Benchmark Finds No Quantum Advantage on Tabular, Omics, and Spatial Cancer Datasets

The study evaluated quantum and classical models on tabular, omics, and spatial cancer data drawn from prior quantum machine learning literature. Researchers applied consistent preprocessing pipelines and resource estimation to test for advantage under controlled conditions. Classical neural networks optimized via AutoML matched or exceeded quantum performance on every task examined.

Prior claims of quantum superiority in oncology often relied on small or synthetic datasets without matched classical baselines. This work highlights how high-dimensional biological noise and limited sample sizes currently mask any potential quantum kernel benefits. It also notes that current NISQ hardware constraints limit expressivity far below the requirements of realistic multi-omics interactions.

Future progress requires datasets with hundreds of biologically grounded features and sample sizes above several hundred patients. Only then can properly powered trials distinguish genuine quantum kernels from classical feature maps. Without such data, incremental hardware improvements alone will not demonstrate clinical utility.

The authors recommend shifting community focus from toy benchmarks toward federated, multi-center oncological repositories that preserve spatial and longitudinal structure.