Development of classification models for predicting inhibition of mitochondrial fusion and fission using machine...

Area of Science:

• Cell Biology
• Toxicology
• Computational Chemistry

Background:

• Mitochondrial fusion and fission are vital for cellular health, especially under stress.
• Disruptions in these processes are linked to neurodegenerative disorders.
• Identifying chemicals that interfere with mitochondrial dynamics is crucial but challenging experimentally.

Purpose of the Study:

• To develop and validate Quantitative Structure-Activity Relationship (QSAR) models for predicting chemical inhibition of mitochondrial fusion and fission.
• To identify structural alerts associated with the disruption of mitochondrial dynamics.

Main Methods:

• Machine learning algorithms including random forest, logistic regression, Bernoulli naive Bayes, and deep neural network were employed.
• Models were rigorously evaluated using 100 iterations of five-fold cross-validation and external validation.
• Structural alerts for fusion and fission inhibition were identified.

Main Results:

• The best QSAR model for mitochondrial fusion achieved an AUC of 82.8% (cross-validation) and 78.1% (external validation).
• The optimal QSAR model for mitochondrial fission demonstrated an AUC of 84.3% (cross-validation) and 97.5% (external validation).
• A total of 45 and 56 structural alerts were identified for mitochondrial fusion and fission inhibition, respectively.

Conclusions:

• Developed QSAR models effectively differentiate chemicals that inhibit mitochondrial fusion and fission.
• The identified structural alerts provide valuable insights into the mechanisms of chemical disruption.
• These computational tools offer a practical and efficient approach for screening chemical impacts on mitochondrial dynamics, aiding in the prevention of related adverse health effects.