Table 4.
cQNN | cQNN | cQNN | Weighted mQNN | Weighted mQNN | Weighted mQNN | ||||
---|---|---|---|---|---|---|---|---|---|
One-layer | Two-layer | Six-layer, 0.15 margin | Six-layer, 0.15 margin | Six-layer, 0.15 margin | |||||
Five-qubit IBM | Five-qubit IBM | PennyLane | DM1 | Rigetti | |||||
Position | Ref | Alt | Ligand | Simulation | 1000 shots | 1000 shots | Simulation | 1000 shots | 1000 shots |
160 | CYS | PHE | X77 | 0 | 0 | 0 | 1 | 1 | 0 |
168 | PRO | SER | X77 | 1 | 1 | 1 | 1 | 1 | 0 |
188 | ARG | SER | X77 | 1 | 1 | 1 | 1 | 1 | 0 |
160 | CYS | PHE | ZINC000016020583 | 0 | 0 | 0 | 1 | 1 | 1 |
168 | PRO | SER | ZINC000016020583 | 1 | 1 | 1 | 1 | 1 | 0 |
188 | ARG | SER | ZINC000016020583 | 1 | 1 | 1 | 1 | 1 | 0 |
160 | CYS | PHE | ZINC000036707984 | 0 | 0 | 0 | 0 | 0 | 0 |
168 | PRO | SER | ZINC000036707984 | 1 | 1 | 1 | 1 | 1 | 0 |
188 | ARG | SER | ZINC000036707984 | 1 | 1 | 1 | 1 | 1 | 0 |
Note: Predictions are indicated as 0 or 1 depending on whether a mutation is predicted to be non-disruptive or disruptive to binding of the ligand, respectively. The results shown include: a three-layer (FN architecture) cQNN circuit where both training and predictions were run using the QASM simulator; and one- and two-layer cQNN circuits trained on the QASM simulator, but with prediction runs on an IBM Quantum five-qubit system (marked by the term ‘real’), with 1000 shots in each case; a weighted mQNN circuit with predictions conducted through simulation (default Pennylane simulator with six-layer and 0.15 margin); a six-layer, 0.15 margin, weighted mQNN circuit with predictions conducted through runs on an AWS’s Braket DM1 system, 1000 shots; a six-layer, 0.15 margin, weighted mQNN circuit with predictions conducted through runs on a Rigetti system, 1000 shots.