Evaluation of\nMachine Learning Based QSAR Models for\nthe Classification of Lung Surfactant Inhibitors

Inhaled\nchemicals can cause dysfunction in the lung surfactant,\na protein–lipid complex with critical biophysical and biochemical\nfunctions. This inhibition has many structure-related and dose-dependent\nmechanisms, making hazard identification challenging. We developed\nquantitative structure–activity relationships for predicting\nlung surfactant inhibition using machine learning. Logistic regression,\nsupport vector machines, random forest, gradient-boosted trees, prior-data-fitted\nnetworks, and multilayer perceptron were evaluated as methods. Multilayer\nperceptron had the strongest performance with 96% accuracy and an\nF1 score of 0.97. Support vector machines and logistic regression\nalso performed well with lower computation costs. This serves as a\nproof-of-concept for efficient hazard screening in the emerging area\nof lung surfactant inhibition.