Integrated QSAR Models for Prediction of Serotonergic Activity: Machine Learning Unveiling Activity and Selectivity Patterns of Molecular Descriptors.

Understanding the features of compounds that determine their high serotonergic activity and selectivity for specific receptor subtypes represents a pivotal challenge in drug discovery, directly impacting the ability to minimize adverse events while maximizing therapeutic efficacy. Up to now, this process has been a puzzle and limited to a few serotonergic targets. One approach represented in the literature focuses on receptor structure whereas in this study, we followed another strategy by creating AI-based models capable of predicting serotonergic activity and selectivity based on ligands' representation by molecular descriptors. Predictive models were developed using Automated Machine Learning provided by Mljar and later analyzed through the SHAP importance analysis, which allowed us to clarify the relationship between descriptors and the effect on activity and what features determine selective affinity for serotonin receptors. Through the experiments, it was possible to highlight the most important features of ligands based on highly efficient models. These features are discussed in this manuscript. The models are available in the additional modules of the SerotoninAI application called "Serotonergic activity" and "Selectivity".

SerotoninAI: Serotonergic System Focused, Artificial Intelligence-Based Application for Drug Discovery.

SerotoninAI is an innovative web application for scientific purposes focused on the serotonergic system. By leveraging SerotoninAI, researchers can assess the affinity (pKi value) of a molecule to all main serotonin receptors and serotonin transporters based on molecule structure introduced as SMILES. Additionally, the application provides essential insights into critical attributes of potential drugs such as blood-brain barrier penetration and human intestinal absorption. The complexity of the serotonergic system demands advanced tools for accurate predictions, which is a fundamental requirement in drug development. SerotoninAI addresses this need by providing an intuitive user interface that generates predictions of pKi values for the main serotonergic targets. The application is freely available on the Internet at https://serotoninai.streamlit.app/, implemented in Streamlit with all major web browsers supported. Currently, to the best of our knowledge, there is no tool that allows users to access affinity predictions for serotonergic targets without registration or financial obligations. SerotoninAI significantly increases the scope of drug development activities worldwide. The source code of the application is available at https://github.com/nczub/SerotoninAI_streamlit.

Artificial Intelligence-Based Quantitative Structure-Property Relationship Model for Predicting Human Intestinal Absorption of Compounds with Serotonergic Activity.

Oral medicines represent the largest pharmaceutical market area. To achieve a therapeutic effect, a drug must penetrate the intestinal walls, the main absorption site for orally delivered active pharmaceutical ingredients (APIs). Indeed, predicting drug absorption can facilitate candidate screening and reduce time to market. Algorithms are available with good prediction accuracy that however focus only on solubility. In this work, we focused on drug permeability looking at human intestinal absorption as a marker for intestinal bioavailability. Being of considerable therapeutic relevance, APIs with serotonergic activity were selected as a dataset. Due to process complexity, experimental data scarcity, and variability, we turned toward an artificial intelligence (AI)-based system, which is a hierarchical combination of classification and regression models. This combination of seemingly two models into a single system widens the space of molecules classified as highly permeable with high accuracy. The specialized and optimized system enables in silico and structure-based prediction with a high degree of certainty. Predictions in external validation allowed correct selection of the 38% of highly permeable molecules without any false positives. The proposed system based on AI represents a promising tool useful for oral drug screening at an early stage of drug discovery and development. Datasets and the obtained models are available on the GitHub platform (https://github.com/nczub/HIA_5-HT).

Development and Pharmacokinetics of a Novel Acetylsalicylic Acid Dry Powder for Pulmonary Administration.

Aspirin is an historic blockbuster product, and it has been proposed in a wide range of formulas. Due to exacerbation risks, the pulmonary route has been seldom considered as an alternative to conventional treatments. Only recently, owing to overt advantages, inhalable acetylsalicylic acid dry powders (ASA DPI) began to be considered as an option. In this work, we developed a novel highly performing inhalable ASA DPI using a nano spray-drying technique and leucine as an excipient and evaluated its pharmacokinetics compared with oral administration. The formulation obtained showed remarkable respirability and quality features. Serum and lung ASA DPI profiles showed faster presentation in blood and higher retention compared with oral administration. The dry powder was superior to the DPI suspension. The relative bioavailability in serum and lungs claimed superiority of ASA DPI over oral administration, notwithstanding a fourfold lower pulmonary dose. The obtained ASA DPI formulation shows promising features for the treatment of inflammatory and infectious lung pathologies.

Do AutoML-Based QSAR Models Fulfill OECD Principles for Regulatory Assessment? A 5-HT1A Receptor Case.

The drug discovery and development process requires a lot of time, financial, and workforce resources. Any reduction in these burdens might benefit all stakeholders in the healthcare domain, including patients, government, and companies. One of the critical stages in drug discovery is a selection of molecular structures with a strong affinity to a particular molecular target. The possible solution is the development of predictive models and their application in the screening process, but due to the complexity of the problem, simple and statistical models might not be sufficient for practical application. The manuscript presents the best-in-class predictive model for the serotonin 1A receptor affinity and its validation according to the Organization for Economic Co-operation and Development guidelines for regulatory purposes. The model was developed based on a database with close to 9500 molecules by using an automatic machine learning tool (AutoML). The model selection was conducted based on the Akaike information criterion value and 10-fold cross-validation routine, and later good predictive ability was confirmed with an additional external validation dataset with over 700 molecules. Moreover, the multi-start technique was applied to test if an automatic model development procedure results in reliable results.

Application and Multi-Stage Optimization of Daylight Polymer 3D Printing of Personalized Medicine Products.

Additive technologies have undoubtedly become one of the most intensively developing manufacturing methods in recent years. Among the numerous applications, the interest in 3D printing also includes its application in pharmacy for production of small batches of personalized drugs. For this reason, we conducted multi-stage pre-formulation studies to optimize the process of manufacturing solid dosage forms by photopolymerization with visible light. Based on tests planned and executed according to the design of the experiment (DoE), we selected the optimal quantitative composition of photocurable resin made of PEG 400, PEGDA MW 575, water, and riboflavin, a non-toxic photoinitiator. In subsequent stages, we adjusted the printer set-up and process parameters. Moreover, we assessed the influence of the co-initiators ascorbic acid or triethanolamine on the resin's polymerization process. Next, based on an optimized formulation, we printed and analyzed drug-loaded tablets containing mebeverine hydrochloride, characterized by a gradual release of active pharmaceutical ingredient (API), reaching 80% after 6 h. We proved the possibility of reusing the drug-loaded resin that was not hardened during printing and determined the linear correlation between the volume of the designed tablets and the amount of API, confirming the possibility of printing personalized modified-release tablets.

Puzzle out Machine Learning Model-Explaining Disintegration Process in ODTs.

Tablets are the most common dosage form of pharmaceutical products. While tablets represent the majority of marketed pharmaceutical products, there remain a significant number of patients who find it difficult to swallow conventional tablets. Such difficulties lead to reduced patient compliance. Orally disintegrating tablets (ODT), sometimes called oral dispersible tablets, are the dosage form of choice for patients with swallowing difficulties. ODTs are defined as a solid dosage form for rapid disintegration prior to swallowing. The disintegration time, therefore, is one of the most important and optimizable critical quality attributes (CQAs) for ODTs. Current strategies to optimize ODT disintegration times are based on a conventional trial-and-error method whereby a small number of samples are used as proxies for the compliance of whole batches. We present an alternative machine learning approach to optimize the disintegration time based on a wide variety of machine learning (ML) models through the H2O AutoML platform. ML models are presented with inputs from a database originally presented by Han et al., which was enhanced and curated to include chemical descriptors representing active pharmaceutical ingredient (API) characteristics. A deep learning model with a 10-fold cross-validation NRMSE of 8.1% and an R2 of 0.84 was obtained. The critical parameters influencing the disintegration of the directly compressed ODTs were ascertained using the SHAP method to explain ML model predictions. A reusable, open-source tool, the ODT calculator, is now available at Heroku platform.

Curated Database and Preliminary AutoML QSAR Model for 5-HT1A Receptor.

Introduction of a new drug to the market is a challenging and resource-consuming process. Predictive models developed with the use of artificial intelligence could be the solution to the growing need for an efficient tool which brings practical and knowledge benefits, but requires a large amount of high-quality data. The aim of our project was to develop quantitative structure-activity relationship (QSAR) model predicting serotonergic activity toward the 5-HT1A receptor on the basis of a created database. The dataset was obtained using ZINC and ChEMBL databases. It contained 9440 unique compounds, yielding the largest available database of 5-HT1A ligands with specified pKi value to date. Furthermore, the predictive model was developed using automated machine learning (AutoML) methods. According to the 10-fold cross-validation (10-CV) testing procedure, the root-mean-squared error (RMSE) was 0.5437, and the coefficient of determination (R2) was 0.74. Moreover, the Shapley Additive Explanations method (SHAP) was applied to assess a more in-depth understanding of the influence of variables on the model's predictions. According to to the problem definition, the developed model can efficiently predict the affinity value for new molecules toward the 5-HT1A receptor on the basis of their structure encoded in the form of molecular descriptors. Usage of this model in screening processes can significantly improve the process of discovery of new drugs in the field of mental diseases and anticancer therapy.

Data-Driven Modeling of the Bicalutamide Dissolution from Powder Systems.

Low solubility of active pharmaceutical compounds (APIs) remains an important challenge in dosage form development process. In the manuscript, empirical models were developed and analyzed in order to predict dissolution of bicalutamide (BCL) from solid dispersion with various carriers. BCL was chosen as an example of a poor water-soluble API. Two separate datasets were created: one from literature data and another based on in-house experimental data. Computational experiments were conducted using artificial intelligence tools based on machine learning (AI/ML) with a plethora of techniques including artificial neural networks, decision trees, rule-based systems, and evolutionary computations. The latter resulting in classical mathematical equations provided models characterized by the lowest prediction error. In-house data turned out to be more homogeneous, as well as formulations were more extensively characterized than literature-based data. Thus, in-house data resulted in better models than literature-based data set. Among the other covariates, the best model uses for prediction of BCL dissolution profile the transmittance from IR spectrum at 1260 cm-1 wavenumber. Ab initio modeling-based in silico simulations were conducted to reveal potential BCL-excipients interaction. All crucial variables were selected automatically by AI/ML tools and resulted in reasonably simple and yet predictive models suitable for application in Quality by Design (QbD) approaches. Presented data-driven model development using AI/ML could be useful in various problems in the field of pharmaceutical technology, resulting in both predictive and investigational tools revealing new knowledge.

Quantitative Assessment of the Physiological Parameters Influencing QT Interval Response to Medication: Application of Computational Intelligence Tools.

Human heart electrophysiology is complex biological phenomenon, which is indirectly assessed by the measured ECG signal. ECG trace is further analyzed to derive interpretable surrogates including QT interval, QRS complex, PR interval, and T wave morphology. QT interval and its modification are the most commonly used surrogates of the drug triggered arrhythmia, but it is known that the QT interval itself is determined by other nondrug related parameters, physiological and pathological. In the current study, we used the computational intelligence algorithms to analyze correlations between various simulated physiological parameters and QT interval. Terfenadine given concomitantly with 8 enzymatic inhibitors was used as an example. The equation developed with the use of genetic programming technique leads to general reasoning about the changes in the prolonged QT. For small changes of the QT interval, the drug-related IKr and ICa currents inhibition potentials have major impact. The physiological parameters such as body surface area, potassium, sodium, and calcium ions concentrations are negligible. The influence of the physiological variables increases gradually with the more pronounced changes in QT. As the significant QT prolongation is associated with the drugs triggered arrhythmia risk, analysis of the role of physiological parameters influencing ECG seems to be advisable.

Empirical search for factors affecting mean particle size of PLGA microspheres containing macromolecular drugs.

BACKGROUND AND OBJECTIVES: Poly(lactic-co-glycolic acid) (PLGA) has become one of the most promising in design, development, and optimization for medical applications polymers. PLGA-based multiparticulate dosage forms are usually prepared as microspheres where the size is from 5 to 100 µm, depending on the route of administration. The main objectives of the study were to develop a predictive model of mean volumetric particle size and on its basis extract knowledge of PLGA containing proteins forming behaviour. METHODS: In the present study, a model for the prediction of mean volumetric particle size developed by an rgp package of R environment is presented. Other tools like fscaret, monmlp, fugeR, MARS, SVM, kNNreg, Cubist, randomForest and piecewise linear regression are also applied during the data mining procedure. RESULTS: The feature selection provided by the fscaret package reduced the original input vector from a total of 295 input variables to 10, 16 and 19. The developed models had good predictive ability, which was confirmed by a normalized root-mean-square error (NRMSE) of 6.8 to 11.1% in 10-fold cross validation training procedure. Moreover, the best models were validated using external experimental data. The superior predictiveness had a model obtained by rgp in the form of a classical equation with a normalized root-mean-squared error (NRMSE) of 6.1%. CONCLUSIONS: A new approach is proposed for computational modelling of the mean particle size of PLGA microspheres and rules extraction from tree-based models. The feature selection leads to revealing chemical descriptor variables which are important in predicting the size of PLGA microspheres. In order to achieve better understanding in the relationships between particle size and formulation characteristics, the surface analysis method and rules extraction procedures were applied.

Empirical modeling of the fine particle fraction for carrier-based pulmonary delivery formulations.

In vitro study of the deposition of drug particles is commonly used during development of formulations for pulmonary delivery. The assay is demanding, complex, and depends on: properties of the drug and carrier particles, including size, surface characteristics, and shape; interactions between the drug and carrier particles and assay conditions, including flow rate, type of inhaler, and impactor. The aerodynamic properties of an aerosol are measured in vitro using impactors and in most cases are presented as the fine particle fraction, which is a mass percentage of drug particles with an aerodynamic diameter below 5 µm. In the present study, a model in the form of a mathematical equation was developed for prediction of the fine particle fraction. The feature selection was performed using the R-environment package "fscaret". The input vector was reduced from a total of 135 independent variables to 28. During the modeling stage, techniques like artificial neural networks, genetic programming, rule-based systems, and fuzzy logic systems were used. The 10-fold cross-validation technique was used to assess the generalization ability of the models created. The model obtained had good predictive ability, which was confirmed by a root-mean-square error and normalized root-mean-square error of 4.9 and 11%, respectively. Moreover, validation of the model using external experimental data was performed, and resulted in a root-mean-square error and normalized root-mean-square error of 3.8 and 8.6%, respectively.

Heuristic modeling of macromolecule release from PLGA microspheres.

Dissolution of protein macromolecules from poly(lactic-co-glycolic acid) (PLGA) particles is a complex process and still not fully understood. As such, there are difficulties in obtaining a predictive model that could be of fundamental significance in design, development, and optimization for medical applications and toxicity evaluation of PLGA-based multiparticulate dosage form. In the present study, two models with comparable goodness of fit were proposed for the prediction of the macromolecule dissolution profile from PLGA micro- and nanoparticles. In both cases, heuristic techniques, such as artificial neural networks (ANNs), feature selection, and genetic programming were employed. Feature selection provided by fscaret package and sensitivity analysis performed by ANNs reduced the original input vector from a total of 300 input variables to 21, 17, 16, and eleven; to achieve a better insight into generalization error, two cut-off points for every method was proposed. The best ANNs model results were obtained by monotone multi-layer perceptron neural network (MON-MLP) networks with a root-mean-square error (RMSE) of 15.4, and the input vector consisted of eleven inputs. The complicated classical equation derived from a database consisting of 17 inputs was able to yield a better generalization error (RMSE) of 14.3. The equation was characterized by four parameters, thus feasible (applicable) to standard nonlinear regression techniques. Heuristic modeling led to the ANN model describing macromolecules release profiles from PLGA microspheres with good predictive efficiency. Moreover genetic programming technique resulted in classical equation with comparable predictability to the ANN model.