Integrating QSP and ML to Facilitate Drug Development and Personalized Medicine

Handbook of experimental pharmacology, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Language: Английский

---

Network pharmacology approach to unravel the neuroprotective potential of natural products: a narrative review

Molecular Diversity, Journal Year: 2025, Volume and Issue: unknown

Published: April 25, 2025

Language: Английский

---

Leveraging large language models to compare perspectives on integrating QSP and AI/ML

Journal of Pharmacokinetics and Pharmacodynamics, Journal Year: 2025, Volume and Issue: 52(3)

Published: May 5, 2025

Language: Английский

---

Quantitative systems pharmacology in the age of artificial intelligence

CPT Pharmacometrics & Systems Pharmacology, Journal Year: 2023, Volume and Issue: 12(12), P. 1823 - 1826

Published: Oct. 1, 2023

In this perspective, we briefly review the state-of-the-art covering interface between quantitative systems pharmacology (QSP) and artificial intelligence (AI) machine learning (ML); in particular, how AI/ML are conceived used as many attempts to address methodological pain points of QSP. a second part, invite reader step out from discipline-centric view discuss paradigm shift consisting repurposing AI into 2022, Journal Pharmacokinetic Pharmacodynamic published special issue edited by Cho, Zhang, Bonate with 10 scientific articles illustrating ways for coupling (AI). The Zang et al. summarizes current state-of-the-art1 explains that AI/machine (ML) currently four main applications related QSP: parameter estimation, model structure, complexity reduction, virtual population generations (see Figure 1). What may strike is – although there good reasons think improving on such technical domains will increase impact QSP these speak toward dimensions discipline. Estimation parameters, writing down optimal structures, reducing complexity, or properly generating terms which talk modeling experts but probably only them. Parameter estimation known be an because models often too large respect available data. modelers typically working engineers; equations describe complex high level granularity, elementary biological processes unknown uncertain. addition that, clinical end aim predict remote expend effort accurately describe. We, community, trying issues AI/ML. On one hand, big data, omics and/or imaging can treated inputs supervised parameters outputs ref. 2 explanation approach oncology). other could linked endpoints similar formalism where predicted. Finding structural also issue, often, structure designed hand; thus subjected subjectivity coming whomever wrote it. We use data-driven approaches inform through Boolean networks, Bayesian even physics-based AI. has received attention especially safety prediction, shown recent initiative identify molecular players involved drug-induced peripheral neuropathy.3 Regarding authors propose apply feature selection techniques reduce priori lead parsimonious more amenable tasks 1 further information). Last not least, creation populations (VPs) been rightly focus much interest ML, prevalence weighting Markov Chain Monte Carlo, proposed relevant plausible ensemble order generate patients resembling cohorts.4 speed up simulations computational-demanding efficient VP generation.5 key drug development safe medicine, two them immediately come mind. First, multiscale nature points, some areas incorporate aspects daily living, dementia rating Alzheimer's disease (AD) unified Parkinsons's scale (UPDRS). Leveraging data notoriously challenging. This difficulty spread therapeutic given increasing availability patients' reported outcomes. sometimes highlight problem being "subjective." author's opinion, less than multidimensional measures. lacking methods bridge traditional level, our models. means have urgency double innovation progression way descriptive information contained predict. Second, lack validated biomarkers, responsibility look beyond classical markers measured blood tissue potential surrogates action. Still, field neuroscience, would cognitive, behavioral, functional levels. How do we, points? news repurposed/reframed stick fundamental focus. Indeed, several recently reports suggest taking advantage algorithm capability perform well human basis "mechanistic" An illustrative example around reframing algorithms performing at facial recognition their analysis shed new light mechanisms happening suffering AD losing capabilities.6 After having trained (mimicking healthy individual) perturbed different ways, either modifying weight mimic brain signal weakening cutting neural network nodes lesions. Simulations performed results were put perspective biomarker neuroimaging initiative. paper presents very clearly. However, deep interpretability, question whether indeed "reverse engineering" informative actual pathological remains fully open. Following strategy technique called reinforcement (RL) promising supported theoretical arguments. it experimentally RL valid analyzing humans animals' learning. It following words Richard Sutton Andrew Barto start book RL7: When infant learns walk, fall try again, no explicit teacher does direct sensorimotor connection its environment. Exercising produces wealth about cause effect, consequences actions, what achieve goals. mimics process consists knowing map situations actions so maximize numerical reward signal. represented Decision Process agent takes sense environment, clarity goal achieve. "computerized" led including health. Computationally, aims estimating value function iteratively, experience, updated term "temporal difference" (TD). TD difference currrent expectation state is, after experience. If positive, increased, if negative, should then decreased. happen until convergence. highlighted interesting precision dosing non-pharmacological interventions computational psychiatry.8 Before reporting definition psychiatry, let us illustrate examples. 2004 Science, David Redish repurposed simulate addiction.9 previous concept TD, addiction modeled always positive when specific taken (i.e., subject addicted to). By way, result corresponding value. turn, most likely chosen subject. Earlier, discussed need biomarkers useful analyze cognitive testing 8 examples). Computational psychiatry nascent area defined characterize mental dysfunction aberrant computations. coupled (pharmacological) description intervention, nothing else approach. To envisioned today reinforces "system" component highly beneficial areas, neurodegeneration limited earlier. conclusion, see topic interfacing first inner view, QSP-centric, leverage seems underdeveloped box reframe (Figure 2). Embracing make critical advance modalities. doing tremendous role supporting therapy. Recently, modality emerging, digital therapeutics (DTx), software treat, manage, prevent condition. DTx considered modalities behaviors, disorders, sleep pain. reviewed identified application principles greatly benefit DTx, better characterizing mechanism action, optimize intervention content, right dose patients.10 drive modalities, becoming instrumental overcoming another limitation field, namely selectivity candidates reflecting interactions circuits. So, challenge? author acknowledges Dr. Cristina Santini UK invitation present 4th Exchange Workshop July 12, 2023; workshop participants valuable questions feedback helped perspective. thanks Lucy Hutchinson exchanges help finalizing manuscript. No funding was work. employee shareholder F. Hoffmann-La Roche Ltd.

Language: Английский

---

Artificial Intelligence and Disease Modeling: Focus on Neurological Disorders

Clinical Pharmacology & Therapeutics, Journal Year: 2024, Volume and Issue: 115(6), P. 1208 - 1211

Published: March 13, 2024

BR, GP, and AS are employees of F. Hoffmann-La Roche Ltd. All other authors declared no competing interests for this work. NK NI PumasAI who involved in the development DeepNLME-based software.

Language: Английский

---

Application of quantitative pharmacology analysis to support early clinical development of oncology drugs: dose selection

Xenobiotica, Journal Year: 2024, Volume and Issue: 54(7), P. 420 - 423

Published: Aug. 8, 2024

The selection of appropriate starting dose and suitable method to predict an efficacious for novel oncology drug in the early clinical development stage poses significant challenges. traditional methods using body surface area transformation from toxicology studies first-in human (FIH) dose, or simply selecting maximum tolerated (MTD) administered (MAD) as recommended phase 2 (RP2D), are usually inadequate risky drugs.Due regulatory efforts aimed at improving optimisation development, is now shifting away these towards a comprehensive benefit/risk assessment-based approach. Quantitative pharmacology analysis (QPA) plays crucial role this new paradigm. This mini-review summarises use QPA FIH potential doses expansion trials. allows more rational scientifically based approach by integrating information across phases.In conclusion, application has significantly enhance success rates trials ultimately support decision-making, particularly selection.

Language: Английский

---

Systems pharmacology – principles, methods and applications

Elsevier eBooks, Journal Year: 2024, Volume and Issue: unknown, P. 193 - 206

Published: Nov. 8, 2024

Language: Английский

---

Future Directions for Quantitative Systems Pharmacology

Handbook of experimental pharmacology, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 1, 2024

Language: Английский

---