Improving In Silico Cardiac Safety Prediction by Consensus Averaging of Transmural Ventricular Cell Models.

Area of Science:

• Computational electrophysiology
• Cardiac safety pharmacology
• In silico modeling

Background:

• Established in silico Torsade de Pointes (TdP) risk assessment models rely on single-cell biomarkers.
• Ventricular transmural heterogeneity influences cardiac repolarization and arrhythmogenic mechanisms.
• Direct integration of multi-cell data can lead to multicollinearity and model instability.

Purpose of the Study:

• To develop a robust ordinal logistic regression (OLR) framework for TdP risk assessment.
• To integrate multi-cell electrophysiological information (endocardium, epicardium, mid-myocardium) while preserving physiological context.
• To mitigate multicollinearity and enhance statistical stability in predictive models.

Main Methods:

• Implemented drug data (IC50, Hill coefficients) in CiPAORdV1.0 and ORd ventricular cell models.
• Computed quantitative network (qNet) biomarkers independently for endocardial, epicardial, and mid-myocardial cells.
• Averaged cell-specific OLR model probabilities for final TdP risk prediction, comparing against single-cell and direct multi-cell approaches.

Main Results:

• The qNet approach achieved high performance (AUCs of 1.000 and 0.958) in the CiPA-ORd v1.0 model using the ChanTest dataset.
• Probability averaging consistently improved performance across Manual and ChanTest datasets in the ORd model.
• The proposed method met seven "excellent" classification criteria for TdP risk.

Conclusions:

• Probability-averaged integration of multi-cell qNet predictions effectively mitigates multicollinearity.
• This approach preserves physiological relevance, leading to more stable and accurate in silico TdP risk classification.
• The framework supports broader applicability in preclinical cardiac safety assessment.