Home

What I Build

Protein Structure Networks

Interactive Tools and Visualizations

Protein Structure Networks

Automated pipelines for constructing and analyzing residue-level interaction graphs, centrality metrics, and allosteric pathways.

Graph Neural Networks

Interactive Tools and Visualizations

Protein Structure Networks

GraphSAGE, GAT, and GCN models tailored to protein systems, with interpretability solutions such as Integrated Gradients and saliency.

Interactive Tools and Visualizations

Dash/Flask apps, PyMOL plugins, and visualization workflows for structural data, network topology, and ligand interactions.

Agentic AI Systems

Interactive Tools and Visualizations

Memory-aware agents, computational reasoning tools, and high-performance local ML workflows for scientific discovery.

Featured Projects

PDB2Graph

PSN Analysis of LRH-1 Conformations

A PyMOL plugin for generating protein structure networks (PSNs) from PDB or CIF files. The tool computes edge definitions, node metrics, betweenness centrality, and pathway topologies, producing graph outputs ready for downstream ML workflows or GNN-based analysis.

Learn More

PSN Analysis of LRH-1 Conformations

Built and analyzed protein structure networks across multiple LRH-1 states to identify distinct allosteric communication routes associated with ligand efficacy. Network differences between high and low ΔΔG structures revealed Helix 6 as a key regulatory conduit for LRH-1 activation.

Learn More

GAT Classifier for LRH-1

PSN Analysis of LRH-1 Conformations

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Built an 8-head GAT model that learns discriminative patterns in LRH-1 protein structure graphs, classifying conformations associated with high vs. low free-energy states. Integrates saliency and Integrated Gradients to produce interpretable residue influence maps for understanding structural regulation.

Learn More

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

A project exploring whether Word2Vec-trained residue sequences ordered by pocket geometry can outperform pooled sequence baselines (ESM, ProtBERT) for kinase binding-site classification. A step toward unifying NLP-style token embeddings with graph-based drug discovery workflows.

Learn More

Agentic Graph AI Workstation Build

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Agentic Graph AI Workstation Build

A purpose-built machine featuring dual Xeon processors, 256GB ECC RAM, and an RTX 3090, optimized for GNN training, PSN analytics, and memory-aware agentic systems. This workstation supports low-latency inference, heavy graph workloads, and full local autonomy for scientific ML development.

Learn More

Graph-Based AI for Structural Biology

What I Build

Protein Structure Networks

Interactive Tools and Visualizations

Protein Structure Networks

Graph Neural Networks

Interactive Tools and Visualizations

Protein Structure Networks

Interactive Tools and Visualizations

Interactive Tools and Visualizations

Interactive Tools and Visualizations

Agentic AI Systems

Interactive Tools and Visualizations

Interactive Tools and Visualizations

Featured Projects

PDB2Graph

PSN Analysis of LRH-1 Conformations

PSN Analysis of LRH-1 Conformations

PSN Analysis of LRH-1 Conformations

PSN Analysis of LRH-1 Conformations

PSN Analysis of LRH-1 Conformations

GAT Classifier for LRH-1

PSN Analysis of LRH-1 Conformations

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Agentic Graph AI Workstation Build

Pocket-Ordered Residue Embeddings (Word2Vec for Kinase Pockets)

Agentic Graph AI Workstation Build

This website uses cookies.