01BioEmu
Target dynamics
AI-estimated conformational ensembles of protein targets, cryptic pockets and druggability per state.
AI Research · Solution Architecture · Computational Chemistry
David Malick Dieng
I architect AI systems that compress the path from protein target to drug candidate — from conformational dynamics and core hopping to ADMET and formulation.
Research focus
Disrupting how molecules are discovered.
My work sits at the intersection of machine learning, structural biology and software architecture. I design end-to-end platforms where generative models, physics-based docking and trained ADMET predictors work as one coherent, auditable system.
A central theme is computational core hopping: replacing a molecule's scaffold while preserving the interactions that make it active — opening new chemical space, improving safety profiles and creating freedom to operate.
Flagship platform
CureIQ Labs
A multi-agent AI drug discovery workspace I designed and built — bilingual (EN/FR), mission-driven, from target to pre-formulation.
02TamGen
Molecule generation
Target-aware generative chemistry producing candidate ligands conditioned on the binding site.
03Standalone engine
Core hopping
Scaffold replacement that preserves pharmacophore geometry while escaping crowded IP and liability-bearing cores.
04AutoDock Vina
Docking & 3D
Pose prediction and interaction analysis with an interactive structural viewer.
05ADMET-AI · Chemprop
ADMET & formulation
22 ADMET endpoints, physico-chemical rules, applicability domain and a pre-formulation engine (BCS, ESOL).
06Agent workspace
Multi-agent orchestration
Missions run by cooperating AI agents that plan, execute and document each discovery step.
22
ADMET endpoints
18
trained model outputs
90
reference drug decisions
2
languages
Inside the platform
AI-estimated conformational states of aspirin-acetylated COX-2 (PDB 5F19) with a 3D structural viewer.
Generated candidate with SMILES, QED, synthetic accessibility and physico-chemical descriptors.
Candidate comparison with Vina scores, ADMET pre-evaluation and space-filling receptor view.
Method
Scientific rigor, engineered at scale.
Calibrated, not cosmetic
Every model is benchmarked against reference drugs — hERG, DILI, Ames, clinical toxicity — and its reliability is shown, not hidden.
Domain-aware
Predictions declare when a molecule is out of the model's applicability domain, and convert uncertainty into concrete validation assays.
Architected for labs
Reproducible sessions, versioned runs, cached results, CSV and PDF decision dossiers that a medicinal chemist can defend.