Machine learning models for predicting pharmacokinetics and pharmacodynamics: A step toward personalized dosing

Machine learning (ML) has become increasingly integrated within pharmacokinetics (PK) and pharmacodynamics (PD) modeling in recent years, and represents a new paradigm in precision medicine, enabling medically-relevant prediction of drug profile and drug response from data. Many conventional PK/PD models and tools have historically been either incapable of accommodating the complexity of PK/PD as demonstrated in individual patients, for whom genetic, physiological, and environmental factors can increase variability with drug disposition, effectiveness, and safety; therefore, requiring some level of medical judgement, often based on rich personal experience, subjectivity, and uncertainty. Machine learning can efficiently and effectively leverage large datasets of heterogeneous data, unveil subtextual behavior patterns, and facilitate the development of predictive models as research tools that are suited for personalized dosage regimens. In this review, we comprehensively summarize advances in recent ML algorithms such as random forests, support vector machines, neural networks and ensemble methods applied to PK/PD prediction and prediction as parameterized, and in estimation, of ADME parameters, as well as efficacy and toxicity profiles. We also discuss critical challenges such as data quality and representativeness, how the properties of medical research impact model interpretation, clinical integration of ML, and future elements of research opportunities. In summary, bridging computational intelligence with pharmacology is an important step forward for inferences for personalized and individualized optimized therapies to improve clinical outcomes via definitive interventions.