Skip to main content
. 2024 May 15;14:11174. doi: 10.1038/s41598-024-61337-3

Figure 6.

Figure 6

Machine learning predicts compounds with high sensitivity and specificity for generating an AA phenotype. (A) ROC curves show the Support Vector Machine (SVM) results for JNK pathway hub compounds to predict an “AA” or “other” phenotype during the Startle Response Behaviour (SRB) and the Post Startle Response Behaviour (PSRB). The number of fish measurements used for testing the machine learning models were JNK-IN-8: 346, SP600125: 224, haloperidol: 220, SB216763: 358, FR236924: 360, R18: 354, SC79: 318, BIM: 216, PMA: 115, SB590885: 206, Bryostatin-1: 359, AKTi: 360. (B) A table summary of the area under the curve (AUC) values from Random Forest, Glmnet and SVM machine learning models are shown. Bold: Ranked highest overall taken from highest score across all machine learning approaches. The number of zebrafish tracks used for ML model training for the SRB class label “AA” were- diazepam: 350, fluoxetine: 335, imipramine: 343, LiCl: 210, and for class label “other”–ketamine: 349 and MK801: fish. During the post-startle response behaviour (PSRB) were as follows: class label “antidepressant or anxiolytic”–diazepam: 337, fluoxetine: 338, imipramine: 335, LiCl: 210; class label “other”–ketamine: 325 and MK801: 323. (C) A schematic summary of the zebrafish larvae behaviour screen for AA phenotype is shown. Beside it is a summary of results obtained using this test to evaluate the effect of JNK1 pathway hub drugs on AA-like behaviour.