Leveraging multi-omics data to empower quantitative systems pharmacology in immuno-oncology.

Understanding the intricate interactions of cancer cells with the tumor microenvironment (TME) is a pre-requisite for the optimization of immunotherapy. Mechanistic models such as quantitative systems pharmacology (QSP) provide insights into the TME dynamics and predict the efficacy of immunotherapy in virtual patient populations/digital twins but require vast amounts of multimodal data for parameterization. Large-scale datasets characterizing the TME are available due to recent advances in bioinformatics for multi-omics data. Here, we discuss the perspectives of leveraging omics-derived bioinformatics estimates to inform QSP models and circumvent the challenges of model calibration and validation in immuno-oncology.

Emulation of Quantitative Systems Pharmacology models to accelerate virtual population inference in immuno-oncology.

Quantitative Systems Pharmacology (QSP) models are increasingly being applied for target discovery and dose selection in immuno-oncology (IO). Typical application involves virtual trial, a simulation of a virtual population of hundreds of model instances with model inputs reflecting individual variability. While the structure of the model and initial parameterisation are based on literature describing the underlying biology, calibration of the virtual population by existing clinical data is frequently required to create tumour and patient population specific model instances. Since comparison of a virtual trial with clinical output requires hundreds of large-scale, non-linear model evaluations, the inference of a virtual population is computationally expensive, frequently becoming a bottleneck. Here, we present novel approach to virtual population inference in IO using emulation of the QSP model and an objective function based on Kolmogorov-Smirnov statistics to maximise congruence of simulated and observed clinical tumour size distributions. We sample the parameter space of a QSP IO model to collect a set of tumour growth time profiles. We evaluate performance of several machine learning approaches in interpolating these time profiles and create a surrogate model, which computes tumor growth profiles faster than the original model and allows examination of tens of millions of virtual patients. We use the surrogate model to infer a virtual population maximising congruence with the waterfall plot of a pembrolizumab clinical trial. We believe that our approach is applicable not only in QSP IO, but also in other applications where virtual populations need to be inferred for computationally expensive mechanistic models.

An In Silico Platform to Predict Cardiotoxicity Risk of Anti-tumor Drug Combination with hiPSC-CMs Based In Vitro Study.

OBJECTIVE: Antineoplastic agent-induced systolic dysfunction is a major reason for interruption of anticancer treatment. Although targeted anticancer agents infrequently cause systolic dysfunction, their combinations with chemotherapies remarkably increase the incidence. Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) provide a potent in vitro model to assess cardiovascular safety. However, quantitatively predicting the reduction of ejection fraction based on hiPSC-CMs is challenging due to the absence of the body's regulatory response to cardiomyocyte injury. METHODS: Here, we developed and validated an in vitro-in vivo translational platform to assess the reduction of ejection fraction induced by antineoplastic drugs based on hiPSC-CMs. The translational platform integrates drug exposure, drug-cardiomyocyte interaction, and systemic response. The drug-cardiomyocyte interaction was implemented as a mechanism-based toxicodynamic (TD) model, which was then integrated into a quantitative system pharmacology-physiological-based pharmacokinetics (QSP-PBPK) model to form a complete translational platform. The platform was validated by comparing the model-predicted and clinically observed incidence of doxorubicin and trastuzumab-induced systolic dysfunction. RESULTS: A total of 33,418 virtual patients were incorporated to receive doxorubicin and trastuzumab alone or in combination. For doxorubicin, the QSP-PBPK-TD model successfully captured the overall trend of systolic dysfunction incidences against the cumulative doses. For trastuzumab, the predicted incidence interval was 0.31-2.7% for single-agent treatment and 0.15-10% for trastuzumab-doxorubicin sequential treatment, covering the observations in clinical reports (0.50-1.0% and 1.5-8.3%, respectively). CONCLUSIONS: In conclusion, the in vitro-in vivo translational platform is capable of predicting systolic dysfunction incidence almost merely depend on hiPSC-CMs, which could facilitate optimizing the treatment protocol of antineoplastic agents.

Quantitative systems pharmacology modeling of HER2-positive metastatic breast cancer for translational efficacy evaluation and combination assessment across therapeutic modalities.

HER2-positive (HER2+) metastatic breast cancer (mBC) is highly aggressive and a major threat to human health. Despite the significant improvement in patients' prognosis given the drug development efforts during the past several decades, many clinical questions still remain to be addressed such as efficacy when combining different therapeutic modalities, best treatment sequences, interindividual variability as well as resistance and potential coping strategies. To better answer these questions, we developed a mechanistic quantitative systems pharmacology model of the pathophysiology of HER2+ mBC that was extensively calibrated and validated against multiscale data to quantitatively predict and characterize the signal transduction and preclinical tumor growth kinetics under different therapeutic interventions. Focusing on the second-line treatment for HER2+ mBC, e.g., antibody-drug conjugates (ADC), small molecule inhibitors/TKI and chemotherapy, the model accurately predicted the efficacy of various drug combinations and dosing regimens at the in vitro and in vivo levels. Sensitivity analyses and subsequent heterogeneous phenotype simulations revealed important insights into the design of new drug combinations to effectively overcome various resistance scenarios in HER2+ mBC treatments. In addition, the model predicted a better efficacy of the new TKI plus ADC combination which can potentially reduce drug dosage and toxicity, while it also shed light on the optimal treatment ordering of ADC versus TKI plus capecitabine regimens, and these findings were validated by new in vivo experiments. Our model is the first that mechanistically integrates multiple key drug modalities in HER2+ mBC research and it can serve as a high-throughput computational platform to guide future model-informed drug development and clinical translation.

Prediction of the liver safety profile of a first-in-class myeloperoxidase inhibitor using quantitative systems toxicology modeling.

The novel myeloperoxidase inhibitor verdiperstat was developed as a treatment for neuroinflammatory and neurodegenerative diseases. During development, a computational prediction of verdiperstat liver safety was performed using DILIsym v8A, a quantitative systems toxicology (QST) model of liver safety.A physiologically-based pharmacokinetic (PBPK) model of verdiperstat was constructed in GastroPlus 9.8, and outputs for liver and plasma time courses of verdiperstat were input into DILIsym. In vitro experiments measured the likelihood that verdiperstat would inhibit mitochondrial function, inhibit bile acid transporters, and generate reactive oxygen species (ROS); these results were used as inputs into DILIsym, with two alternate sets of parameters used in order to fully explore the sensitivity of model predictions. Verdiperstat dosing protocols up to 600 mg BID were simulated for up to 48 weeks using a simulated population (SimPops) in DILIsym.Verdiperstat was predicted to be safe, with only very rare, mild liver enzyme increases as a potential possibility in highly sensitive individuals. Subsequent Phase 3 clinical trials found that ALT elevations in the verdiperstat treatment group were generally similar to those in the placebo group. This validates the DILIsym simulation results and demonstrates the power of QST modelling to predict the liver safety profile of novel therapeutics.

An industry perspective on current QSP trends in drug development.

2023 marks the 10th anniversary of Natpara's submission to the US FDA, which led to the first recorded regulatory interaction where a decision was supported by Quantitative and Systems Pharmacology (QSP) simulations. It had taken about 5 years for the timid QSP discipline to emerge as an effective Model-Informed Drug Development (MIDD) tool with visible impact in the pharmaceutical industry. Since then, the presence of QSP in the regulatory environment has continued to increase, to the point that the Agency reported 60 QSP submissions in 2020 alone, representing ~ 4% of their annual IND submissions [1]. What sort of industry mindset has enabled QSP to reach this level of success? How does QSP fit within the MIDD paradigm? Does QSP mean the same to Discovery and to Clinical Development projects? How do 'platforms' compare to 'fit-for-purpose' QSP models in an industrial setting? Can QSP and empirical Pharmacokinetic-Pharmacodynamic (PKPD) modelling be complementary? What level of validation is required to inform drug development decisions? This article reflects on all these questions, in particular addressing those audiences with limited line-of-sight into the drug industry decision-making machinery.

Virtual clinical trials via a QSP immuno-oncology model to simulate the response to a conditionally activated PD-L1 targeting antibody in NSCLC.

Recently, immunotherapies for antitumoral response have adopted conditionally activated molecules with the objective of reducing systemic toxicity. Amongst these are conditionally activated antibodies, such as PROBODY® activatable therapeutics (Pb-Tx), engineered to be proteolytically activated by proteases found locally in the tumor microenvironment (TME). These PROBODY® therapeutics molecules have shown potential as PD-L1 checkpoint inhibitors in several cancer types, including both effectiveness and locality of action of the molecule as shown by several clinical trials and imaging studies. Here, we perform an exploratory study using our recently published quantitative systems pharmacology model, previously validated for triple-negative breast cancer (TNBC), to computationally predict the effectiveness and targeting specificity of a PROBODY® therapeutics drug compared to the non-modified antibody. We begin with the analysis of anti-PD-L1 immunotherapy in non-small cell lung cancer (NSCLC). As a first contribution, we have improved previous virtual patient selection methods using the omics data provided by the iAtlas database portal compared to methods previously published in literature. Furthermore, our results suggest that masking an antibody maintains its efficacy while improving the localization of active therapeutic in the TME. Additionally, we generalize the model by evaluating the dependence of the response to the tumor mutational burden, independently of cancer type, as well as to other key biomarkers, such as CD8/Treg Tcell and M1/M2 macrophage ratio. While our results are obtained from simulations on NSCLC, our findings are generalizable to other cancer types and suggest that an effective and highly selective conditionally activated PROBODY® therapeutics molecule is a feasible option.

A quantitative systems pharmacology model of plasma kallikrein-kinin system dysregulation in hereditary angioedema.

Hereditary angioedema (HAE) due to C1-inhibitor deficiency is a rare, debilitating, genetic disorder characterized by recurrent, unpredictable, attacks of edema. The clinical symptoms of HAE arise from excess bradykinin generation due to dysregulation of the plasma kallikrein-kinin system (KKS). A quantitative systems pharmacology (QSP) model that mechanistically describes the KKS and its role in HAE pathophysiology was developed based on HAE attacks being triggered by autoactivation of factor XII (FXII) to activated FXII (FXIIa), resulting in kallikrein production from prekallikrein. A base pharmacodynamic model was constructed and parameterized from literature data and ex vivo assays measuring inhibition of kallikrein activity in plasma of HAE patients or healthy volunteers who received lanadelumab. HAE attacks were simulated using a virtual patient population, with attacks recorded when systemic bradykinin levels exceeded 20 pM. The model was validated by comparing the simulations to observations from lanadelumab and plasma-derived C1-inhibitor clinical trials. The model was then applied to analyze the impact of nonadherence to a daily oral preventive therapy; simulations showed a correlation between the number of missed doses per month and reduced drug effectiveness. The impact of reducing lanadelumab dosing frequency from 300 mg every 2 weeks (Q2W) to every 4 weeks (Q4W) was also examined and showed that while attack rates with Q4W dosing were substantially reduced, the extent of reduction was greater with Q2W dosing. Overall, the QSP model showed good agreement with clinical data and could be used for hypothesis testing and outcome predictions.

Assessing the performance of QSP models: biology as the driver for validation.

Validation of a quantitative model is a critical step in establishing confidence in the model's suitability for whatever analysis it was designed. While processes for validation are well-established in the statistical sciences, the field of quantitative systems pharmacology (QSP) has taken a more piecemeal approach to defining and demonstrating validation. Although classical statistical methods can be used in a QSP context, proper validation of a mechanistic systems model requires a more nuanced approach to what precisely is being validated, and what role said validation plays in the larger context of the analysis. In this review, we summarize current thoughts of QSP validation in the scientific community, contrast the aims of statistical validation from several contexts (including inference, pharmacometrics analysis, and machine learning) with the challenges faced in QSP analysis, and use examples from published QSP models to define different stages or levels of validation, any of which may be sufficient depending on the context at hand.

Towards a platform quantitative systems pharmacology (QSP) model for preclinical to clinical translation of antibody drug conjugates (ADCs).

A next generation multiscale quantitative systems pharmacology (QSP) model for antibody drug conjugates (ADCs) is presented, for preclinical to clinical translation of ADC efficacy. Two HER2 ADCs (trastuzumab-DM1 and trastuzumab-DXd) were used for model development, calibration, and validation. The model integrates drug specific experimental data including in vitro cellular disposition data, pharmacokinetic (PK) and tumor growth inhibition (TGI) data for T-DM1 and T-DXd, as well as system specific data such as properties of HER2, tumor growth rates, and volumes. The model incorporates mechanistic detail at the intracellular level, to account for different mechanisms of ADC processing and payload release. It describes the disposition of the ADC, antibody, and payload inside and outside of the tumor, including binding to off-tumor, on-target sinks. The resulting multiscale PK model predicts plasma and tumor concentrations of ADC and payload. Tumor payload concentrations predicted by the model were linked to a TGI model and used to describe responses following ADC administration to xenograft mice. The model was translated to humans and virtual clinical trial simulations were performed that successfully predicted progression free survival response for T-DM1 and T-DXd for the treatment of HER2+ metastatic breast cancer, including differential efficacy based upon HER2 expression status. In conclusion, the presented model is a step toward a platform QSP model and strategy for ADCs, integrating multiple types of data and knowledge to predict ADC efficacy. The model has potential application to facilitate ADC design, lead candidate selection, and clinical dosing schedule optimization.

Systems Pharmacology: Defining the Interactions of Drug Combinations.

The majority of diseases are associated with alterations in multiple molecular pathways and complex interactions at the cellular and organ levels. Single-target monotherapies therefore have intrinsic limitations with respect to their maximum therapeutic benefits. The potential of combination drug therapies has received interest for the treatment of many diseases and is well established in some areas, such as oncology. Combination drug treatments may allow us to identify synergistic drug effects, reduce adverse drug reactions, and address variability in disease characteristics between patients. Identification of combination therapies remains challenging. We discuss current state-of-the-art systems pharmacology approaches to enable rational identification of combination therapies. These approaches, which include characterization of mechanisms of disease and drug action at a systems level, can enable understanding of drug interactions at the molecular, cellular, physiological, and organismal levels. Such multiscale understanding can enable precision medicine by promoting the rational development of combination therapy at the level of individual patients for many diseases.

Quantitative Systems Pharmacology Modeling Framework of Autophagy in Tuberculosis: Application to Adjunctive Metformin Host-Directed Therapy.

Quantitative systems pharmacology (QSP) modeling of the host immune response against Mycobacterium tuberculosis can inform the rational design of host-directed therapies (HDTs). We aimed to develop a QSP framework to evaluate the effects of metformin-associated autophagy induction in combination with antibiotics. A QSP framework for autophagy was developed by extending a model for host immune response to include adenosine monophosphate-activated protein kinase (AMPK)-mTOR-autophagy signaling. This model was combined with pharmacokinetic-pharmacodynamic models for metformin and antibiotics against M. tuberculosis. We compared the model predictions to mice infection experiments and derived predictions for the pathogen- and host-associated dynamics in humans treated with metformin in combination with antibiotics. The model adequately captured the observed bacterial load dynamics in mice M. tuberculosis infection models treated with metformin. Simulations for adjunctive metformin therapy in newly diagnosed patients suggested a limited yet dose-dependent effect of metformin on reduction of the intracellular bacterial load when the overall bacterial load is low, late during antibiotic treatment. We present the first QSP framework for HDTs against M. tuberculosis, linking cellular-level autophagy effects to disease progression and adjunctive HDT treatment response. This framework may be extended to guide the design of HDTs against M. tuberculosis.

A mathematical model to identify optimal combinations of drug targets for dupilumab poor responders in atopic dermatitis.

BACKGROUND: Several biologics for atopic dermatitis (AD) have demonstrated good efficacy in clinical trials, but with a substantial proportion of patients being identified as poor responders. This study aims to understand the pathophysiological backgrounds of patient variability in drug response, especially for dupilumab, and to identify promising drug targets in dupilumab poor responders. METHODS: We conducted model-based meta-analysis of recent clinical trials of AD biologics and developed a mathematical model that reproduces reported clinical efficacies for nine biological drugs (dupilumab, lebrikizumab, tralokinumab, secukinumab, fezakinumab, nemolizumab, tezepelumab, GBR 830, and recombinant interferon-gamma) by describing system-level AD pathogenesis. Using this model, we simulated the clinical efficacy of hypothetical therapies on virtual patients. RESULTS: Our model reproduced reported time courses of %improved EASI and EASI-75 of the nine drugs. The global sensitivity analysis and model simulation indicated the baseline level of IL-13 could stratify dupilumab good responders. Model simulation on the efficacies of hypothetical therapies revealed that simultaneous inhibition of IL-13 and IL-22 was effective, whereas application of the nine biologic drugs was ineffective, for dupilumab poor responders (EASI-75 at 24 weeks: 21.6% vs. max. 1.9%). CONCLUSION: Our model identified IL-13 as a potential predictive biomarker to stratify dupilumab good responders, and simultaneous inhibition of IL-13 and IL-22 as a promising drug therapy for dupilumab poor responders. This model will serve as a computational platform for model-informed drug development for precision medicine, as it allows evaluation of the effects of new potential drug targets and the mechanisms behind patient variability in drug response.

A framework for simplification of quantitative systems pharmacology models in clinical pharmacology.

Quantitative systems pharmacology (QSP) is a relatively new discipline within modelling and simulation that has gained wide attention over the past few years. The application of QSP models spans drug-target identification and validation, through all drug development phases as well as clinical applications. Due to their detailed mechanistic nature, QSP models are capable of extrapolating knowledge to predict outcomes in scenarios that have not been tested experimentally, making them an important resource in experimental and clinical pharmacology. However, these models are complicated to work with due to their size and inherent complexity. This makes many applications of QSP models for simulation, parameter estimation and trial design computationally intractable. A number of techniques have been developed to simplify QSP models into smaller models that are more amenable to further analyses while retaining their accurate predictive capabilities. Different simplification techniques have different strengths and weaknesses and hence different utilities. Understanding the utilities of different methods is essential for selection of the best method for a particular situation. In this paper, we have created an overall framework for model simplification techniques that allows a natural categorisation of methods based on their utility. We provide a brief description of the concept underpinning the different methods and example applications. A summary of the utilities of methods is intended to provide a guide to modellers in their model endeavours to simplify these complicated models.

A Quantitative Systems Pharmacology Perspective on the Importance of Parameter Identifiability.

There is an inherent tension in Quantitative Systems Pharmacology (QSP) between the need to incorporate mathematical descriptions of complex physiology and drug targets with the necessity of developing robust, predictive and well-constrained models. In addition to this, there is no "gold standard" for model development and assessment in QSP. Moreover, there can be confusion over terminology such as model and parameter identifiability; complex and simple models; virtual populations; and other concepts, which leads to potential miscommunication and misapplication of methodologies within modeling communities, both the QSP community and related disciplines. This perspective article highlights the pros and cons of using simple (often identifiable) vs. complex (more physiologically detailed but often non-identifiable) models, as well as aspects of parameter identifiability, sensitivity and inference methodologies for model development and analysis. The paper distills the central themes of the issue of identifiability and optimal model size and discusses open challenges.

Towards a comprehensive assessment of QSP models: what would it take?

Quantitative Systems Pharmacology (QSP) has emerged as a powerful ensemble of approaches aiming at developing integrated mathematical and computational models elucidating the complex interactions between pharmacology, physiology, and disease. As the field grows and matures its applications expand beyond the boundaries of research and development and slowly enter the decision making and regulatory arenas. However, widespread acceptance and eventual adoption of a new modeling approach requires assessment criteria and quantifiable metrics that establish credibility and increase confidence in model predictions. QSP aims to provide an integrated understanding of pathology in the context of therapeutic interventions. Because of its ambitious nature and the fact that QSP emerged in an uncoordinated manner as a result of activities distributed across organizations and academic institutions, high entropy characterizes the tools, methods, and computational methodologies and approaches used. The eventual acceptance of QSP model predictions as supporting material for an application to a regulatory agency will require that two key aspects are considered: (1) increase confidence in the QSP framework, which drives standardization and assessment; and (2) careful articulation of the expectations. Both rely heavily on our ability to rigorously and consistently assess QSP models. In this manuscript, we wish to discuss the meaning and purpose of such an assessment in the context of QSP model development and elaborate on the differentiating features of QSP that render such an endeavor challenging. We argue that QSP establishes a conceptual, integrative framework rather than a specific and well-defined computational methodology. QSP elicits the use of a wide variety of modeling and computational methodologies optimized with respect to specific applications and available data modalities, which exceed the data structures employed by chemometrics and PK/PD models. While the range of options fosters creativity and promises to substantially advance our ability to design pharmaceutical interventions rationally and optimally, our expectations of QSP models need to be clearly articulated and agreed on, with assessment emphasizing the scope of QSP studies rather than the methods used. Nevertheless, QSP should not be considered an independent approach, rather one of many in the broader continuum of computational models.

Accelerating robust plausible virtual patient cohort generation by substituting ODE simulations with parameter space mapping.

The generation of plausible virtual patients (VPs) is an important step in most quantitative systems pharmacology (QSP) workflows, which requires time-intensive solving of ordinary differential equations (ODEs). However, non-physiological profiles of outputs of interest (OoI) are frequently produced, and additional acceptance/rejection steps are needed for comparing and removing VPs with predicted values outside a pre-defined range. Here, a new approach is developed to accelerate the acceptance/rejection steps by leveraging patterns of parameter associations with OoI. In most models, some parameters are monotonic with respect to OoI, such that an increase in a parameter value always induces an increase or decrease in the OoI. This monotonic property can be used to replace some ODE-solving steps with appropriate monotonic parameter value comparisons to extrapolate the rejection or interpolate the acceptance of some VPs (after simulation) to others. Two algorithms were built that directly extract plausible VPs from a pre-defined initial cohort. These algorithms were first tested using a simple tumor growth inhibition model. Analyzing 200,000 VPs took 50 s with a reference method and 3 to 41 s (depending on the initial set-up) with the first algorithm. The method was then applied to an apoptosis QSP model, in which the clinical phenotypes (i.e., treatment sensitive or resistant) of 200,000 VPs were fully characterized for four different drug regimens in 12 min as compared to over 80 min with the reference approach. Extraction of each phenotype can also be performed individually in 34 s to 8 min, demonstrating the time benefit and flexibility of this approach.

Recent applications of quantitative systems pharmacology and machine learning models across diseases.

Quantitative systems pharmacology (QSP) is a quantitative and mechanistic platform describing the phenotypic interaction between drugs, biological networks, and disease conditions to predict optimal therapeutic response. In this meta-analysis study, we review the utility of the QSP platform in drug development and therapeutic strategies based on recent publications (2019-2021). We gathered recent original QSP models and described the diversity of their applications based on therapeutic areas, methodologies, software platforms, and functionalities. The collection and investigation of these publications can assist in providing a repository of recent QSP studies to facilitate the discovery and further reusability of QSP models. Our review shows that the largest number of QSP efforts in recent years is in Immuno-Oncology. We also addressed the benefits of integrative approaches in this field by presenting the applications of Machine Learning methods for drug discovery and QSP models. Based on this meta-analysis, we discuss the advantages and limitations of QSP models and propose fields where the QSP approach constitutes a valuable interface for more investigations to tackle complex diseases and improve drug development.

Application of different approaches to generate virtual patient populations for the quantitative systems pharmacology model of erythropoiesis.

In a standard situation, a quantitative systems pharmacology model describes a "reference patient," and the model parameters are fixed values allowing only the mean values to be described. However, the results of clinical trials include a description of variability in patients' responses to a drug, which is typically expressed in terms of conventional statistical parameters, such as standard deviations (SDs) from mean values. Therefore, in this study, we propose and compare four different approaches: (1) Monte Carlo Markov Chain (MCMC); (2) model fitting to Monte Carlo sample; (3) population of clones; (4) stochastically bounded selection to generate virtual patient populations based on experimentally measured mean data and SDs. We applied these approaches to generate virtual patient populations in the QSP model of erythropoiesis. According to the results of our research, stochastically bounded selection showed slightly better results than the other three methods as it allowed the description of any number of patients from clinical trials and could be applied in the case of complex models with a large number of variable parameters.

Understanding virtual patients efficiently and rigorously by combining machine learning with dynamical modelling.

Individual biological organisms are characterized by daunting heterogeneity, which precludes describing or understanding populations of 'patients' with a single mathematical model. Recently, the field of quantitative systems pharmacology (QSP) has adopted the notion of virtual patients (VPs) to cope with this challenge. A typical population of VPs represents the behavior of a heterogeneous patient population with a distribution of parameter values over a mathematical model of fixed structure. Though this notion of VPs is a powerful tool to describe patients' heterogeneity, the analysis and understanding of these VPs present new challenges to systems pharmacologists. Here, using a model of the hypothalamic-pituitary-adrenal axis, we show that an integrated pipeline that combines machine learning (ML) and bifurcation analysis can be used to effectively and efficiently analyse the behaviors observed in populations of VPs. Compared with local sensitivity analyses, ML allows us to capture and analyse the contributions of simultaneous changes of multiple model parameters. Following up with bifurcation analysis, we are able to provide rigorous mechanistic insight regarding the influences of ML-identified parameters on the dynamical system's behaviors. In this work, we illustrate the utility of this pipeline and suggest that its wider adoption will facilitate the use of VPs in the practice of systems pharmacology.