01
Molecular Data
Molecular Data
& Curation
Bioactivity data (ChEMBL, etc.), molecular descriptors, fingerprints, scaffold analysis, normalization, and SAR exploration.
COMPUTATIONAL DRUG DISCOVERY • AI/ML • MOLECULAR SCIENCE
Maryam
Taherzadeh
AI/ML Scientist in Computational Drug Discovery
Molecular machine learning, hit generation, generative design, hit-to-lead and lead optimization, ADMET modeling, and structure-based evaluation to accelerate therapeutic discovery.
Drug Discovery Pipeline
End-to-end computational workflow from molecular data to lead optimization and structural validation.
→
02
Predictive Modeling
QSAR/QSPR, potency prediction, ADMET property modeling, molecular property modeling, GNNs, and applicability-domain analysis.
→
03
Hit Generation
Virtual screening, multi-criteria design, reinforcement learning and diffusion models for de novo molecule generation.
→
04
Hit-to-Lead &
Hit-to-Lead &
Lead Optimization
Potency improvement, scaffold exploration, multi-parameter optimization, ADMET trade-offs, developability filtering, and candidate prioritization.
→
05
Structural Evaluation
Docking, protein–ligand interaction analysis, molecular dynamics, MM/GBSA, binding-mode analysis, and candidate ranking.
Featured Research Projects
Selected projects demonstrating applications of AI and computational chemistry across the drug discovery pipeline.
FLAGSHIP
PROJECT 01
MDM2–p53 Inhibitor Design
AI-driven hit generation, hit-to-lead & lead optimization
End-to-end computational drug-discovery workflow integrating curated bioactivity data, QSAR modeling, REINVENT4 molecular generation, ADMET-aware prioritization, docking, molecular dynamics, protein–ligand interaction analysis, and MM/GBSA.
4,146curated compounds
5,734unique molecules
36.77%Gen3 pass rate
0.516 Åredocking RMSD
PROJECT 02
MDM2 Potency Prediction
Scaffold-aware QSAR for pIC50 prediction
Curated ChEMBL bioactivity data and RDKit molecular features used with Extra Trees regression to predict MDM2 inhibitor potency under scaffold-aware evaluation.
R² 0.732scaffold test
RMSE 0.612pIC50
Extra Treesfinal regressor
PROJECT 03
Agentic AI Drug Discovery: MDM2–p53
Interactive pIC50 prediction & candidate prioritization
Deployed decision-support workflow accepting SMILES input, generating RDKit descriptors, predicting pIC50 with XGBoost, assessing activity and Lipinski drug-likeness, and producing a rule-based recommendation.
R² 0.769deployment model
RMSE 0.696pIC50
MAE 0.499pIC50
PROJECT 04
GNN Solubility Prediction
Graph neural networks for aqueous solubility
GCN and GAT models trained on AqSolDB and evaluated under random and Bemis–Murcko scaffold splits, with the scaffold-trained GCN deployed as an interactive Streamlit application.
R² 0.786scaffold test
RMSE 1.095logS
MAE 0.751logS
PROJECT 05
DiffSBDD Molecular Generation
3D structure-based generative modeling
Structure-conditioned molecular generation for the MDM2 binding pocket with validity filtering, ADMET assessment, chemical-space analysis, docking, and protein–ligand interaction analysis.
2,910unique valid molecules · 3,000 requested
About Me
I am a computational scientist with dual Master’s degrees in Computer Science and Physical/Computational Chemistry. My work focuses on machine learning, cheminformatics, generative molecular design, hit-to-lead and lead optimization, ADMET-aware prioritization, and structure-based modeling to accelerate therapeutic discovery.
◈M.S.Computer Science
⚗M.Sc.Physical/Computational Chemistry
▥AI for DrugDiscovery
⌖Based inCalifornia, USA
Core Expertise
View All Skills →- Generative AI for Molecular Design (REINVENT, DiffSBDD)
- Hit Generation, Hit-to-Lead and Lead Optimization
- QSAR/QSPR & ADMET Property Prediction
- Structural Bioinformatics & Protein–Ligand Modeling
- Molecular Docking, PLIP, Molecular Dynamics, MM/GBSA
- Python, RDKit, PyTorch, PyG, scikit-learn, Streamlit
- HPC / Cloud (Docker, SLURM, Kubernetes, AWS, GCP)
Latest Technical Notes
View All Notes →Designing Multi-Objective Rewards for Molecular Generation
Balancing potency, molecular properties, structural support, and developability in generative models.
Read More →Why Applicability Domain Matters in Generative Drug Design
How applicability-domain analysis can reveal when generated chemistry moves beyond model support.
Read More →LET'S COLLABORATE
Let’s accelerate drug discovery together.
I’m open to research collaborations, full-time opportunities, and AI/ML consulting in computational chemistry and drug discovery.