Machine learning to predict developmental neurotoxicity with high-throughput data from 2D bio-engineered tissues.

Area of Science:

• Toxicology
• Neuroscience
• Biotechnology

Background:

• Current developmental neurotoxicity testing methods (animal studies, cell cultures) are time-consuming, costly, and lack human physiological relevance.
• Machine learning models using gene expression data from 3D human neural tissue models show promise but require significant expertise.
• A simpler, more accessible assay is needed for efficient neurotoxicity screening.

Purpose of the Study:

• To compare the predictive accuracy of machine learning models for developmental neurotoxicity using 2D versus 3D human neural tissue models.
• To evaluate the robustness of these models under varying gene set selection criteria.
• To determine if a simplified 2D model can provide sufficient accuracy for prioritizing chemical safety.

Main Methods:

• Gene expression data was collected from human 2D and 3D neural tissue models exposed to various chemical compounds.
• Machine learning models were trained using this gene expression data to predict developmental neurotoxicity.
• Model performance was compared between the 2D and 3D tissue models, with varying gene set selection stringency.

Main Results:

• The predictive models trained on data from the 2D tissue model demonstrated substantially higher accuracy compared to those trained on the 3D model.
• The 2D model approach remained robust even with stringent gene set selection.
• The 3D model approach showed significant accuracy degradation under stringent gene set selection.

Conclusions:

• A 2D human neural tissue model provides a more accurate and robust platform for predicting developmental neurotoxicity than a 3D model.
• The proposed 2D assay offers a valuable, efficient tool for decision-makers in prioritizing chemical neurotoxicity screening.
• This simplified approach addresses the need for faster, cost-effective, and human-relevant developmental neurotoxicity testing.