#18 · ML / AI methods
Graph models encode relational topology in spatial, glycan and drug-target data
The dominant thread is graph-structured representation learning applied to biological entities, with relational topology encoded explicitly. stMFG partitions spatial transcriptomics spots into domains via multi-scale graph fusion and inter-view contrastive learning, while MAGNET uses a multi-view graph autoencoder with cell-gene attention to reconstruct cell-cell interaction networks and is reported to outperform TENET. GlycanGT applies a graph transformer to glycan monosaccharide graphs, pretrained with masked language modeling. LGABAN, DCGAT-DTI, and HGANMDA apply graph attention or adversarial graph encoding to drug-protein, drug-target, and microbe-drug association prediction, with LGABAN and DCGAT-DTI further integrating sequence or pretrained language-model embeddings. The new element across these items is explicit relational structure, spatial, glycan, or interaction, paired with multi-view or multimodal fusion. The set is heterogeneous, so this describes a shared approach rather than a uniform method.
Published papers
LGABAN: An Integrated Multi-Scale Approach Combining Graph and Sequence Features for Enhanced Prediction of Drug-Protein Interactions.
LGABAN, a dual-branch deep learning framework combining local and global drug and protein features with bilinear and graph attention, improves drug-target…
HGANMDA: A Heterogeneous Graph Adversarial Network for Multimodal Microbe-Drug Association Prediction.
HGANMDA, a heterogeneous graph adversarial network integrating multimodal data with attention-based multichannel encoding, predicts microbe-drug associations…
DCGAT-DTI: dynamic cross-graph attention network for drug-target interaction prediction.
DCGAT-DTI, a dynamic cross-graph attention network using pretrained language-model embeddings and GCN-based cross-neighborhood selection, predicts drug-target…
MAGNET: Multi-view graph autoencoder with cell-gene attention for cell interaction network reconstruction from spatial transcriptomics.
MAGNET, a multi-view graph autoencoder with cell-gene attention, reconstructs cell-cell interaction networks from spatial transcriptomics and outperforms TENET…
Preprints
A Multi-scale Fused Graph Neural Network with Inter-view Contrastive Learning for Spatial Transcriptomics Data Clustering
stMFG, a multi-scale fused graph neural network with inter-view contrastive learning, clusters spatial transcriptomics spots into domains, reporting up to 14%…
GlycanGT: A Foundation Model for Glycan Graphs with Pretrained Representation and Generative Learning
GlycanGT is a graph-transformer foundation model for glycans, pretrained with masked language modeling on monosaccharide graphs; it outperforms prior methods…
Topic history
| Date | Rank |
|---|---|
| #18 | |
| #19 | |
| #14 |