Abstract
The detection and identification of intracranial tumours is limited by the lack of accurate biomarkers and requires invasive biopsy procedures. We investigated a machine perception liquid biopsy approach to detect and identify intracranial tumours from peripheral blood and to discover biomarkers responsible for the predictions. Quantum well defect-modified single-walled carbon nanotubes stabilized with single-stranded DNA, interrogating 739 plasma samples from brain tumour patients, were used to train and validate machine-learning models to detect intracranial tumours with 98% accuracy and identify tumour type. The protein corona of the top model-contributing nanosensor was interrogated using quantitative proteomics, resulting in the identification of tumour ecosystem-secreted factors, both previously reported and newly discovered, originating from intracranial tumour cells, the tumour microenvironment and the innate immune system of patients with glioblastoma and meningioma. Newly discovered factors elicited linear nanosensor responses and were elevated in one or both tumour types, matching the original protein corona enrichment. This investigation reveals that a perception-based detection of disease in blood can identify biomarkers responsible for the signal and also amplify cancer detection signals by detecting factors beyond tumour cells, thereby recruiting the entire tumour ecosystem for cancer diagnosis.