graph neural network

This paper develops a graph-based model to predict small-molecule partitioning into biomolecular condensates using existing experimental data and applies it to identify chemically diverse partitioners.

GraphTox predicts peptide toxicity using a geometric graph transformer and LORA-based finetuning, outperforming existing methods by incorporating structural information.

GAE-Δ uses a graph autoencoder to analyze multi-omics data and predict clinical outcomes while identifying gene network rewiring associated with cancer progression.

The paper introduces triAMPh, a framework for predicting pathogen-specific antimicrobial peptide activity using biological large language models, showing it matches or outperforms existing methods.

HapFold improves haplotype reconstruction for diploid genomes by combining graph-based and sequence-based scaffolding methods, achieving high accuracy and contiguity.

AE-PocketMiner predicts cryptic protein pockets and their allosteric coupling using a graph neural network, outperforming existing methods and validating predictions experimentally.

This paper introduces HAST, a framework for spatial transcriptomics that adapts graph learning scales to tissue heterogeneity, improving spatial structure identification and tumor region characterization.

DGAT imputes spatial protein expression from transcriptomics-only data, revealing cell states and tissue organization not visible from RNA alone.

DMG-PPI is a dual-channel graph neural network that models similarity and complementarity in PPI networks, improving prediction accuracy and providing interpretable insights into binding mechanisms.

The paper predicts age and MMSE scores from brain connectivity graphs using a graph-based deep learning model, showing improved performance over existing methods.

The paper introduces a deep learning model to predict peptide retention times in liquid chromatography and identifies key factors influencing retention through saliency mapping.

This paper introduces a graph-based framework for correcting nucleus segmentation errors in 3D reconstructions, achieving high accuracy on brain-scale benchmarks.

This paper introduces F2S, a method that predicts drug side effect frequencies by accounting for chirality, achieving comparable performance to state-of-the-art methods with fewer false positives.

The paper investigates the relationship between cell shape, protein expression, and tissue state using spatial imaging and machine learning, finding a bidirectional link and identifying p53-positive tumor cells.

DynoSys 2.0 models dynamic behavioral risk states using graph-based temporal modeling with longitudinal data from the ABCD Study, showing better prediction for externalizing behaviors and substance use initiation.

This paper develops an uncertainty-aware framework for PAD biomarker discovery using graph neural networks and positive-unlabeled learning, identifying novel and validated candidate proteins.

The paper develops a closed-form equation to predict curvature changes in thoracic aortas from CT scans, showing strong performance on synthetic and clinical data.

pyTrance identifies co-localizing RNAs in spatial transcriptomics data with higher accuracy than existing methods, revealing biologically relevant RNA co-localization patterns in mouse brain tissue.

InterScale models both local and global cellular interactions in spatial transcriptomics data, revealing spatially restricted differentiation programs and disease-associated tissue remodeling.

This paper introduces a signed multi-omics knowledge graph foundation model for drug action and clinical response predictions, achieving high accuracy and efficiency.

G-SPRI predicts cancer driver events from missense mutations by integrating structural and mutation-specific features using a graph model.

The paper proposes a computational framework for zero-shot control in dynamic pursuit tasks and demonstrates its implementation in a graph convolutional network, showing transferability and change-of-mind behavior, supported by neural recordings.

PEARL is a deep graph learning method for multi-omics integration that outperforms existing methods and identifies functionally important genes in Alzheimer's disease.

TopoFuseNet improves drug synergy prediction by integrating hierarchical molecular structures and multi-scale topological features, outperforming existing methods on benchmark datasets.

The paper introduces immuneKG, a knowledge graph framework for immune-mediated diseases, achieving high performance in target discovery using a graph neural network and novel features.

KG-Bench benchmarks graph neural networks for drug repurposing using the Open Targets dataset, finding TransformerConv achieves the highest performance.

MUSE integrates multi-omics data across species to align cellular states while preserving species-specific features and regulatory module conservation.

The paper introduces PEPR-GNN, a graph neural network framework to model regulome responses during genetic perturbations using Multiome Perturb-Seq data.

GatorSC improves single-cell transcriptomic analysis by integrating multi-scale cell and gene graphs with a Mixture-of-Experts fusion framework, outperforming existing methods in clustering, imputation, and annotation.

Spartan identifies spatial domains and variable genes by integrating spatial topology and activation signals in spatial transcriptomics data.