Quantity
Quality quantum chemistry yields ΔG‡ directly comparable to experiment — not a heuristic score.
Flagship · Quantum-classical platform
Covalent drug discovery, decided at quantum accuracy.
Sub-kcal/mol activation energies for covalent inhibitors, computed from first principles — quantities that map one-to-one to reaction rates measured in the lab.
One hybrid pipeline orchestrating MD, QM/MM and a quantum core, executing across CPU, GPU and QPU through CUDA-Q.
Covalent
QM/MM
ECC-DMET
CUDA-Q
N 50°01′47″ · E 19°56′25″
BEIT / 002
< 1 kcal/mol
Target accuracy
~30%
of FDA drugs are covalent
CPU · GPU · QPU
Hybrid execution
CUDA-Q
Native integration
At a glance
The CovAngelo story, in three frames.
Engine
Quantum solvers, HPC, AI/ML and data management orchestrated through one unified engine.
Impact
Higher hit rates across the discovery pipeline — powering next-gen ML scoring.
01 / The challenge
Covalent docking has been stuck between inaccurate and intractable.
01
Low-level QM can't be trusted
DFT and semi-empirical methods (PM6) are inaccurate and case-specific — false positives and false negatives propagate straight through the CADD pipeline.
02
High-level QM is out of reach
Coupled-cluster and FCI deliver the accuracy, but cost and expert dependence make them intractable for routine virtual screening.
03
The goal
Automation, error control, and sub-kcal/mol accuracy on activation energies — delivered as a turnkey ranking, not a research project.
02 / Why CovAngelo
First-principles quantum chemistry, productionised.
Current covalent docking tools rely on heuristic scoring. CovAngelo computes activation energies directly — quantities that map one-to-one to reaction rates measured in the lab.
Quantity, not score
Activation energies map one-to-one to reaction rates measured in the lab.
Transferable
First-principles scoring generalises across protein families and chemotypes.
Integrated stack
MD, QM/MM, ECC-DMET and quantum core orchestrated through a single pipeline.
Hybrid execution
CUDA-Q unifies CPU, GPU and QPU backends — no rewrite to change hardware.
03 / How it works
From receptor and ligands to a ranked, rate-aware shortlist.
Step 1
Molecular Dynamics Solver
Phase-space sampling of protein-ligand conformations.
Step 2
Quantum Chemistry Solver (QM/MM)
Electronic structure with electrostatic embedding.
Step 3
Quantum-Information Optimizer
Orbital entanglement optimisation — fewer qubits, same answer.
Step 4
Quantum Core Solver (QM/QM)
ECC-DMET with FCI, CCSD(T) and VQE on the bond-forming core.
Step 5
CPU / GPU / QPU backend
Unified execution via CUDA-Q across NVIDIA, IonQ, IBM and IQM.
Input
Receptor (PDB) · ligand dataset (mol2)
Output
Ranked ligands · reaction energy barrier · rate constants · molecular features
04 / The science
A QM/QM/MM model built on BEIT's Entanglement-Consistent DMET.
Classical molecular dynamics samples phase space across the full protein. A low-cost quantum-mechanical region (Hartree-Fock / DFT) handles the broader chemistry, while the bond-forming core is solved at high accuracy with FCI, CCSD(T) and VQE on top of our Entanglement-Consistent DMET — including explicit water networks stabilizing transition states.
US Patent App. #64 026,210
05 / Proof of concept
Zanubrutinib · covalent BTK inhibitor, end-to-end.
CovAngelo reproduces the full reaction profile of zanubrutinib binding to CYS481 — Michael addition - and recovers experimentally measured rate constants via the Eyring equation.
k(T) = κ · (kBT / h) · exp(−ΔG‡ / RT)
Hardware
IQM Garnet
Ansatz
UCCSD · 26 variational parameters (VQE)
Validated
CPU · GPU · QPU
Backends
NVIDIACUDA-QIonQIBMIQMAWS
Resources
Read the research behind CovAngelo.
Bring quantum accuracy to your covalent inhibitors.
We work with discovery teams on lead optimization, hit triage, and difficult covalent targets. Tell us about yours.