News

Qubit Pharmaceuticals, Sorbonne Université, and Q-CTRL Demonstrate 3D Protein Hydration-Site Prediction on IBM Heron Processors

Researchers from Qubit Pharmaceuticals, Sorbonne Université (CNRS Laboratoire de Chimie Théorique), and Q-CTRL have published a peer-reviewed study in Physical Review Research demonstrating three-dimensional protein-pocket hydration-site prediction on 156-qubit IBM Heron quantum processors. By fitting continuous 3D Reference Interaction Site Model (3D-RISM) water densities to a Gaussian Mixture Model (GMM) cast as a Quadratic Unconstrained Binary Optimization (QUBO) problem, the team executed hardware experiments up to 123 qubits across five FDA-approved drug-target protein structures using Q-CTRL’s Fire Opal optimization solver.

3D Protein Hydration-Site Quantum Optimization & Resource Scaling Summary
Benchmark / RegimeScale & Hardware / SolverKey Findings & Resource Metrics
11-Instance Protein Benchmark Set41 to 116 qubits across 5 PDB targets (1f9g, 1×70, 2f9w, 3b7e, 4h2f/3beq) on IBM Kingston (Heron r2)Q-CTRL Fire Opal identified exact CPLEX-verified optimal water placements across all 11 instances, reaching ~9% optimal sampling probability at 116 qubits (vs. ~2% for simulated annealing).
123-Qubit Hard QUBO Benchmark123 qubits (PDB 3b7e with ligand at 1.9 Å grid spacing) on IBM Pittsburgh (Heron r3)Found a lower-cost solution (C = −321.32) in 25 minutes on QPU than CPLEX reached before timing out at 3 hours (C = −319.90, 40% optimality gap).
Classical Scale-Up & Comparison900 to 6,824 binary variables via simulated annealing vs. Hydraprot, Watgen, Placevent, and Dowser++Crystal-water recovery rose from ~60% at 123 variables to ~90% at ≥1,450 variables, matching Hydraprot precision and outperforming Placevent, Watgen, and Dowser++.
Quantum Utility ForecastQuadratic two-qubit gate scaling extrapolated from 156-qubit Heron compilationsProjects ~900 qubits and ~100,000 two-qubit gates (reducible to ~15,000 with higher connectivity) for practical drug-discovery utility.

Across 11 discretized protein-ligand test instances ranging from 41 to 116 variables on the IBM Kingston Heron r2 processor, Q-CTRL’s automated error-suppressed variational solver recovered the exact classically verified ground-state water configurations in every case, achieving roughly 9% sampling probability for the optimal solution on the 116-qubit instance compared with 2% for classical simulated annealing. Scaling the 3b7e protein-ligand complex to 123 qubits on IBM’s newer Heron r3 processor (IBM Pittsburgh) yielded a lower-energy solution in 25 minutes of QPU runtime than the exact classical solver CPLEX achieved before hitting a three-hour timeout.

To chart the path toward practical computer-aided drug discovery (CADD) utility, the authors evaluated larger problem grids classically up to 6,824 variables and extrapolated circuit compilation requirements from Heron hardware. Their analysis indicates that hydration-site prediction becomes competitive with leading classical tools at roughly 900 qubits and 100,000 two-qubit gates, placing full-pocket water placement for molecular docking and lead optimization within reach of early error-corrected quantum processors.

Read the open-access paper in Physical Review Research here.

In GQI Portal

The players behind the news

The team that writes QCR tracks every company, deal and technology in the GQI Factory, GQI's verified database of the quantum industry. Next up: every story linked to its players, coming to QCR's paid plans.

Leave a comment

All fields are required. Your email address will not be published.