Hierarchical Bayesian inference for ion channel screening dose-response data

Drug combinations screening using a Bayesian ranking approach based on dose-response models

Drug combinations have been of increasing interest in recent years for the treatment of complex diseases such as cancer, as they could reduce the risk of drug resistance. Moreover, in oncology, combining drugs may allow tackling tumor heterogeneity. Identifying potent combinations can be an arduous task since exploring the full dose-response matrix of candidate combinations over a large number of drugs is costly and sometimes unfeasible, as the quantity of available biological material is limited and may vary across patients. Our objective was to develop a rank-based screening approach for drug combinations in the setting of limited biological resources. A hierarchical Bayesian 4-parameter log-logistic (4PLL) model was used to estimate dose-response curves of dose-candidate combinations based on a parsimonious experimental design. We computed various activity ranking metrics, such as the area under the dose-response curve and Bliss synergy score, and we used the posterior distributions of ranks and the surface under the cumulative ranking curve to obtain a comprehensive final ranking of combinations. Based on simulations, our proposed method achieved good operating characteristics to identifying the most promising treatments in various scenarios with limited sample sizes and interpatient variability. We illustrate the proposed approach on real data from a combination screening experiment in acute myeloid leukemia.

Inclusion of Control Data in Fits to Concentration-Response Curves Improves Estimates of Half-Maximal Concentrations

Concentration-response curves, in which the effect of varying the concentration on the response of an assay is measured, are widely used to evaluate biological effects of chemical compounds. While National Center for Advancing Translational Sciences guidelines specify that readouts should be normalized by the controls, recommended statistical analyses do not explicitly fit to the control data. Here, we introduce a nonlinear regression procedure based on maximum likelihood estimation that determines parameters for the classical Hill equation by fitting the model to both the curve and the control data. Simulations show that the proposed procedure provides more precise parameters compared with previously prescribed practices. Analysis of enzymatic inhibition data from the COVID Moonshot demonstrates that the proposed procedure yields a lower asymptotic standard error for estimated parameters. Benefits are most evident in the analysis of the incomplete curves. We also find that Lenth's outlier detection method appears to determine parameters more precisely.

Considering population variability of electrophysiological models improves the in silico assessment of drug-induced torsadogenic risk

BACKGROUND AND OBJECTIVE

In silico tools are known to aid in drug cardiotoxicity assessment. However, computational models do not usually consider electrophysiological variability, which may be crucial when predicting rare adverse events such as drug-induced Torsade de Pointes (TdP). In addition, classification tools are usually binary and are not validated using an external data set. Here we analyze the role of incorporating electrophysiological variability in the prediction of drug-induced arrhythmogenic-risk, using a ternary classification and two external validation datasets.

METHODS

The effects of the 12 training CiPA drugs were simulated at three different concentrations using a single baseline model and an electrophysiologically calibrated population of models. 9 biomarkers related with action potential (AP), calcium dynamics and net charge were measured for each simulated concentration. These biomarkers were used to build ternary classifiers based on Support Vector Machines (SVM) methodology. Classifiers were validated using two external drug sets: the 16 validation CiPA drugs and 81 drugs from CredibleMeds database.

RESULTS

Population of models allowed to obtain different AP responses under the same pharmacological intervention and improve the prediction of drug-induced TdP with respect to the baseline model. The classification tools based on population of models achieve an accuracy higher than 0.8 and a mean classification error (MCE) lower than 0.3 for both validation drug sets and for the two electrophysiological action potential models studied (Tomek et al. 2020 and a modified version of O'Hara et al. 2011). In addition, simulations with population of models allowed the identification of individuals with lower conductances of I_Kr, I_Ks, and I_NaK and higher conductances of I_CaL, I_NaL, and I_NCX, which are more prone to develop TdP.

CONCLUSIONS

The methodology presented here provides new opportunities to assess drug-induced TdP-risk, taking into account electrophysiological variability and may be helpful to improve current cardiac safety screening methods.

Use of Patient Health Records to Quantify Drug-Related Pro-arrhythmic Risk

There is an increasing expectation that computational approaches may supplement existing human decision-making. Frontloading of models for cardiac safety prediction is no exception to this trend, and ongoing regulatory initiatives propose use of high-throughput in vitro data combined with computational models for calculating proarrhythmic risk. Evaluation of these models requires robust assessment of the outcomes. Using FDA Adverse Event Reporting System reports and electronic healthcare claims data from the Truven-MarketScan US claims database, we quantify the incidence rate of arrhythmia in patients and how this changes depending on patient characteristics. First, we propose that such datasets are a complementary resource for determining relative drug risk and assessing the performance of cardiac safety models for regulatory use. Second, the results suggest important determinants for appropriate stratification of patients and evaluation of additional drug risk in prescribing and clinical support algorithms and for precision health.

In vitro data and computational models can assist with calculating pro-arrhythmic risk

We use patient health records and FDA Adverse Event Reporting System reports

Use of such datasets helps assess relative drug risk and cardiac safety models

We quantify how patient characteristics can affect arrhythmia incidence

A systematic strategy for estimating hERG block potency and its implications in a new cardiac safety paradigm

Introduction

hERG block potency is widely used to calculate a drug's safety margin against its torsadogenic potential. Previous studies are confounded by use of different patch clamp electrophysiology protocols and a lack of statistical quantification of experimental variability. Since the new cardiac safety paradigm being discussed by the International Council for Harmonisation promotes a tighter integration of nonclinical and clinical data for torsadogenic risk assessment, a more systematic approach to estimate the hERG block potency and safety margin is needed.

Methods

A cross-industry study was performed to collect hERG data on 28 drugs with known torsadogenic risk using a standardized experimental protocol. A Bayesian hierarchical modeling (BHM) approach was used to assess the hERG block potency of these drugs by quantifying both the inter-site and intra-site variability. A modeling and simulation study was also done to evaluate protocol-dependent changes in hERG potency estimates.

Results

A systematic approach to estimate hERG block potency is established. The impact of choosing a safety margin threshold on torsadogenic risk evaluation is explored based on the posterior distributions of hERG potency estimated by this method. The modeling and simulation results suggest any potency estimate is specific to the protocol used.

Discussion

This methodology can estimate hERG block potency specific to a given voltage protocol. The relationship between safety margin thresholds and torsadogenic risk predictivity suggests the threshold should be tailored to each specific context of use, and safety margin evaluation may need to be integrated with other information to form a more comprehensive risk assessment.

•

hERG potency/safety margin is a widely used nonclinical cardiac safety strategy.

•

A new regulatory paradigm promotes the integration of nonclinical and clinical data.

•

Lack of uncertainty quantification hindered using hERG potency in the new paradigm.

•

A systematic method was established to address this limitation.

•

Analysis supports using different safety margin thresholds in different context.

Classifying Drugs by their Arrhythmogenic Risk Using Machine Learning

All medications have adverse effects. Among the most serious of these are cardiac arrhythmias. Current paradigms for drug safety evaluation are costly, lengthy, conservative, and impede efficient drug development. Here, we combine multiscale experiment and simulation, high-performance computing, and machine learning to create a risk estimator to stratify new and existing drugs according to their proarrhythmic potential. We capitalize on recent developments in machine learning and integrate information across 10 orders of magnitude in space and time to provide a holistic picture of the effects of drugs, either individually or in combination with other drugs. We show, both experimentally and computationally, that drug-induced arrhythmias are dominated by the interplay between two currents with opposing effects: the rapid delayed rectifier potassium current and the L-type calcium current. Using Gaussian process classification, we create a classifier that stratifies drugs into safe and arrhythmic domains for any combinations of these two currents. We demonstrate that our classifier correctly identifies the risk categories of 22 common drugs exclusively on the basis of their concentrations at 50% current block. Our new risk assessment tool explains under which conditions blocking the L-type calcium current can delay or even entirely suppress arrhythmogenic events. Using machine learning in drug safety evaluation can provide a more accurate and comprehensive mechanistic assessment of the proarrhythmic potential of new drugs. Our study paves the way toward establishing science-based criteria to accelerate drug development, design safer drugs, and reduce heart rhythm disorders.

Classifying Drugs by their Arrhythmogenic Risk Using Machine Learning

All medications have adverse effects. Among the most serious of these are cardiac arrhythmias. Current paradigms for drug safety evaluation are costly, lengthy, conservative, and impede efficient drug development. Here, we combine multiscale experiment and simulation, high-performance computing, and machine learning to create a risk estimator to stratify new and existing drugs according to their proarrhythmic potential. We capitalize on recent developments in machine learning and integrate information across 10 orders of magnitude in space and time to provide a holistic picture of the effects of drugs, either individually or in combination with other drugs. We show, both experimentally and computationally, that drug-induced arrhythmias are dominated by the interplay between two currents with opposing effects: the rapid delayed rectifier potassium current and the L-type calcium current. Using Gaussian process classification, we create a classifier that stratifies drugs into safe and arrhythmic domains for any combinations of these two currents. We demonstrate that our classifier correctly identifies the risk categories of 22 common drugs exclusively on the basis of their concentrations at 50% current block. Our new risk assessment tool explains under which conditions blocking the L-type calcium current can delay or even entirely suppress arrhythmogenic events. Using machine learning in drug safety evaluation can provide a more accurate and comprehensive mechanistic assessment of the proarrhythmic potential of new drugs. Our study paves the way toward establishing science-based criteria to accelerate drug development, design safer drugs, and reduce heart rhythm disorders.

General Principles for the Validation of Proarrhythmia Risk Prediction Models: An Extension of the CiPA In Silico Strategy

This white paper presents principles for validating proarrhythmia risk prediction models for regulatory use as discussed at the In Silico Breakout Session of a Cardiac Safety Research Consortium/Health and Environmental Sciences Institute/US Food and Drug Administration-sponsored Think Tank Meeting on May 22, 2018. The meeting was convened to evaluate the progress in the development of a new cardiac safety paradigm, the Comprehensive in Vitro Proarrhythmia Assay (CiPA). The opinions regarding these principles reflect the collective views of those who participated in the discussion of this topic both at and after the breakout session. Although primarily discussed in the context of in silico models, these principles describe the interface between experimental input and model-based interpretation and are intended to be general enough to be applied to other types of nonclinical models for proarrhythmia assessment. This document was developed with the intention of providing a foundation for more consistency and harmonization in developing and validating different models for proarrhythmia risk prediction using the example of the CiPA paradigm.

Global Sensitivity Analysis of Ventricular Myocyte Model-Derived Metrics for Proarrhythmic Risk Assessment

Multiscale computational models of the heart are being extensively investigated for improved assessment of drug-induced torsades de pointes (TdP) risk, a fatal side effect of many drugs. Model-derived metrics such as action potential duration and net charge carried by ionic currents (qNet) have been proposed as potential candidates for TdP risk stratification after being tested on small datasets. Unlike purely statistical approaches, model-derived metrics are thought to provide mechanism-based classification. In particular, qNet has been recently proposed as a surrogate metric for early afterdepolarizations (EADs), which are known to be cellular triggers of TdP. Analysis of critical model components and of the ion channels that have major impact on model-derived metrics can lead to improvements in the confidence of the prediction. In this paper, we analyze large populations of virtual drugs to systematically examine the influence of different ion channels on model-derived metrics that have been proposed for proarrhythmic risk assessment. We demonstrate via global sensitivity analysis (GSA) that model-derived metrics are most sensitive to different sets of input parameters. Similarly, important differences in sensitivity to specific channel blocks are highlighted when classifying drugs into different risk categories by either qNet or a metric directly based on simulated EADs. In particular, the higher sensitivity of qNet to the block of the late sodium channel might explain why its classification accuracy is better than that of the EAD-based metric, as shown for a small set of known drugs. Our results highlight the need for a better mechanistic interpretation of promising metrics like qNet based on a formal analysis of models. GSA should, therefore, constitute an essential component of the in silico workflow for proarrhythmic risk assessment, as an improved understanding of the structure of model-derived metrics could increase confidence in model-predicted risk.

Four Ways to Fit an Ion Channel Model

Mathematical models of ionic currents are used to study the electrophysiology of the heart, brain, gut, and several other organs. Increasingly, these models are being used predictively in the clinic, for example, to predict the risks and results of genetic mutations, pharmacological treatments, or surgical procedures. These safety-critical applications depend on accurate characterization of the underlying ionic currents. Four different methods can be found in the literature to fit voltage-sensitive ion channel models to whole-cell current measurements: method 1, fitting model equations directly to time-constant, steady-state, and I-V summary curves; method 2, fitting by comparing simulated versions of these summary curves to their experimental counterparts; method 3, fitting to the current traces themselves from a range of protocols; and method 4, fitting to a single current trace from a short and rapidly fluctuating voltage-clamp protocol. We compare these methods using a set of experiments in which hERG1a current was measured in nine Chinese hamster ovary cells. In each cell, the same sequence of fitting protocols was applied, as well as an independent validation protocol. We show that methods 3 and 4 provide the best predictions on the independent validation set and that short, rapidly fluctuating protocols like that used in method 4 can replace much longer conventional protocols without loss of predictive ability. Although data for method 2 are most readily available from the literature, we find it performs poorly compared to methods 3 and 4 both in accuracy of predictions and computational efficiency. Our results demonstrate how novel experimental and computational approaches can improve the quality of model predictions in safety-critical applications.

Enhancing the drug discovery process: Bayesian inference for the analysis and comparison of dose-response experiments

MOTIVATION

The efficacy of a chemical compound is often tested through dose-response experiments from which efficacy metrics, such as the IC50, can be derived. The Marquardt-Levenberg algorithm (non-linear regression) is commonly used to compute estimations for these metrics. The analysis are however limited and can lead to biased conclusions. The approach does not evaluate the certainty (or uncertainty) of the estimates nor does it allow for the statistical comparison of two datasets. To compensate for these shortcomings, intuition plays an important role in the interpretation of results and the formulations of conclusions. We here propose a Bayesian inference methodology for the analysis and comparison of dose-response experiments.

RESULTS

Our results well demonstrate the informativeness gain of our Bayesian approach in comparison to the commonly used Marquardt-Levenberg algorithm. It is capable to characterize the noise of dataset while inferring probable values distributions for the efficacy metrics. It can also evaluate the difference between the metrics of two datasets and compute the probability that one value is greater than the other. The conclusions that can be drawn from such analyzes are more precise.

AVAILABILITY AND IMPLEMENTATION

We implemented a simple web interface that allows the users to analyze a single dose-response dataset, as well as to statistically compare the metrics of two datasets.

Machine learning in drug development: Characterizing the effect of 30 drugs on the QT interval using Gaussian process regression, sensitivity analysis, and uncertainty quantification

Prolonged QT intervals are a major risk factor for ventricular arrhythmias and a leading cause of sudden cardiac death. Various drugs are known to trigger QT interval prolongation and increase the proarrhythmic potential. Yet, how precisely the action of drugs on the cellular level translates into QT interval prolongation on the whole organ level remains insufficiently understood. Here we use machine learning techniques to systematically characterize the effect of 30 common drugs on the QT interval. We combine information from high fidelity three-dimensional human heart simulations with low fidelity one-dimensional cable simulations to build a surrogate model for the QT interval using multi-fidelity Gaussian process regression. Once trained and cross-validated, we apply our surrogate model to perform sensitivity analysis and uncertainty quantification. Our sensitivity analysis suggests that compounds that block the rapid delayed rectifier potassium current I Kr have the greatest prolonging effect of the QT interval, and that blocking the L-type calcium current I CaL and late sodium current I NaL shortens the QT interval. Our uncertainty quantification allows us to propagate the experimental variability from individual block-concentration measurements into the QT interval and reveals that QT interval uncertainty is mainly driven by the variability in I Kr block. In a final validation study, we demonstrate an excellent agreement between our predicted QT interval changes and the changes observed in a randomized clinical trial for the drugs dofetilide, quinidine, ranolazine, and verapamil. We anticipate that both the machine learning methods and the results of this study will have great potential in the efficient development of safer drugs.

Reproducible model development in the cardiac electrophysiology Web Lab

The modelling of the electrophysiology of cardiac cells is one of the most mature areas of systems biology. This extended concentration of research effort brings with it new challenges, foremost among which is that of choosing which of these models is most suitable for addressing a particular scientific question. In a previous paper, we presented our initial work in developing an online resource for the characterisation and comparison of electrophysiological cell models in a wide range of experimental scenarios. In that work, we described how we had developed a novel protocol language that allowed us to separate the details of the mathematical model (the majority of cardiac cell models take the form of ordinary differential equations) from the experimental protocol being simulated. We developed a fully-open online repository (which we termed the Cardiac Electrophysiology Web Lab) which allows users to store and compare the results of applying the same experimental protocol to competing models. In the current paper we describe the most recent and planned extensions of this work, focused on supporting the process of model building from experimental data. We outline the necessary work to develop a machine-readable language to describe the process of inferring parameters from wet lab datasets, and illustrate our approach through a detailed example of fitting a model of the hERG channel using experimental data. We conclude by discussing the future challenges in making further progress in this domain towards our goal of facilitating a fully reproducible approach to the development of cardiac cell models.

Tailoring Mathematical Models to Stem-Cell Derived Cardiomyocyte Lines Can Improve Predictions of Drug-Induced Changes to Their Electrophysiology

Human induced pluripotent stem cell derived cardiomyocytes (iPSC-CMs) have applications in disease modeling, cell therapy, drug screening and personalized medicine. Computational models can be used to interpret experimental findings in iPSC-CMs, provide mechanistic insights, and translate these findings to adult cardiomyocyte (CM) electrophysiology. However, different cell lines display different expression of ion channels, pumps and receptors, and show differences in electrophysiology. In this exploratory study, we use a mathematical model based on iPSC-CMs from Cellular Dynamic International (CDI, iCell), and compare its predictions to novel experimental recordings made with the Axiogenesis Cor.4U line. We show that tailoring this model to the specific cell line, even using limited data and a relatively simple approach, leads to improved predictions of baseline behavior and response to drugs. This demonstrates the need and the feasibility to tailor models to individual cell lines, although a more refined approach will be needed to characterize individual currents, address differences in ion current kinetics, and further improve these results.

Uncertainty Quantification Reveals the Importance of Data Variability and Experimental Design Considerations for in Silico Proarrhythmia Risk Assessment

The Comprehensive in vitro Proarrhythmia Assay (CiPA) is a global initiative intended to improve drug proarrhythmia risk assessment using a new paradigm of mechanistic assays. Under the CiPA paradigm, the relative risk of drug-induced Torsade de Pointes (TdP) is assessed using an in silico model of the human ventricular action potential (AP) that integrates in vitro pharmacology data from multiple ion channels. Thus, modeling predictions of cardiac risk liability will depend critically on the variability in pharmacology data, and uncertainty quantification (UQ) must comprise an essential component of the in silico assay. This study explores UQ methods that may be incorporated into the CiPA framework. Recently, we proposed a promising in silico TdP risk metric (qNet), which is derived from AP simulations and allows separation of a set of CiPA training compounds into Low, Intermediate, and High TdP risk categories. The purpose of this study was to use UQ to evaluate the robustness of TdP risk separation by qNet. Uncertainty in the model parameters used to describe drug binding and ionic current block was estimated using the non-parametric bootstrap method and a Bayesian inference approach. Uncertainty was then propagated through AP simulations to quantify uncertainty in qNet for each drug. UQ revealed lower uncertainty and more accurate TdP risk stratification by qNet when simulations were run at concentrations below 5× the maximum therapeutic exposure (C_max). However, when drug effects were extrapolated above 10× C_max, UQ showed that qNet could no longer clearly separate drugs by TdP risk. This was because for most of the pharmacology data, the amount of current block measured was <60%, preventing reliable estimation of IC₅₀-values. The results of this study demonstrate that the accuracy of TdP risk prediction depends both on the intrinsic variability in ion channel pharmacology data as well as on experimental design considerations that preclude an accurate determination of drug IC₅₀-values in vitro. Thus, we demonstrate that UQ provides valuable information about in silico modeling predictions that can inform future proarrhythmic risk evaluation of drugs under the CiPA paradigm.

Novel Two-Step Classifier for Torsades de Pointes Risk Stratification from Direct Features

While pre-clinical Torsades de Pointes (TdP) risk classifiers had initially been based on drug-induced block of hERG potassium channels, it is now well established that improved risk prediction can be achieved by considering block of non-hERG ion channels. The current multi-channel TdP classifiers can be categorized into two classes. First, the classifiers that take as input the values of drug-induced block of ion channels (direct features). Second, the classifiers that are built on features extracted from output of the drug-induced multi-channel blockage simulations in the in-silico models (derived features). The classifiers built on derived features have thus far not consistently provided increased prediction accuracies, and hence casts doubt on the value of such approaches given the cost of including biophysical detail. Here, we propose a new two-step method for TdP risk classification, referred to as Multi-Channel Blockage at Early After Depolarization (MCB@EAD). In the first step, we classified the compound that produced insufficient hERG block as non-torsadogenic. In the second step, the role of non-hERG channels to modulate TdP risk are considered by constructing classifiers based on direct or derived features at critical hERG block concentrations that generates EADs in the computational cardiac cell models. MCB@EAD provides comparable or superior TdP risk classification of the drugs from the direct features in tests against published methods. TdP risk for the drugs highly correlated to the propensity to generate EADs in the model. However, the derived features of the biophysical models did not improve the predictive capability for TdP risk assessment.

Quantitative Comparison of Effects of Dofetilide, Sotalol, Quinidine, and Verapamil between Human Ex vivo Trabeculae and In silico Ventricular Models Incorporating Inter-Individual Action Potential Variability

Background:In silico modeling could soon become a mainstream method of pro-arrhythmic risk assessment in drug development. However, a lack of human-specific data and appropriate modeling techniques has previously prevented quantitative comparison of drug effects between in silico models and recordings from human cardiac preparations. Here, we directly compare changes in repolarization biomarkers caused by dofetilide, dl-sotalol, quinidine, and verapamil, between in silico populations of human ventricular cell models and ex vivo human ventricular trabeculae. Methods and Results:Ex vivo recordings from human ventricular trabeculae in control conditions were used to develop populations of in silico human ventricular cell models that integrated intra- and inter-individual variability in action potential (AP) biomarker values. Models were based on the O'Hara-Rudy ventricular cardiomyocyte model, but integrated experimental AP variability through variation in underlying ionic conductances. Changes to AP duration, triangulation and early after-depolarization occurrence from application of the four drugs at multiple concentrations and pacing frequencies were compared between simulations and experiments. To assess the impact of variability in IC50 measurements, and the effects of including state-dependent drug binding dynamics, each drug simulation was repeated with two different IC50 datasets, and with both the original O'Hara-Rudy hERG model and a recently published state-dependent model of hERG and hERG block. For the selective hERG blockers dofetilide and sotalol, simulation predictions of AP prolongation and repolarization abnormality occurrence showed overall good agreement with experiments. However, for multichannel blockers quinidine and verapamil, simulations were not in agreement with experiments across all IC50 datasets and IKr block models tested. Quinidine simulations resulted in overprolonged APs and high incidence of repolarization abnormalities, which were not observed in experiments. Verapamil simulations showed substantial AP prolongation while experiments showed mild AP shortening. Conclusions: Results for dofetilide and sotalol show good agreement between experiments and simulations for selective compounds, however lack of agreement from simulations of quinidine and verapamil suggest further work is needed to understand the more complex electrophysiological effects of these multichannel blocking drugs.