Elevating performance and interpretability of in silico classifiers for drug proarrhythmia risk evaluations using...

Area of Science:

• Computational toxicology
• Pharmacology
• Machine learning in drug safety

Background:

• In silico drug assessment systems predict proarrhythmia risk.
• Current leading systems use single biomarkers like qNet, offering interpretability.
• Advanced models use multiple biomarkers for better prediction but lack intuition.

Purpose of the Study:

• To develop a method for proarrhythmia risk assessment using multiple biomarkers.
• To maintain interpretability while enhancing predictive capabilities.
• To provide practical alternatives for drug safety evaluations.

Main Methods:

• Enhanced ordinal logistic regression (OLR) with additional physiological biomarkers.
• Introduced a general torsade metric score (TMS) for multi-biomarker interpretability.
• Employed a multi-criteria decision analysis ranking algorithm for classifier evaluation.

Main Results:

• Multi-biomarker approaches show superior performance compared to single qNet.
• Some OLR models without qNet outperform those including it.
• Individual poorly performing biomarkers can improve in combination.

Conclusions:

• The proposed method effectively utilizes multiple biomarkers for proarrhythmia risk assessment.
• The TMS provides straightforward interpretability for multi-biomarker OLR models.
• Offers practical and interpretable alternatives for drug safety evaluations.