Using machine learning surrogate modeling for faster QSP VP cohort generation.

Virtual patients (VPs) are widely used within quantitative systems pharmacology (QSP) modeling to explore the impact of variability and uncertainty on clinical responses. In one method of generating VPs, parameters are sampled randomly from a distribution, and possible VPs are accepted or rejected based on constraints on model output behavior. This approach works but can be inefficient (i.e., the vast majority of model runs typically do not result in valid VPs). Machine learning surrogate models offer an opportunity to improve the efficiency of VP creation significantly. In this approach, surrogate models are trained using the full QSP model and subsequently used to rapidly pre-screen for parameter combinations that result in feasible VPs. The overwhelming majority of parameter combinations pre-vetted using the surrogate models result in valid VPs when tested in the original QSP model. This tutorial presents this novel workflow and demonstrates how a surrogate model software application can be used to select and optimize the surrogate models in a case study. We then discuss the relative efficiency of the methods and scalability of the proposed method.

Quantitative systems pharmacology model of the amyloid pathway in Alzheimer's disease: Insights into the therapeutic mechanisms of clinical candidates.

Despite considerable investment into potential therapeutic approaches for Alzheimer's disease (AD), currently approved treatment options are limited. Predictive modeling using quantitative systems pharmacology (QSP) can be used to guide the design of clinical trials in AD. This study developed a QSP model representing amyloid beta (Aß) pathophysiology in AD. The model included mechanisms of Aß monomer production and aggregation to form insoluble fibrils and plaques; the transport of soluble species between the compartments of brain, cerebrospinal fluid (CSF), and plasma; and the pharmacokinetics, transport, and binding of monoclonal antibodies to targets in the three compartments. Ordinary differential equations were used to describe these processes quantitatively. The model components were calibrated to data from the literature and internal studies, including quantitative data supporting the underlying AD biology and clinical data from clinical trials for anti-Aß monoclonal antibodies (mAbs) aducanumab, crenezumab, gantenerumab, and solanezumab. The model was developed for an apolipoprotein E (APOE) ɛ4 allele carrier and tested for an APOE ɛ4 noncarrier. Results indicate that the model is consistent with data on clinical Aß accumulation in untreated individuals and those treated with monoclonal antibodies, capturing increases in Aß load accurately. This model may be used to investigate additional AD mechanisms and their impact on biomarkers, as well as predict Aß load at different dose levels for mAbs with known targets and binding affinities. This model may facilitate the design of scientifically enriched and efficient clinical trials by enabling a priori prediction of biomarker dynamics in the brain and CSF.

Quantitative systems pharmacology model-based investigation of adverse gastrointestinal events associated with prolonged treatment with PI3-kinase inhibitors.

Several PI3K inhibitors are in clinical development for the treatment of various forms of cancers, including pan-PI3K inhibitors targeting all four PI3K isoforms (α, ß, γ, and δ), and isoform-selective inhibitors. Diarrhea and immune-mediated colitis are among the adverse events observed with PI3K inhibition which limits the maximal tolerated dose. A quantitative systems pharmacology model was developed to investigate PI3K-inhibitor-induced colitis. The effects of individual PI3K isoforms on relevant cellular pathways were incorporated into a mechanistic representation of mucosal inflammation. A virtual clinical population captures the observed clinical variability in the onset timing and rates of diarrhea and colitis for seven clinically tested PI3K inhibitors. Model-based analysis suggests that colitis development is governed by both the inhibition of PI3Kδ, which drives T cell differentiation and proliferation, and PI3Kα, which regulates epithelial barrier integrity. Specifically, when PI3Kα is inhibited below a given threshold, epithelial barrier dysfunction precipitates an exaggerated T effector response due to PI3Kδ-inhibition, leading to risk of diarrhea and colitis. This synergy explains why the lowest diarrhea and colitis rates are seen with the weakest PI3Kδ inhibition (alpelisib), and higher rates are seen with strong PI3Kδ inhibition if PI3Kα is even mildly inhibited (e.g., idelalisib), whereas strong PI3Kδ inhibition in the absence of PI3Kα inhibition does not result in high colitis rates (umbralisib). Thus, the model-based analysis suggests that PI3Kα and δ inhibition play unique but synergistic roles in driving colitis. Finally, we explore if and how dose-regimen might influence colitis rates for molecules that inhibit both PI3Kα and PI3Kδ.

FDA-Industry Scientific Exchange on assessing quantitative systems pharmacology models in clinical drug development: a meeting report, summary of challenges/gaps, and future perspective.

The pharmaceutical industry is actively applying quantitative systems pharmacology (QSP) to make internal decisions and guide drug development. To facilitate the eventual development of a common framework for assessing the credibility of QSP models for clinical drug development, scientists from US Food and Drug Administration and the pharmaceutical industry organized a full-day virtual Scientific Exchange on July 1, 2020. An assessment form was used to ensure consistency in the evaluation process. Among the cases presented, QSP was applied to various therapeutic areas. Applications mostly focused on phase 2 dose selection. Model transparency, including details on expert knowledge and data used for model development, was identified as a major factor for robust model assessment. The case studies demonstrated some commonalities in the workflow of QSP model development, calibration, and validation but differ in the size, scope, and complexity of QSP models, in the acceptance criteria for model calibration and validation, and in the algorithms/approaches used for creating virtual patient populations. Though efforts are being made to build the credibility of QSP models and the confidence is increasing in applying QSP for internal decisions at the clinical stages of drug development, there are still many challenges facing QSP application to late stage drug development. The QSP community needs a strategic plan that includes the ability and flexibility to Adapt, to establish Common expectations for model Credibility needed to inform drug Labeling and patient care, and to AIM to achieve the goal (ACCLAIM).

Higher naloxone dosing in a quantitative systems pharmacology model that predicts naloxone-fentanyl competition at the opioid mu receptor level.

Rapid resuscitation of an opioid overdose with naloxone, an opioid antagonist, is critical. We developed an opioid receptor quantitative systems pharmacology (QSP) model for evaluation of naloxone dosing. In this model we examined three opioid exposure levels that have been reported in the literature (25 ng/ml, 50 ng/ml, and 75 ng/ml of fentanyl). The model predicted naloxone-fentanyl interaction at the mu opioid receptor over a range of three naloxone doses. For a 2 mg intramuscular (IM) dose of naloxone at lower fentanyl exposure levels (25 ng/ml and 50 ng/ml), the time to decreasing mu receptor occupancy by fentanyl to 50% was 3 and 10 minutes, respectively. However, at a higher fentanyl exposure level (75 ng/ml), a dose of 2 mg IM of the naloxone failed to reduce mu receptor occupancy by fentanyl to 50%. In contrast, naloxone doses of 5 mg and 10 mg IM reduced mu receptor occupancy by fentanyl to 50% in 5.5 and 4 minutes respectively. These results suggest that the current doses of naloxone (2 mg IM or 4 mg intranasal (IN)) may be inadequate for rapid reversal of toxicity due to fentanyl exposure and that increasing the dose of naloxone is likely to improve outcomes.

In silico prediction of clinical efficacy.

Drug development is a high risk and costly process, and the ability to predict clinical efficacy in silico (in a computer) can save the pharmaceutical industry time and resources. Additionally, such an approach will result in more targeted, personalized therapies. To date, a number of in silico strategies have been developed to provide better information about the human response to novel therapies earlier in the drug development process. Some of the most prominent include physiological modeling of disease and disease processes, analytical tools for population pharmacodynamics, tools for the analysis of genomic expression data, Monte Carlo simulation technologies, and predictive biosimulation. These strategies are likely to contribute significantly to reducing the failure rate of drugs entering clinical trials.

A Quantitative Systems Pharmacology Platform to Investigate the Impact of Alirocumab and Cholesterol-Lowering Therapies on Lipid Profiles and Plaque Characteristics.

Reduction in low-density lipoprotein cholesterol (LDL-C) is associated with decreased risk for cardiovascular disease. Alirocumab, an antibody to proprotein convertase subtilisin/kexin type 9 (PCSK9), significantly reduces LDL-C. Here, we report development of a quantitative systems pharmacology (QSP) model integrating peripheral and liver cholesterol metabolism, as well as PCSK9 function, to examine the mechanisms of action of alirocumab and other lipid-lowering therapies, including statins. The model predicts changes in LDL-C and other lipids that are consistent with effects observed in clinical trials of single or combined treatments of alirocumab and other treatments. An exploratory model to examine the effects of lipid levels on plaque dynamics was also developed. The QSP platform, on further development and qualification, may support dose optimization and clinical trial design for PCSK9 inhibitors and lipid-modulating drugs. It may also improve our understanding of factors affecting therapeutic responses in different phenotypes of dyslipidemia and cardiovascular disease.