a computational framework to identify and characterize cell niches from spatial omics data at single-cell resolution

A comprehensive (masked) graph autoencoders benchmark.

This project detects structural network anomalies using a GNN autoencoder. It contrasts this deep learning approach with the classic DBSCAN method. While DBSCAN only uses node features (CPU, RAM), the GNN learns the graph's topology to identify statistically improbable links, proving superior for structural analysis.

This project models the GitHub collaboration network to predict potential partnerships. It applies Social Network Analysis (SNA) and a Graph Autoencoder (GAE) to tackle the link prediction task, identifying future collaborators with high accuracy.

Open-source pipeline for neural decoding via structure-from-function inference on connectome data. Phases: C. elegans validation → pre-registered FlyWire→MICrONS cross-species transfer test → disentanglement extension → parallel sleep-EEG track. Six-month independent solo research.

A comprehensive bridge impact analysis system combining heterogeneous graph neural networks (HGNN) for closure-impact prediction with graph autoencoders (GAE/VGAE/HetVGAE) for unsupervised bridge similarity learning. This extends the system with heterogeneous graph variational autoencoders for metapath-based bridge classification.

This repository contains the Hopf-GAE, a physics-informed deep learning architecture that detects depression-related dynamical abnormalities without ever training on depressed brains.