Figure 8

Performance of the machine learning method trained on different sized sets of data from SAAPdb. In each case, a balanced dataset of the required size was extracted at random from the SAAPdb dataset of mutations mapped to protein chains (Table 2) and random forests were trained and tested using 10-fold cross-validation. The graph clearly shows that performance drops as the dataset size decreases, showing a marked drop in performance with datasets below 10,000 samples in size (5,000 SNPs and 5,000 PDs).