Figure 5.
Comparing and selecting machine learning models and the result of a gene-level prediction. (A) The measurement of feature codependency in model training. We tested the error rate for models trained with all selected features (Table 1) as well as by removing one feature at a time (see Methods). (B) The frequency distribution of the gene-level AAMR risk scores for 12,074 OMIM genes. (C) The frequency distribution of the gene-level AAMR risk scores for 133 genes that have been involved in AAMR more than once.
