Uncertainty assessment of proarrhythmia predictions derived from multi-level in silico models

Figures

• 

  Fig. 1. A simplified schema of our multi-level in silico proarrhythmia model. For a single compound, the input comprises a set of IC50 values for the currents IKr, INaL, ICaL, a drug concentration, and a set of electrophysiological simulation parameters. The model translates these inputs to an APD90 prediction

• 

  Fig. 2. Identified sources of aleatory and epistemic uncertainty affecting elements of in silico multi-level proarrhythmia models

• 

  Fig. 3. Structure of our multi-level APD90 prediction model showing the sources of variability that affect model inputs addressed in this work. Δ-IC50s represents the variability in the determined inhibitory drug effects on ion channels involved in physiological action potential generation. Δ-Parameters describes the variability in the electrophysiological model parameters due to inter-individual differences. Δ-[D] is the variability of the drug concentration obtained after the administration of the drug at therapeutic dosage due to inter-individual pharmacokinetic differences. Each of the input variability sources contributes to the overall level of output variability, indicated as (Δ-Prediction)

• 

  Fig. 4. Distributions of APD90 values generated by the population of 860 electrophysiological models for input values #0 (left), #3 (middle), and #60 (right), from the input value combinations shown in the graphics

• 

  Fig. 5. Schema of the three simulation types carried out in this work. Simulation A—propagation of experimental variability associated with pIC50 values, Simulation B—propagation of inter-individual variability arising at the level of electrophysiological model parameters, and Simulation C—propagation of combined experimental and inter-individual variability

• 

  Fig. 6. Violin plots showing distributions of APD90 values obtained in different runs of Monte Carlo simulations introducing the following variability types. Simulation A: Experimental variability (Δ-pIC50); Simulation B: Inter-individual variability (Δ-Parameters); Simulation C: Combination of experimental and inter-individual variability

• 

  Fig. 7. Bar plots showing the median of the APD90 predictions obtained for the 12 CIPA compounds, using three simulation types. Simulation A: Experimental variability (Δ-pIC50); Simulation B: Inter-individual variability (Δ-Parameters); Simulation C: Combination of experimental and inter-individual variability. The intervals represent the 10th and 90th percentiles obtained from the distributions shown in Fig. 6

• 

  Fig. 8. Predicted APD90 values and their corresponding 80% intervals for three selected CiPA compounds assigned to the following arrhythmogenic risk classes: Quinidine as a high-risk drug (red); Ondansetron as an intermediate-risk drug (orange); Mexiletine as a low-risk drug (green). Intervals shown here were obtained in MC Simulations A–C as described in Fig. 4. (color figure online)