Skip to content
View abinittio's full-sized avatar

Highlights

  • Pro

Block or report abinittio

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
abinittio/README.md

Typing SVG


Python PyTorch TypeScript RDKit Next.js


What I work on

I develop graph neural networks and numerical simulation tools that encode real chemistry — stereochemistry, quantum descriptors, pharmacokinetics — into predictive models. My focus is bridging the gap between physical chemistry and modern ML, where most tools ignore 3D molecular structure entirely.


Drug discovery & ADMET

Full ADMET prediction platform — monoamine transporter substrate vs blocker classification with stereochemistry-aware GNN.

0.974 AUC · GAT + GIN · Deployed on HuggingFace Spaces

Drug abuse risk classification combining MAT activity with SMARTS-based structural pattern recognition and SAR pharmacology rules.

Validated on 80 DEA-scheduled compounds · HIGH/MODERATE/LOW risk tiers

Cardiac safety prediction via hERG channel inhibition modelling. Focal loss for class imbalance, K-fold ensembling with test-time augmentation.

Trained on 7,000+ ChEMBL compounds

Multi-task GNN for drug-drug interaction screening across 5 cytochrome P450 enzymes (1A2, 2C9, 2C19, 2D6, 3A4).

Shared representation · Task-specific heads

BBB permeability & quantum descriptors

Blood-Brain Barrier permeability predictor — hybrid GAT → GCN → GraphSAGE with focal loss and stereo-aware encoding.

0.9612 AUC on 7,807-compound external validation · Outperforms ADMETlab 2.0

Quantum-enhanced GNN with 34-dimensional features from 3D conformers — HOMO/LUMO, Fukui indices, chemical hardness.

GATv2 + TransformerConv · Pretrained on 320k ZINC molecules

Simulation & other

PK/PD simulation engine solving coupled ODEs for prodrug systems with saturable enzymatic conversion.

RK4 · Michaelis-Menten · Sigmoid Emax · Bayesian personalisation

Precision agriculture system — soil analysis, Monte Carlo yield prediction, and fertiliser recommendations.

244,000+ soil samples · Real-time market data · Uncertainty quantification

Technical stack

Molecular ML        PyTorch · PyTorch Geometric · RDKit · DGL
Quantum descriptors HOMO/LUMO · Fukui indices · Gasteiger charges · ETKDG conformers
Numerical methods   RK4 · Michaelis-Menten · Hill equation · Sigmoid Emax
Languages           Python · TypeScript
Frameworks          Streamlit · Gradio · Next.js · React
Deployment          HuggingFace Spaces · Vercel

Popular repositories Loading

  1. BBB-Permeability-GNN BBB-Permeability-GNN Public

    Interpretable GNN for Blood-Brain Barrier prediction with molecular generation and structure-activity analysis.

    Python

  2. StereoAwareGNN StereoAwareGNN Public

    Blood-Brain Barrier permeability predictor using hybrid GAT-GCN-GraphSAGE architecture. 0.9612 AUC on 7,807-compound external validation.

    Python

  3. SoilIntelligence SoilIntelligence Public

    Predicts quality of soil for any coordinate in Africa and recommends fertiliser to maximise yield

    Python

  4. Insilico-Drug-Discovery-Toolkit Insilico-Drug-Discovery-Toolkit Public

    Comprehensive ADMET prediction platform: monoamine transporter classification (0.974 AUC), abuse liability, hERG cardiotoxicity, CYP450 metabolism.

    Python

  5. abinittio abinittio Public

  6. DoseTrack DoseTrack Public

    Pharmacokinetic/Pharmacodynamic simulation engine with RK4 ODE solver, Michaelis-Menten kinetics, and adaptive personalisation. Built with Next.js and TypeScript.

    TypeScript