Leveraging developmental neurotoxicity data in zebrafish embryos through the use of artificial intelligence methods

Area of Science:

• Developmental toxicology
• Neuroscience
• Computational biology

Background:

• Zebrafish are a key model for studying chemical developmental neurotoxicity (DNT).
• The zebrafish light-dark transition test (LDT) is an emerging assay for DNT screening, but requires refined analytical tools.
• Artificial intelligence (AI) offers potential for advanced analysis of LDT data.

Purpose of the Study:

• To evaluate the efficacy of AI methods in detecting DNT effects in zebrafish exposed to various pharmaceuticals.
• To compare the performance of different AI algorithms (GLM, RF, GBM, DL) in analyzing LDT data.
• To assess the sensitivity of AI in identifying DNT compared to conventional methods.

Main Methods:

• Zebrafish were exposed to vehicle controls or pharmaceuticals (fluoxetine, paroxetine, carbamazepine, phenytoin) at 5, 10, or 14 days post-fertilization (dpf).
• Four AI methods (GLM, RF, GBM, DL) were trained to distinguish between control and exposed zebrafish based on locomotory behavior in the LDT.
• A five-fold cross-validation approach was used to assess prediction accuracy.

Main Results:

• AI prediction accuracy for DNT detection increased from 67% at 5 dpf to 76% at 10 and 14 dpf.
• Antidepressants (fluoxetine, paroxetine) showed persistent DNT effects detectable by AI even after shorter exposure durations.
• AI methods, particularly GLM, RF, and DL, demonstrated comparable and high performance, outperforming GBM.

Conclusions:

• AI methods show enhanced sensitivity for detecting DNT effects in zebrafish compared to univariate analysis.
• The study highlights the potential of AI for robust DNT screening and the refinement of the LDT assay.
• AI implementation can advance the assessment of chemical safety during development.