Predicting blood-to-plasma concentration ratios of drugs from chemical structures and volumes of distribution in humans

A Combination of Machine Learning and PBPK Modeling Approach for Pharmacokinetics Prediction of Small Molecules in Humans

PURPOSE

Recently, there has been rapid development in model-informed drug development, which has the potential to reduce animal experiments and accelerate drug discovery. Physiologically based pharmacokinetic (PBPK) and machine learning (ML) models are commonly used in early drug discovery to predict drug properties. However, basic PBPK models require a large number of molecule-specific inputs from in vitro experiments, which hinders the efficiency and accuracy of these models. To address this issue, this paper introduces a new computational platform that combines ML and PBPK models. The platform predicts molecule PK profiles with high accuracy and without the need for experimental data.

METHODS

This study developed a whole-body PBPK model and ML models of plasma protein fraction unbound ( f up ), Caco-2 cell permeability, and total plasma clearance to predict the PK of small molecules after intravenous administration. Pharmacokinetic profiles were simulated using a "bottom-up" PBPK modeling approach with ML inputs. Additionally, 40 compounds were used to evaluate the platform's accuracy.

RESULTS

Results showed that the ML-PBPK model predicted the area under the concentration-time curve (AUC) with 65.0 % accuracy within a 2-fold range, which was higher than using in vitro inputs with 47.5 % accuracy.

CONCLUSION

The ML-PBPK model platform provides high accuracy in prediction and reduces the number of experiments and time required compared to traditional PBPK approaches. The platform successfully predicts human PK parameters without in vitro and in vivo experiments and can potentially guide early drug discovery and development.

Investigating pharmacokinetic profiles of Centella asiatica using machine learning and PBPK modelling

Physiologically based pharmacokinetic (PBPK) modeling serves as a valuable tool for determining the distribution and disposition of substances in the body of an organism. It involves a mathematical representation of the interrelationships among crucial physiological, biochemical, and physicochemical parameters. A lack of the values of pharmacokinetic parameters can be challenging in constructing a PBPK model. Herein, we propose an artificial intelligence framework to evaluate a key pharmacokinetic parameter, the intestinal effective permeability (Peff). The publicly available Peff dataset was utilized to develop regression machine learning models. The XGBoost model demonstrates the best test accuracy of R-squared (R2, coefficient of determination) of 0.68. The model is then applied to compute the Peff of asiaticoside and madecassoside, the parent compounds found in Centella asiatica. Subsequently, PBPK modeling was conducted to evaluate the biodistribution of the herbal substances following oral administration in a rat model. The simulation results were evaluated and validated, which agreed with the existing in vivo studies in rats. This in silico pipeline presents a potential approach for investigating the pharmacokinetic parameters and profiles of drugs or herbal substances, which can be used independently or integrated into other modeling systems.

Fingerprick volumetric absorptive microsampling for therapeutic drug monitoring of antiseizure medications: Reliability and real-life feasibility in epilepsy patients

Volumetric absorptive microsampling (VAMS) is increasingly proposed as a clinically reliable therapeutic drug monitoring (TDM) sampling methodology. The study aimed to establish the reliability and real-life feasibility of patient self-collected capillary VAMS for TDM of antiseizure medication (ASMs), using plasma ASMs concentrations from venous blood as a reference standard. Nurses collected venous and capillary blood samples using VAMS. Afterward, persons with epilepsy (PWE) performed VAMS sampling by themselves. All samples were analyzed by UHPLC-MS/MS. We performed a cross-validation study, comparing ASMs concentrations obtained by VAMS nurses and patients' self-collected versus plasma through Bland-Altman analysis and Passing-Bablok regression. We enrolled 301 PWE (M: F 42.5%:57.5%; mean age 44±16 years), treated with 13 ASMs, providing a total of 464 measurements. Statistical analysis comparing VAMS self-collected versus plasma ASMs concentrations showed a bias close to zero and slope and intercept values indicating a good agreement for CBZ, LCS, LEV, LTG, OXC, PB, and PHT, while a systematic difference between the two methods was found for VPA, PMP, TPM and ZNS. This is the first study showing the reliability and feasibility of the real-world application of PWE self-collected VAMS for most of the ASMs considered, giving a promising basis for at-home VAMS applications.

A Tissue Distribution Study of Propafenone in an Intentional Fatal Poisoning Case

Propafenone (PPF) belongs to the class 1C antiarrhythmics and can cause electrocardiogram-associated adverse/toxic effects. Cases of PPF intoxication are rarely investigated. We developed a novel and selective GC-MS/MS method for the determination of PPF and its tissue distribution in an intentional fatal poisoning case, which is applicable to PPF quantification in the range of therapeutic to lethal concentrations in complex post-mortem samples. A simple and effective sample pretreatment was applied to all analyzed samples. PPF was determined without the need for dilution, even in highly complex samples containing a wide range of analyte concentrations. Quantification was performed using the standard addition method, developed and validated according to the ICH M10 guidelines. The obtained results indicated that the PPF concentration in the serum from blood taken while alive, before therapy, was the highest ever reported in the literature. Despite the intensive therapy after the patients' admission, the PPF concentrations in the lungs, spleen, femoral blood and cardiac blood were fatal or abnormally high. On the other hand, the concentrations in the liver and skeletal muscle were lower or approximately the same as observed in cases with therapeutic doses. To the best of our knowledge, the distribution of PPF has not been investigated in fatal intoxication cases and can be helpful in clinical or forensic toxicology.

Comparison of maternal venous blood metabolomics collected as dried blood spots, dried blood microsamplers, and plasma for integrative environmental health research

Use of capillary blood devices for exposome research can deepen our understanding of the intricate relationship between environment and health, and open up new avenues for preventive and personalized medicine, particularly for vulnerable populations. While the potential of these whole blood devices to accurately measure chemicals and metabolites has been demonstrated, how untargeted metabolomics data from these samplers can be integrated with previous and ongoing environmental health studies that have used conventional blood collection approaches is not yet clear. Therefore, we performed a comprehensive comparison between relative-quantitative metabolite profiles measured in venous blood collected with dried whole blood microsamplers (DBM), dried whole blood spots (DBS), and plasma from 54 mothers in an ethnically diverse population. We determined that a majority of the 309 chemicals and metabolites showed similar median intensity rank, moderate correlation, and moderate agreement between participant-quantiled intraclass correlation coefficients (ICCs) for pair-wise comparisons among the three biomatrices. In particular, whole blood sample types, DBM and DBS, were in highest agreement across metabolite comparison metrics, followed by metabolites measured in DBM and plasma, and then metabolites measured in DBS and plasma. We provide descriptive characteristics and measurement summaries as a reference database. This includes unique metabolites that were particularly concordant or discordant in pairwise comparisons. Our results demonstrate that the range of metabolites from untargeted metabolomics data collected with DBM, DBS, and plasma provides biologically relevant information for use in independent exposome investigations. However, before meta-analysis with combined datasets are performed, robust statistical approaches that integrate untargeted metabolomics data collected on different blood matrices need to be developed.

Predicting pharmacodynamic effects through early drug discovery with artificial intelligence-physiologically based pharmacokinetic (AI-PBPK) modelling

A mechanism-based pharmacokinetic/pharmacodynamic (PK/PD) model links the concentration-time profile of a drug with its therapeutic effects based on the underlying biological or physiological processes. Clinical endpoints play a pivotal role in drug development. Despite the substantial time and effort invested in screening drugs for favourable pharmacokinetic (PK) properties, they may not consistently yield optimal clinical outcomes. Furthermore, in the virtual compound screening phase, researchers cannot observe clinical outcomes in humans directly. These uncertainties prolong the process of drug development. As incorporation of Artificial Intelligence (AI) into the physiologically based pharmacokinetic/pharmacodynamic (PBPK) model can assist in forecasting pharmacodynamic (PD) effects within the human body, we introduce a methodology for utilizing the AI-PBPK platform to predict the PK and PD outcomes of target compounds in the early drug discovery stage. In this integrated platform, machine learning is used to predict the parameters for the model, and the mechanism-based PD model is used to predict the PD outcome through the PK results. This platform enables researchers to align the PK profile of a drug with desired PD effects at the early drug discovery stage. Case studies are presented to assess and compare five potassium-competitive acid blocker (P-CAB) compounds, after calibration and verification using vonoprazan and revaprazan.

Quetiapine-Related Deaths: In Search of a Surrogate Endpoint

Quetiapine is a second-generation antipsychotic drug available for two and half decades. Due to increased misuse, prescription outside the approved indications, and availability on the black market, it is being encountered in medicolegal autopsies more frequently. For instance, it has been linked to increased mortality rates, most likely due to its adverse effects on the cardiovascular system. Its pharmacokinetic features and significant postmortem redistribution challenge traditional sampling in forensic toxicology. Therefore, a systematic literature review was performed, inclusive of PubMed, the Web of Science-core collection, and the Scopus databases; articles were screened for the terms "quetiapine", "death", and "autopsy" to reevaluate each matrix used as a surrogate endpoint in the forensic toxicology of quetiapine-related deaths. Ultimately, this review considers the results of five studies that were well presented (more than two matrices, data available for all analyses, for instance). The highest quetiapine concentrations were usually measured in the liver tissue. As interpreted by their authors, the results of the considered studies showed a strong correlation between some matrices, but, unfortunately, the studies presented models with poor goodness of fit. The distribution of quetiapine in distinct body compartments/tissues showed no statistically significant relationship with the length of the postmortem interval. Furthermore, this study did not confirm the anecdotal correlation of peripheral blood concentrations with skeletal muscle concentrations. Otherwise, there was no consistency regarding selecting an endpoint for analysis.

Preclinical pharmacokinetic exploration of a novel osteoporotic quinazolinone-benzopyran-indole hybrid (S019-0385) using LC-MS/MS

1. The current investigation was to develop and validate the LC-MS/MS method in order to analyse the various pharmacokinetic parameters of S019-0385. A sensitive, selective, and robust LC-MS/MS approach was established and validated for measuring S019-0385 in female mice plasma and tissue, using optimal multiple reaction monitoring (MRM) transition m/z 488.25/329.12 on positive mode. On a Waters Symmetry Shield C18 column, the analyte was separated using acetonitrile and deionised water with formic acid within 6 min at 0.7 mL/min. Linearity (R2 ≥ 0.99) was observed across 0.195-100 ng/mL concentration range using linear least-squares regression.2. Blood-to-plasma ratio and plasma protein drug binding (%) in mice and human was assessed and found to be less than 1 and >83%, respectively. Absolute bioavailability (%F) of S019-0385 in female Swiss mice was exhibited to be 6.90%. Percent dose excreted S019-0385 in unchanged form through urine and faecal was found to be less than 2% and 0.5%, respectively.3. Following oral administration at 5 mg/kg, the concentration of S019-0385 in tissue distribution was found to be in the order of C small intestine > C bone > C lung > C spleen > C kidney > C liver > C heart > C brain.

Predictive Models Based on Molecular Images and Molecular Descriptors for Drug Screening

Various toxicity and pharmacokinetic evaluations as screening experiments are needed at the drug discovery stage. Currently, to reduce the use of animal experiments and developmental expenses, the development of high-performance predictive models based on quantitative structure-activity relationship analysis is desired. From these evaluation targets, we selected 50% lethal dose (LD50), blood-brain barrier penetration (BBBP), and the clearance (CL) pathway for this investigation and constructed predictive models for each target using 636-11,886 compounds. First, we constructed predictive models using the DeepSnap-deep learning (DL) method and images of compounds as features. The calculated area under the curve (AUC) and balanced accuracy (BAC) were, respectively, 0.887 and 0.818 for LD50, 0.893 and 0.824 for BBBP, and 0.883 and 0.763 for the CL pathway. Next, molecular descriptors (MDs) of compounds were calculated using Molecular Operating Environment, alvaDesc, and ADMET Predictor to construct predictive models using the MD-based method. Using these MDs, we constructed predictive models using DataRobot. The calculated AUC and BAC were, respectively, 0.931 and 0.805 for LD50, 0.919 and 0.849 for BBBP, and 0.900 and 0.807 for the CL pathway. In this investigation, we constructed predictive models combining the DeepSnap-DL and MD-based methods. In ensemble models using the mean predictive probability of the DeepSnap-DL and MD-based methods, the calculated AUC and BAC were, respectively, 0.942 and 0.842 for LD50, 0.936 and 0.853 for BBBP, and 0.908 and 0.832 for the CL pathway, with improved predictive performance observed for all variables compared with either single method alone. Moreover, in consensus models that adopted only compounds for which the results of the two methods agreed, the calculated BAC for LD50, BBBP, and the CL pathway were 0.916, 0.918, and 0.847, respectively, indicating higher predictive performance than the ensemble models for all three variables. The predictive models combining the DeepSnap-DL and MD-based methods displayed high predictive performance for LD50, BBBP, and the CL pathway. Therefore, the application of this approach to prediction targets in various drug discovery screenings is expected to accelerate drug discovery.

Predicting Volume of Distribution in Neonates: Performance of Physiologically Based Pharmacokinetic Modelling

The volume of distribution at steady state (Vss) in neonates is still often estimated through isometric scaling from adult values, disregarding developmental changes beyond body weight. This study aimed to compare the accuracy of two physiologically based pharmacokinetic (PBPK) Vss prediction methods in neonates (Poulin & Theil with Berezhkovskiy correction (P&T+) and Rodgers & Rowland (R&R)) with isometrical scaling. PBPK models were developed for 24 drugs using in-vitro and in-silico data. Simulations were done in Simcyp (V22) using predefined populations. Clinical data from 86 studies in neonates (including preterms) were used for comparison, and accuracy was assessed using (absolute) average fold errors ((A)AFEs). Isometric scaling resulted in underestimated Vss values in neonates (AFE: 0.61), and both PBPK methods reduced the magnitude of underprediction (AFE: 0.82-0.83). The P&T+ method demonstrated superior overall accuracy compared to isometric scaling (AAFE of 1.68 and 1.77, respectively), while the R&R method exhibited lower overall accuracy (AAFE: 2.03). Drug characteristics (LogP and ionization type) and inclusion of preterm neonates did not significantly impact the magnitude of error associated with isometric scaling or PBPK modeling. These results highlight both the limitations and the applicability of PBPK methods for the prediction of Vss in the absence of clinical data.

Machine Learning Methods for Small Data Challenges in Molecular Science

Small data are often used in scientific and engineering research due to the presence of various constraints, such as time, cost, ethics, privacy, security, and technical limitations in data acquisition. However, big data have been the focus for the past decade, small data and their challenges have received little attention, even though they are technically more severe in machine learning (ML) and deep learning (DL) studies. Overall, the small data challenge is often compounded by issues, such as data diversity, imputation, noise, imbalance, and high-dimensionality. Fortunately, the current big data era is characterized by technological breakthroughs in ML, DL, and artificial intelligence (AI), which enable data-driven scientific discovery, and many advanced ML and DL technologies developed for big data have inadvertently provided solutions for small data problems. As a result, significant progress has been made in ML and DL for small data challenges in the past decade. In this review, we summarize and analyze several emerging potential solutions to small data challenges in molecular science, including chemical and biological sciences. We review both basic machine learning algorithms, such as linear regression, logistic regression (LR), k-nearest neighbor (KNN), support vector machine (SVM), kernel learning (KL), random forest (RF), and gradient boosting trees (GBT), and more advanced techniques, including artificial neural network (ANN), convolutional neural network (CNN), U-Net, graph neural network (GNN), Generative Adversarial Network (GAN), long short-term memory (LSTM), autoencoder, transformer, transfer learning, active learning, graph-based semi-supervised learning, combining deep learning with traditional machine learning, and physical model-based data augmentation. We also briefly discuss the latest advances in these methods. Finally, we conclude the survey with a discussion of promising trends in small data challenges in molecular science.

Pharmacokinetic Profile Evaluation of Novel Combretastatin Derivative, LASSBio-1920, as a Promising Colorectal Anticancer Agent

LASSBio-1920 was synthesized due to the poor solubility of its natural precursor, combretastatin A4 (CA4). The cytotoxic potential of the compound against human colorectal cancer cells (HCT-116) and non-small cell lung cancer cells (PC-9) was evaluated, yielding IC₅₀ values of 0.06 and 0.07 μM, respectively. Its mechanism of action was analyzed by microscopy and flow cytometry, where LASSBio-1920 was found to induce apoptosis. Molecular docking simulations and the enzymatic inhibition study with wild-type (wt) EGFR indicated enzyme-substrate interactions similar to other tyrosine kinase inhibitors. We suggest that LASSBio-1920 is metabolized by O-demethylation and NADPH generation. LASSBio-1920 demonstrated excellent absorption in the gastrointestinal tract and high central nervous system (CNS) permeability. The pharmacokinetic parameters obtained by predictions indicated that the compound presents zero-order kinetics and, in a human module simulation, accumulates in the liver, heart, gut, and spleen. The pharmacokinetic parameters obtained will serve as the basis to initiate in vivo studies regarding LASSBio-1920's antitumor potential.

Therapeutic Monitoring of Orally Administered, Small-Molecule Anticancer Medications with Tumor-Specific Cellular Protein Targets in Peripheral Fluid Spaces-A Review

Orally administered, small-molecule anticancer drugs with tumor-specific cellular protein targets (OACD) have revolutionized oncological pharmacotherapy. Nevertheless, the differences in exposure to these drugs in the systemic circulation and extravascular fluid compartments have led to several cases of therapeutic failure, in addition to posing unknown risks of toxicity. The therapeutic drug monitoring (TDM) of OACDs in therapeutically relevant peripheral fluid compartments is therefore essential. In this work, the available knowledge regarding exposure to OACD concentrations in these fluid spaces is summarized. A review of the literature was conducted by searching Embase, PubMed, and Web of Science for clinical research articles and case reports published between 10 May 2001 and 31 August 2022. Results show that, to date, penetration into cerebrospinal fluid has been studied especially intensively, in addition to breast milk, leukocytes, peripheral blood mononuclear cells, peritoneal fluid, pleural fluid, saliva and semen. The typical clinical indications of peripheral fluid TDM of OACDs were (1) primary malignancy, (2) secondary malignancy, (3) mental disorder, and (4) the assessment of toxicity. Liquid chromatography-tandem mass spectrometry was most commonly applied for analysis. The TDM of OACDs in therapeutically relevant peripheral fluid spaces is often indispensable for efficient and safe treatments.

Improved Predictability of Hepatic Clearance with Optimal pH for Acyl-Glucuronidation in Liver Microsomes

The purpose of this study was to investigate the optimal pH for acyl-glucuronidation formation with carboxylic acid-containing compounds in human and rat liver microsomes to improve the predictability of their hepatic clearance. The optimal pH for acyl-glucuronidation of all 17 compounds was around pH 6.0 in human and rat liver microsomes. Correlation analysis was done with the predicted in vitro intrinsic clearance (CL_{int,in vitro}) and in vivo intrinsic clearance (CL_{int,in vivo}) calculated from available reported data of total clearance (CL_tot) of 11 compounds in humans. For 8 of the 11 compounds, under the pH 6.0 condition, the CL_{int,in vitro} were within 1/3 to 3-fold error of the observed CL_{int,in vivo} whereas, the error was within 1/3 to 3-fold of the observed CL_{int,in vivo} for only 3 of the 11 under the pH 7.4 condition. The intracellular pH in human and rat hepatocytes decreased in the presence of a carboxylic acid-containing compound. These findings suggest that acyl-glucuronidation in liver microsomes at pH 6.0 is closer to physiological conditions in the presence of carboxylic acid compounds, and thus, use of this pH condition is important for physiological interpretation and predictability of intrinsic clearance.

Construction of a prediction model for drug removal rate in hemodialysis based on chemical structures

In designing drug dosing for hemodialysis patients, the removal rate (RR) of the drug by hemodialysis is important. However, acquiring the RR is difficult, and there is a need for an estimation method that can be used in clinical settings. In this study, the RR predictive model was constructed using the RR of known drugs by quantitative structure-activity relationship (QSAR) analysis. Drugs were divided into a model construction drug set (75%) and a model validation drug set (25%). The RR was collected from 143 medicines. The objective variable (RR) and chemical structural characteristics (descriptors) of the drug (explanatory variable) were used to construct a prediction model using partial least squares (PLS) regression and artificial neural network (ANN) analyses. The determination coefficients in the PLS and ANN methods were 0.586 and 0.721 for the model validation drug set, respectively. QSAR analysis successfully constructed dialysis RR prediction models that were comparable or superior to those using pharmacokinetic parameters. Considering that the RR dataset contains potential errors, we believe that this study has achieved the most reliable RR prediction accuracy currently available. These predictive RR models can be achieved using only the chemical structure of the drug. This model is expected to be applied at the time of hemodialysis.

Novel QSAR Approach for a Regression Model of Clearance That Combines DeepSnap-Deep Learning and Conventional Machine Learning

The toxicity, absorption, distribution, metabolism, and excretion properties of some targets are difficult to predict by quantitative structure-activity relationship analysis. Therefore, there is a need for a new prediction method that performs well for these targets. The aim of this study was to develop a new regression model of rat clearance (CL). We constructed a regression model using 1545 in-house compounds for which we had rat CL data. Molecular descriptors were calculated using molecular operating environment, alvaDesc, and ADMET Predictor software. The classification model of DeepSnap and Deep Learning (DeepSnap-DL) with images of the three-dimensional chemical structures of compounds as features was constructed, and the prediction probabilities for each compound were calculated. For molecular descriptor-based methods that use molecular descriptors and conventional machine learning algorithms selected by DataRobot, the correlation coefficient (R ²) and root mean square error (RMSE) were 0.625-0.669 and 0.295-0.318, respectively. We combined molecular descriptors and prediction probability of DeepSnap-DL as features and developed a novel regression method we called the combination model. In the combination model with these two types of features and conventional algorithms selected by DataRobot, R ² and RMSE were 0.710-0.769 and 0.247-0.278, respectively. This finding shows that the combination model performed better than molecular descriptor-based methods. Our combination model will contribute to the design of more rational compounds for drug discovery. This method may be applicable not only to rat CL but also to other pharmacokinetic and pharmacological activity and toxicity parameters; therefore, applying it to other parameters may help to accelerate drug discovery.

The blood-to-plasma ratio and predicted GABAA-binding affinity of designer benzodiazepines

PURPOSE

The number of benzodiazepines appearing as new psychoactive substances (NPS) is continually increasing. Information about the pharmacological parameters of these compounds is required to fully understand their potential effects and harms. One parameter that has yet to be described is the blood-to-plasma ratio. Knowledge of the pharmacodynamics of designer benzodiazepines is also important, and the use of quantitative structure-activity relationship (QSAR) modelling provides a fast and inexpensive method of predicting binding affinity to the GABA_A receptor.

METHODS

In this work, the blood-to-plasma ratios for six designer benzodiazepines (deschloroetizolam, diclazepam, etizolam, meclonazepam, phenazepam, and pyrazolam) were determined. A previously developed QSAR model was used to predict the binding affinity of nine designer benzodiazepines that have recently appeared.

RESULTS

Blood-to-plasma values ranged from 0.57 for phenazepam to 1.18 to pyrazolam. Four designer benzodiazepines appearing since 2017 (fluclotizolam, difludiazepam, flualprazolam, and clobromazolam) had predicted binding affinities to the GABA_A receptor that were greater than previously predicted binding affinities for other designer benzodiazepines.

CONCLUSIONS

This work highlights the diverse nature of the designer benzodiazepines and adds to our understanding of their pharmacology. The greater predicted binding affinities are a potential indication of the increasing potency of designer benzodiazepines appearing on the illicit drugs market.

Prediction Model of Clearance by a Novel Quantitative Structure-Activity Relationship Approach, Combination DeepSnap-Deep Learning and Conventional Machine Learning

Some targets predicted by machine learning (ML) in drug discovery remain a challenge because of poor prediction. In this study, a new prediction model was developed and rat clearance (CL) was selected as a target because it is difficult to predict. A classification model was constructed using 1545 in-house compounds with rat CL data. The molecular descriptors calculated by Molecular Operating Environment (MOE), alvaDesc, and ADMET Predictor software were used to construct the prediction model. In conventional ML using 100 descriptors and random forest selected by DataRobot, the area under the curve (AUC) and accuracy (ACC) were 0.883 and 0.825, respectively. Conversely, the prediction model using DeepSnap and Deep Learning (DeepSnap-DL) with compound features as images had AUC and ACC of 0.905 and 0.832, respectively. We combined the two models (conventional ML and DeepSnap-DL) to develop a novel prediction model. Using the ensemble model with the mean of the predicted probabilities from each model improved the evaluation metrics (AUC = 0.943 and ACC = 0.874). In addition, a consensus model using the results of the agreement between classifications had an increased ACC (0.959). These combination models with a high level of predictive performance can be applied to rat CL as well as other pharmacokinetic parameters, pharmacological activity, and toxicity prediction. Therefore, these models will aid in the design of more rational compounds for the development of drugs.