Fig. 6

A shallow neural network-based classification of synthetic and real datasets with 37 and 5 protein candidates. a The dissimilarity matrix (top left corner) and multi-dimensional scaling (MDS) scatter plot for the triplicates of pooled CRC plasma samples (e.g., healthy control and stages I–IV). b The dissimilarity matrix and MDS plot of a synthetic dataset of a panel of 37 protein candidates. A total of 5000 synthetic patients (1000 per healthy control and the 4 CRC stages) were created from random numbers falling within a normal distribution of 10 times the standard deviation (SD) of the pooled real CRC plasma samples. c Confusion matrix of the synthetic dataset (for 37 protein candidates) for the test phase of the training of the classifier achieved 99.6% success. d Confusion matrix for the testing of the classifier on the real dataset kept out of training achieved 80% correct classification. e Dissimilarity matrix and MDS plot of the synthetic dataset for a panel of 5 protein candidates (SAA2, APCS, APOA4, F2 and AMB) with a total of 5000 synthetic patients. f Confusion matrix of the synthetic dataset (for 5 protein candidates) for the test phase of the training of the classifier achieved 94% success. g Confusion matrix for the testing of the classifier on the real dataset kept out of training achieved 100% correct classification