Skip to main content
. 2021 May 25;17(5):e10267. doi: 10.15252/msb.202110267

Figure 3. Classifiers trained on behavioral fingerprints can predict the mode of action of unseen test compounds.

Figure 3

  1. Toy data illustrating the potential benefit of normalization in correcting for potency differences within mode‐of‐action classes. Following normalization, each behavioral fingerprint exists on a hypersphere in the phenotype space regardless of effect size in the original space. Nonlinear dose–response curves will not collapse perfectly following normalization, which is a linear transformation.
  2. The confusion matrix obtained through cross‐validation for the best performing feature set (1,024 features) and logistic regression classifier following feature selection and hyperparameter tuning on the training data.
  3. The confusion matrix for the classifier trained in (B) applied to previously unseen test compounds without any further tuning.
  4. The novelty score assigned to novel test compounds with a mode of action not seen during training compared to the novelty score of compounds from the test set in (C). Novel compounds tend to have higher novelty scores than compounds from previously seen modes of action. The non‐novel compounds with high novelty scores include the two incorrectly classified test compounds (in red box).