Data by OrandiAILLMsdrug discoveryquantum-AI modelsmolecular property prediction
From Simulations to Patients: How Hybrid Quantum-AI Models Are Accelerating Drug Discovery in Africa
8 min read · 1,453 wordsBy Orandi Felix
On the cluster A100 the training loop runs in ~3 min for 200 epochs; on a local RTX 3060 (12 GB) it takes ~7 min. The quantum latency is dominated by IBM API overhead (2–3 s per molecule), so we cache the expectations and only recompute if descriptor space drifts.
The quantum hardware RMSE increases because NISQ devices introduce shot noise and gate errors. Still, even the noisiest runs beat classical-only baselines—proof that classical nets can “learn around” quantum imperfections.
IBM Quantum Experience credits are free for African universities; apply through their Qiskit Global Summer School. Google Quantum AI gives free quota on cirq backend; the pennylane wrapper above works with both.