Fig 1.
Frequency of appearance of amino acids in beta-lactamase dataset.
The amino acid distribution is plotted for the functionally neutral/beneficial mutations (left) and the functionally disrupting mutations (right). The blue bars in the plot show the number of mutations from that wildtype amino acid. And the orange bars show the number of mutations to that amino acid. Functionally disrupting mutations that occur a statistical significant number of times (p < 0.01) are show in bold, with an asterisk under the amino acid.
Fig 2.
ROC curves for classifying functionally disrupting mutations on the beta lactamase dataset.
This dataset is composed of 990 single-point mutations with measured changes in their minimum inhibitory concentration (MIC). The five models include energy scores from FoldX, the solvent accessibility of a side chain, the score from the BLOSUM62 substitution matrix, the Prime Stability score, and the Prime stability score with a 5Å minimization cutoff. The area under the curve (AUC) for each ROC curve is shown on the right along with its 95% confidence interval (CI).
Fig 3.
ROC curves for individual Prime terms on the beta lactamase dataset.
The ROC curve for the Prime Stability energy function (black) and individual terms of the Prime Stability energy function (gray). The table on the right list the corresponding area under the curve (AUC) values and the 95% confidence intervals for the AUC values.
Fig 4.
ROC curves separated by mutations that change in charge from mutations that conserve charge.
(A) The charge-change dataset consisted of 369 mutations (B) The conserve-charge dataset consists of 623 mutations that conserve the net charge of the protein. The tables on the right list the area under the curve (AUC) values along with 95% confidence interval (CI) for the AUC values.
Fig 5.
ROC curves separated by mutations that decrease in volume from mutations that increase in volume.
(A) The big->small dataset consists of 126 mutations that decrease the net volume of the side chain (see Methods). (B) The small->big dataset consists of 141 mutations that increase the volume of the mutated side chain (see Methods). The tables on the right list the area under the curve (AUC) values along with corresponding 95% confidence interval (CI) for the AUCs.
Fig 6.
ROC curves separated by mutations that are buried in the protein from solvent exposed mutations.
(A) The buried dataset consists of 292 mutations that are buried in the core of beta lactamase (< = 5% solvent exposed). (B) The solvent exposed dataset consists of 367 mutations that are solvent exposed (>20% solvent exposed). The tables on the right list the area under the curve (AUC) values along with corresponding 95% confidence interval (CI) for the AUCs.
Fig 7.
Two confusion matrices classifying functionally disrupting mutations on the surface using the Prime stability score.
The left matrix uses a Prime energy cutoff of 10 prime energy units to classify functionally disrupting mutations from neutral/beneficial mutations. The matrix on the right uses a Prime energy cutoff of 20 prime energy units. The precision, sensitivity, and specificity are shown below the matrices.
Fig 8.
ROC curves using 495 mutations randomly sampled from the beta lactamase dataset.
Prime Max refers to the maximum prime energy between the predicted change in stability and predicted change in affinity. The machine learning (ML) model refers to a single layer neural network trained on the other 495 mutations not included in this dataset. The area under the curve (AUC) values along with the corresponding 95% confidence intervals are shown in the table on the right.