Fig. 5

Data clustering and sample identification. On the basis of PCA analysis, we selected the best combinations of voltage, sensor position, modulation and time constant that optimize the response of the device and maximize separation of the sample of secretome made by cultures of non-haematological cells isolated by liquid biopsy of control and cancer patients, between variables (a–f). For all considered configurations, data were clustered into control and patient groups with a specificity up to 100% with the exception of configuration (c), where matching the efficiency of classification is nearly 93%