Quantum drug-design preprint reports comparable performance with 110 generator parameters

Researchers report in a preprint that a quantum generative model for molecular design achieved statistically comparable overall performance to a classical baseline with about 700,000 generator parameters, using 110 trainable generator parameters. It also showed statistically significant gains on selected drug-relevant metrics against that baseline and a classical model with matched generator capacity. The system maps molecular structures into a compressed representation and uses a hybrid quantum generator to produce new examples. Training and large-scale sampling used noiseless quantum simulators with circuits of up to 15 qubits. A separate inference-only demonstration on an IBM Heron quantum computer used 5 qubits to sample 2,500 new molecules and investigate noise effects.