graph neural network

DeepDrug uses graph-based deep learning to predict drug-drug and drug-target interactions, outperforming existing methods and identifying potential SARS-CoV-2 treatments.

Neurospectrum extracts interpretable spatiotemporal features from neural activity using a geometric and topological deep learning framework, outperforming traditional methods in simulations and fMRI data.

This study uses computational models to analyze how mutations in SARS-CoV-2 Omicron variants balance protein stability and ACE2 binding through epistatic interactions and dynamic changes.

The paper compares sequence- and structure-based ML models for predicting protein residue properties, finding they excel at different tasks and that combining them improves accuracy.

This study evaluates how different node features affect GCN performance in brain network classification, finding that node correlation provides higher accuracy than other features.

The paper improves drug repurposing for SARS-CoV-2 by enhancing a graph neural network model, generating candidate drugs, and optimizing computational efficiency.

The paper introduces two deep learning models to predict carbohydrate binding sites on proteins, with one model outperforming the other and showing effectiveness on both experimental and predicted structures.

This paper introduces STdGCN, a graph neural network for spatial transcriptomic cell-type deconvolution, outperforming existing methods and revealing spatial patterns in cancer and heart tissue.

E2EGraph predicts prostate cancer pathological stages using gene expression data and graph neural networks, achieving state-of-the-art performance with interpretable marker gene identification.

The paper constructs a graph-based sheep pan-genome to identify structural variations and their association with tail phenotypes in domestic and wild sheep.

GraphST improves spatial transcriptomics analysis by integrating clustering, integration, and deconvolution through graph self-supervised contrastive learning, achieving higher accuracy than existing methods.

The paper introduces ChemWalker, a method that improves annotation propagation on molecular networks using random walks, enhancing scalability and efficiency compared to previous approaches.

Verkko automates telomere-to-telomere diploid genome assembly using long-read sequencing and graph-based methods, achieving high accuracy in chromosome assembly.

This paper introduces a graph-based method to estimate the number and strength of synaptic inputs to motor neurons using discharge rate correlations, validated through simulations and applied to experimental data.

This paper develops a hierarchical graph attention network to identify PTSD-associated genetic biomarkers by integrating domain knowledge and deep learning.

The paper identifies 17 up-regulated biomarkers for psoriasis-related melanoma using computational analysis of gene expression data and network biology.

scGREAT is a tool for analyzing regulatory elements in single-cell multi-omics data, enabling heterogeneity analysis and regulatory state inference.

The paper develops a graph neural network (GNN) that incorporates quantum chemical features to improve mass spectrum prediction, achieving better performance than baseline models.

The paper introduces OBNB, a benchmarking resource for GNN-based network biology tasks, providing datasets and a Python toolkit for evaluating gene classification models.

This paper applies graph convolutional neural networks to multimodal brain connectome data to classify schizophrenia, achieving an F1 score of 0.75.

The paper introduces a graph-wavelet model to predict protein biophysical properties and assess mutational effects by analyzing amino-acid features in protein structures.

PanPA generates and aligns panproteome graphs to improve alignment accuracy in prokaryotic genomes by leveraging amino acid similarity over DNA sequence diversity.

ReTrace evaluates tractography algorithms using Reeb graphs, showing no single algorithm consistently outperforms others and deterministic methods better capture fiber bundle properties.

The paper uses deep learning to model mutational effects on Aβ42 peptide aggregation traits from DMS data, showing improved performance with geometric-structural features.

This paper introduces CAME, a heterogeneous graph neural network for cross-species cell-type assignment in scRNA-seq data, showing improved performance and revealing evolutionary patterns.

The paper introduces a graph-based method to reconstruct molecular networks from DNA sequencing data, achieving accurate spatial recovery with minimal assumptions.

The paper introduces EnQA, a 3D-equivariant graph neural network for assessing protein model quality, achieving state-of-the-art performance and outperforming AlphaFold2's own scores.

This paper introduces HiC-GNN, a graph neural network for 3D chromosome reconstruction from Hi-C data, showing superior accuracy and generalization across resolutions and cell lines.

Stardust is a visualization method that embeds cells and marker genes together on a 2D map, performing competitively with existing tools on scRNA-seq and spatial transcriptomics data.

The paper uses graph-based pangenomics to improve genotyping density and reveal structural variation's role in fungal resistance in melon.