Fig. 5

MODICT scores of ACADM mutations. a. Mutation pairs were plotted based on their enzymatic activity and the average of their MODICT scores. MODICT scores or residual activities that are 2 standard deviations away from the data average was excluded which corresponded to exclusion of only 1 data point (residual activity 60, modict score 53.5). The remaining data points had a correlation coefficient of –0.488 with a p-value of 0.044 according to 1 tailed t-distribution. b. Same mutations were plotted with POLYPHEN2 scores instead which yielded a positive correlation coefficient of 0.211 with p-value of 0.244. c. Eight out of 14 mutation pairs in table 2 harbored a p.K329E variant where homozygotes for this mutation only had 5 percent of wildtype activity. Assuming significant portion of residual activity coming from the other variants, these 8 variants (lower left) were used as a training dataset for MODICT. After training, MODICT was able to find a weight score combination with a correlation coefficient of –0.959 (lower mid). Using the trendline obtained by least squares method, the residual activity of 6 other mutation pairs (that did not include the trained mutations) were guessed. MODICT was able to achieve 91 percent accuracy (lower right). (∗∗=p<0.05;∗∗∗=p<0.001)