology

NucXplore converts H&E-stained nuclei into 129 explicit features, outperforming classical and deep learning methods in histopathology cohorts and revealing aging and exposure-related cellular changes.

Capsid-specialized protein language models reveal higher-order viral architecture from sequence

This paper introduces ESMCapsid, a protein language model that detects viral capsids and infers their architecture from sequences, revealing hidden structural patterns in metagenomic data.

miRstring: An RNA language model enables mature miRNA decoding and artificial small RNA design across species

miRstring decodes mature miRNA boundaries from precursor context across species and designs artificial miRNAs effectively.

GapDream: Active Synthetic Evidence Acquisition for Long-Tailed Medical Imaging

GAP DREAM improves medical image generation by actively acquiring synthetic evidence that reduces evidence gaps, achieving better performance with fewer samples than random or balanced methods.

Leakage-controlled benchmarking reveals generalization limits of deep learning for protein-ligand binding affinity prediction

This paper introduces PLABench, a benchmarking framework for protein-ligand binding affinity prediction, revealing that structural input and training scale alone do not ensure cross-target generalization.

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.

Mechanism of GPR84 allosteric modulation at a helix 8-proximate site

This paper identifies a helix 8-proximate allosteric site in GPR84 and shows how it enables Gi-biased signalling through structural and dynamic mechanisms.

High-resolution postmortem MRI of the human infant brain: cortical reconstruction and anatomical parcellation

This paper introduces a computational framework for automated anatomical segmentation and cortical reconstruction in high-resolution postmortem infant MRI, showing improved performance across diverse datasets.

Machine learning of honey bee olfactory behavior identifies repellent odorants in free flying bees in the field

The paper uses machine learning to identify repellent odorants for honey bees, validated in both lab and field settings, showing potential for pesticide mitigation.

DIRTNet: Enabling Root Phenotyping with Fiber Bragg Grating Sensors

DIRTNet, a hybrid deep learning model, accurately predicts root depth and diameter from FBG sensor data in both simulated and real maize root growth scenarios.

Generative Access to Make-on-Demand Chemical Space Enables Ultra-Large Virtual Screening

The paper introduces REAL-SWIT, a generative framework for ultra-large virtual screening that efficiently explores chemical space by generating molecules with favorable docking scores.

Multi-Stain Fusion of Histopathology Images Using Deep Learning for Pediatric Brain Tumor Classification

This study uses deep learning to classify pediatric brain tumors using H&E and Ki-67 images, finding that multi-stain fusion improves classification accuracy compared to single-stain models.

ErrorRoute: Budgeted Utility-Aware Resolution Routingfor 3D Medical Image Segmentation

ERRORROUTE improves 3D medical image segmentation by selectively allocating computation to regions where additional processing provides the most utility, achieving higher accuracy with lower computational cost.

The generation and validation of a de novo-designed titin binder for use in fluorescence microscopy

The paper designs and validates a de novo protein binder for the titin M10 domain, demonstrating its utility in fluorescence microscopy without antibodies.

An interpretable deep learning framework for classifying neuronal morphologies using topology and graph neural networks

This paper presents a deep learning framework that classifies neuronal morphologies using topology and graph neural networks, achieving accuracy comparable to human experts.

Characterizing Compounds Targeting Colorectal Cancer Derived From Monastrol Using High-Through Screening of an Extensive Combinatorial Library

This paper identifies two fascin-binding compounds using a virtual screening approach combined with experimental validation, showing they inhibit CRC cell proliferation and migration.

Improving racial fairness in brain age models using style-transfer synthesis

This paper improves racial fairness in brain age models by using synthetic T1 images generated via style-transfer, reducing disparities without requiring race as an input.

Generative Design of New-to-nature Biosynthetic Assembly Lines with Genomic Language Modeling

This paper uses a genomic language model to design and optimize a synthetic biosynthetic pathway for producing a non-natural molecule, achieving a 9.4-fold increase in production.

AutoWrinkleID: a machine learning pipeline for biofilm wrinkle identification and quantitative analysis

AutoWrinkleID is a machine learning pipeline for automated biofilm wrinkle detection and quantification, showing that motility-based annotations outperform manual and constitutive fluorescence annotations.

Molecular insights into ATP synthase c-ring accumulation in BMP-deficient lysosomes in Batten disease

This study investigates how BMP deficiency in Batten disease leads to lysosomal accumulation of ATP synthase c-rings through molecular simulations showing BMP's preferential interaction with the c-ring.

EZSolver: Template-free prediction of polar enzymatic mechanisms via bidirectional flow matching and search

EZSolver predicts polar enzymatic mechanisms without templates using flow matching and beam search, achieving 60% accuracy on unseen reactions.

Optimizing MR-based gaze-decoding for eyes-closed eye-tracking in fMRI

This paper improves gaze decoding in fMRI during eyes-closed states using a deep learning framework, showing it can generalize to eyes-closed conditions with task-specific fine-tuning.