Conformal prediction to define applicability domain - A case study on predicting ER and AR binding.

Area of Science:

• Computational toxicology
• cheminformatics
• Endocrine disruption research

Background:

• Robust in silico models require accurate statistical quality and predictive boundary estimation.
• Applicability domain estimation is vital for reliable computational toxicology predictions.

Purpose of the Study:

• To introduce and evaluate conformal prediction for applicability domain estimation in endocrine disruptor research.
• To assess the method's performance with oestrogen and androgen receptor binders and non-binders.

Main Methods:

• Application of conformal prediction to ensembles of decision trees.
• Investigation of dragon, RDKit, and signature fingerprints as chemical descriptors.
• Validation using internal training and external test sets.

Main Results:

• Conformal prediction yielded valid models with balanced quality, sensitivity, and specificity.
• The method allows user-adjustable confidence levels with immediate inspection of consequences.
• Predictive boundaries were rigorously defined, eliminating ambiguity for new compound assessments.

Conclusions:

• Conformal prediction provides a robust framework for applicability domain estimation in endocrine disruptor modeling.
• The method enhances the reliability and interpretability of in silico predictions.
• This approach ensures clear definition of model applicability for new chemical entities.