Cleveland Clinic and IBM aren't chasing speedup. They're chasing 150 peptides.

Personalized cancer vaccines live or die on a ranking problem: of the thousands of mutant peptides a single tumor throws off, only a tiny fraction will actually be flagged by the immune system. The hard part is teaching a model to find them.

Cleveland Clinic and IBM's Q-CHIPP paper, published in Science Advances and surfaced through Quantum Computing Report's re-report, treats a regime of roughly 150 labeled peptides per HLA type as the design problem. Two parallel quantum convolutional networks split the task in half. One scores MHC binding. The other scores T-cell recognition, trained only on confirmed binders so binding does not contaminate the immunogenicity signal. A peptide is called a candidate only when both channels agree.

That is the reusable category: when biological data is the binding constraint, the winning move is an architecture that splits a tangled signal into two cleaner subproblems. Q-CHIPP's value is the inductive bias for the small-data regime, not qubit count and not a speedup claim.

The falsifier is a future benchmark where a well-tuned classical baseline on the same roughly 150-sample regime ties or beats the dual-QCNN split. Until that paper lands, Q-CHIPP is a tool being added to the prioritization workflow, not a bedside answer.