|
|
|
Figure 7
Normalized confusion matrices comparing the classification performance of the MLE-Matthews model against the Matthews model on test data. The heat maps show per-class prediction distributions for 12 asymmetric unit categories (1–10, 12 and X). Rows represent the true classes and columns the predicted classes, with values normalized by row to highlight recall performance. Colour intensity corresponds to the percentage of predictions within each true class, with darker shades indicating higher proportions. (a) The Matthews model displays widespread off-diagonal elements, indicating substantial misclassification, particularly as the number of copies increases. (b) The MLE-Matthews Control model also displays widespread off-diagonal elements, indicating substantial misclassification. (c) The MLE-Matthews model has strong diagonal dominance, reflecting the accurate and specific classification of structural classes. |
Open
access
access
journal menu![[Figure 7]](has5001fig7.jpg)



