deep learning

AlphaGenome Atlas: in silico mutagenesis of the entire human genome improves prioritization and interpretation of non-coding variants

AlphaGenome Atlas predicts regulatory effects of non-coding variants across the human genome, improving their prioritization and interpretation.

CUBE: Multimodal Representation Learning Reveals Biological Structure Across Histomorphology, Spatial Protein Phenotypes, and Transcriptome-Associated Signals

CUBE integrates spatial protein and transcriptomic data via histology, creating a multimodal representation that captures biological structure from incompletely paired data.

Segment Any Plant (SAP): Foundation-Model Segmentation for Plant Time-Series Phenotyping

SAP provides a transferable, training-free framework for automated segmentation of plant time-series imagery across species and conditions.

Multi-model biological and sequence information fusion for gene regulatory network inference from single-cell transcriptomics

This paper introduces scMGFGRN, a deep learning framework for gene regulatory network inference from single-cell transcriptomics, which outperforms existing methods by integrating multimodal biological data.

Ancestral Sequences Cannot be Accurately Reconstructed via Interpolation in a Variational Autoencoder's Latent Space

This study tests using a variational autoencoder to reconstruct ancestral sequences by interpolating in its latent space, finding it less accurate than standard methods despite containing evolutionary signal.

One-dimensional CNNs for Near-Infrared Prediction of Protein and Moisture in Cereal Grains: The Effects of Architecture and Input Preparation

This study compares CNNs, PLSR, SVR, and XGBoost for NIR prediction of protein and moisture in cereal grains, finding that model accuracy depends more on input preparation and sample size than network architecture.

Pep-PU-GAN: Positive-Unlabeled Adversarial Learning for Peptide Function Prediction

Pep-PU-GAN improves peptide function prediction using PU learning and GANs with graph neural networks, achieving high F1 and AUROC scores on neuropeptide classification.

EM3DFold: accurate de novo protein and nucleic acid model building for cryo-EM maps using language model-powered deep learning

EM3DFold improves cryo-EM model building for proteins, nucleic acids, and complexes using a language model-powered deep learning framework, achieving high accuracy on benchmark datasets.

Img2EEG: A Scalable and Interpretable Encoding Framework for Simulating Human EEG Responses to Visual Inputs

Img2EEG generates synthetic EEG responses to visual inputs, generalizing across stimuli and participants while preserving neural structure and enabling targeted interventions.

A Unified 3D Generative Model for Synthesizable Structure-Based Drug Design

This paper introduces LDDM, a generative model for drug design that produces synthetically accessible compounds validated experimentally for five protein targets.

Virtual experiments bridge sequence and microscopy with generative models

The paper introduces a generative model framework that maps protein sequences to microscopy images and vice versa, enabling virtual mutagenesis and functional motif analysis.

Machine Learning for Toxicity Prediction in Low-Sample Molecular Classes

The paper proposes a framework to improve toxicity prediction for small molecular classes by fine-tuning a general model on class-specific data, outperforming classical methods and the EPA TEST tool.

AI-Powered Discovery of Novel RNA Viruses from the Permafrost of a 14,300-Year-Old Pleistocene Wolf

This study identifies two novel RNA viruses from a 14,300-year-old wolf specimen using AI and metatranscriptomic analysis, revealing their association with fungal microbiomes.

Machine learning reveals sequence and genomic context features underlying Alu-specific effects on genome folding

The paper uses deep learning to analyze Alu elements' impact on genome folding, identifying sequence and genomic context features that influence chromatin interactions and boundary formation.

Data-driven predictive design of engineered living hydrogels

The paper uses a tabular foundation model to predict material properties of engineered living hydrogels from genetic and process parameters, achieving high accuracy in predicting storage modulus.

LLMsFold: Integrating Large Language Models and Biophysical Simulations for De Novo Drug Design

LLMsFold designs and validates new small molecules targeting pathogenic proteins using LLMs and biophysical simulations, showing strong in silico binding and drug-like profiles.

Time-Series Foundation Models for Cognitive Workload Classification using Eye-Tracking Data

This study evaluates two time-series foundation models for cognitive workload classification from eye-tracking data, showing better generalization to unseen individuals than task-specific models.

Joint inference of paired dynamical gene regulatory networks reveals distinct cell-state landscapes of neutrophil reprogramming

The paper introduces NetDes-Duo, a method to jointly infer gene regulatory networks from scRNA-seq data, revealing distinct cell-state landscapes in neutrophil reprogramming under different conditions.

ConvexGating infers gating strategies from clusters in single cell cytometry data

ConvexGating automatically learns unbiased, interpretable gating strategies for flow cytometry data, improving consistency and reducing contamination in cell population identification.

A Hybrid Residual-Swin Transformer Design with Attention for Prostate Cancer Segmentation

The paper introduces RSAUNet, a deep learning model for prostate cancer segmentation in MRI scans, achieving high Dice and Jaccard scores.

Self-supervised representations reveal the genetic architecture of human cortical folding

The paper uses self-supervised deep learning representations of cortical folding to identify more genetic loci associated with cortical shape than classical morphometry in a large cohort.

Novel two-stage deep learning-based approach applied to gene expression data pertaining to esophageal adenocarcinoma boosting biological knowledge discovery

The paper introduces a two-stage deep learning approach for gene expression analysis in esophageal adenocarcinoma, identifying more expressed genes than traditional methods.

ReScale4DL: Balancing Pixel and Contextual Information for Enhanced Bioimage Segmentation

This paper evaluates how reducing image resolution affects deep learning performance in bioimage segmentation, finding that lower resolution can improve accuracy or cause minimal degradation.

Independent noise realizations enable morphologically agnostic image reconstruction in single-photon-sensitive microscopy

This paper introduces a morphology-agnostic regularization framework for image deconvolution in single-photon-sensitive microscopy, eliminating semi-convergence and improving reconstruction quality.

Balanced contractility and adhesion drive polarization in a minimal elastic actomyosin network

This paper shows that a minimal mechanical model of an actomyosin network can spontaneously polarize and move through balanced contractility and adhesion, without external cues or complex signaling.

Interrogating contrastive learning embeddings for structure-based virtual screening: a case study on DrugCLIP

This paper evaluates DrugCLIP's contrastive learning embeddings for virtual screening, showing strong performance but limitations under realistic conditions.

Mechanistic Interpretability of Fine-Tuned Protein Language Models for Nanobody Thermostability Prediction

This paper fine-tunes a protein language model on nanobody thermostability data and uses sparse autoencoders to uncover biophysical rules governing stability, identifying both known and novel stabilizing features.

BFVD v3-UniProt-complete, improved viral protein structure predictions

5,776,417 ColabFold-AlphaFold2 structures cover 98.9% of viral UniProt 2025_03 sequences, 16.4-fold more than earlier versions, with 75.3% high-confidence and 72.6% of viral reference proteomes fully covered.

A novel, high-density CRISPR activation platform for mapping cancer dependencies and resistance pathways ex vivo and in vivo

This paper introduces Partita, a high-density CRISPRa platform for comprehensive gene activation studies, demonstrating its utility in identifying cancer dependencies and resistance pathways in vitro and in vivo.

Automated acoustic classifiers provide insights into calling patterns of cicadas in tropical Australia

The paper evaluates automated acoustic classifiers for detecting cicada species in tropical Australia, finding high accuracy with a precision-prioritised threshold.