Artificial Intelligence in Drug Delivery and Development
The wonder of AI has spanned numerous sectors, from business to education to crypto and the most efficient healthcare. Drug delivery is a therapeutic way of delivering doses of medication to the sick in a practiced and managed manner. More than just administering drugs for recovery, the right dosage is key to targeting the illness in the bodies of sick people. Over the last decade, increasing interest has been attracted towards the application of artificial intelligence (AI) technology for analyzing and interpreting biological or genetic information, accelerated drug discovery, and identification of selective small-molecule modulators or rare molecules and prediction of their behavior. In the era of antimicrobial resistance, the prevalence of multidrug-resistant microorganisms that resist conventional antibiotic treatment has steadily increased. Emerging studies have demonstrated that artificial intelligence (AI) can transform drug delivery to promote the effective treatment of infectious diseases.
Application of automated workflows and databases for rapid analysis of the huge amounts of data and artificial neural networks (ANNs) for development of the novel hypotheses and treatment strategies, prediction of disease progression, and evaluation of the pharmacological profiles of drug candidates may significantly improve treatment outcomes. implantable drug delivery systems necessitate consideration of a number of points such as dose adjustment, targeted delivery, sustained release, and intelligent control system. Today, drug delivery is rapidly expanding, surpassing the borders between pharmaceutical sciences, biosciences, and engineering. This unprecedented transformation has been prompted by the evolution of computational methods that enable nanosystem design, engineering, and manufacturing.
Artificial intelligence (AI) is poised to advance the design, characterization, and manufacturing of drug delivery nanosystems. Further, the power of big data is becoming a relevant tool to accomplish reverse engineering and continuous optimization of these systems. In this chapter, the implementation of AI in the R&D and manufacturing of drug delivery systems is covered, focusing on pharmaceutical development and close-to-market applications. AI-assisted drug delivery for infectious diseases treatment is achieved mainly through machine learning, which is the most successful technique of AI that could capture the complex relationship between variables. There is a wide range of machine learning models that have been investigated to improve drug delivery for infectious disease treatment.
The evolving pharmaceutical field
The pharmaceutical industry has traditionally been a conservative sector, researching and developing preferentially small-molecule drugs intrinsically featuring (1) stability, (2) adequate potency for therapeutic purposes, and (3) acceptable toxicity for the vast majority of consumers. Among the most powerful approaches in pharmaceutical R&D is the systematic chemical screening of molecular variants in combinatorial libraries, with the aim to provide novel molecules with positive features to be exploited in healthcare. However, this approach is insufficient in providing novel pharmaceutics to the increasingly demanding healthcare industry. Machine learning is an AI procedure composed of four major parts: (1) variable input, (2) feature engineering (i.e., feature extraction and selection), (3) model or algorithm training, and (4) result output. Variables that can be the input of a machine learning model are versatile, examples of which include pathogen genomics, clinical data, and patient medical records.
Pharmaceutical research seeking to enhance drug bioavailability, increase stability and improve organ targeting has been progressively advanced. Pharmaceutical nanocarriers are drug delivery vehicles of submicron size and high versatility. They include polymeric, lipidic, and inorganic nanoparticles, liposomes, nanotubes, nano complexes, niosomes, and many others. In principle, ligands can be attached to the surface of nanocarriers for better uptake and targetability. Even though drug delivery has become more and more important in the pharmaceutical industry due to the extended time, increased cost, and lower productivity of recent molecular commodities. However, even existing formulation development depends on classic trial and error experiments, which are time-consuming, expensive, and unpredictable. With the explosive growth of computing power and algorithms over the past decade, a new system called “computational pharmaceutics” is integrating big data, AI, and multiscale modeling approach into pharmaceutics, proposing a significant potential change to the drug delivery paradigm. Nowadays, some actions are made to apply AI strategies to pharmaceutical product development.
Clinical features such as abnormal systemic immune responses, renal dysfunction, and organ failure, which are typically associated with poor prognostic outcomes of infectious diseases, are often late findings that generally do not allow for an early prediction of adverse outcomes thereby delaying the prompt initiation of new treatments. With the advent of artificial intelligence (AI) methodologies, these above-mentioned challenges could largely be overcome. By integrating insights from diverse disciplines including computer science, neuroscience, and mathematics, AI aims to allow the machine to mimic human intelligence The most successful application of AI is machine learning, which utilizes the machine and algorithms to automatically interpret and learn from pools of data to predict the output of unseen data. Machine learning by AI is well-suited to assist drug delivery for infectious disease treatment due to the following reasons: (1) It can find the applicable features in large-scale and complex datasets to make an accurate prediction without the brute force of repetitive biological tests. Accordingly, it would be a desirable tool for drug development, drug combination selection, and drug dosing optimization, all of which require a huge test space that covers the possible therapeutic compounds, the screening concentration matrices, the required replicates, and perhaps the variety of the pathogenic strains. (2) It can form new rules, detect unforeseen patterns, and disclose hidden knowledge from the data. It would also allow us to optimize the drug delivery system, the drug delivery route, and the drug pharmacokinetics even when the related biological processes and pathways are unspecified. (3) It has a remarkable data processing and analysis speed. (4) By embedding computational software in mobile or clinically available devices, it can operate at the point of care, becoming a practical tool for clinical decision-making on anti-infective drug delivery methods. (5) It can incorporate and learn from new variables of microorganisms, patients, and antimicrobial agents, to refine its performance in real-time and generate a drug delivery plan that is sufficiently flexible to accommodate the common and continuous pathogen evolution.
Commonly used machine learning models and their properties
The most popular machine learning models that have been used in studies on drug delivery for infectious disease treatment have been illustrated in These models include:
Boost: Boost is a type of ensemble learning that aims to train multiple models with an identical structure to improve the robustness of the prediction. Ensemble learning trains the multiple models with randomly initialized parameters to yield a set of alternative models. The final prediction combines the predictions from these different models to reduce their noise, bias, and variance. In general, the boost method can improve the accuracy and reliability compared to the single model. Various boosting algorithms have been introduced, such as adaptive boosting (AdaBoost).
Decision Trees and Random Forest: Decision trees are rule-based machine learning methods. A typical decision tree is a tree-like representation of chains where one root node exemplifies the whole training data set, several branch nodes represent decisions, and several leaf nodes denote the outcomes of the cumulative choices. Each branch node splits the data into homogenous subsets based on the splitting rules defined on the node. The splitting will continue to divide the data into progressively smaller subsets until either all samples within the subsets carry the same label or specific criteria are reached (e.g., the number of the depth reaches the given maximum value).
Logistic Regression: Logistic regression is a widely used machine learning model in medical applications to interpret clinical data. It is a linear method that learns the probability of a sample belonging to a class. Logistic regression is a discriminative model, which directly computes the posterior probability by mapping the input feature vector to the output, without considering the joint distribution of the input and output.
Feedback System Control (FSC): is a closed-loop search algorithm aiming to find the optimal combinations of the input variables by the differential evolution (DE) search method. It is suitable for use in an array of different applications. The key idea of FSC is that it directly focuses on the feedback from the system, i.e., the differences between the target and real system response. The feedback is used as the optimization criteria by the DE search method, which iteratively drives the system toward the desired output [96] or finds the optimal combinations. A typical FSC system usually contains four main steps: (1) input variables that affect the behavior of the system, (2) generate the output of the system given the input variables, (3) use the search algorithm to find the optimal inputs, (4) analyze input-output using a regression method to provide the feedback for the system.
Prediction of treatment outcome
Given AI’s excellent predictive efficacy, researchers have been setting out to build AI tools for clinical outcome prediction of infectious diseases. These AI tools are mostly based on the features of the etiological agent and/or the individual patient (input). For the etiological agent, the commonly used features include genotype, drug resistance-relevant mutations, and drug susceptibility levels. With respect to patients, the commonly used features, in general, fell into four categories, which were patient demographic information (e.g., country, age of onset, sex, ethnicity, and social risk factors), clinical characteristics (e.g., comorbidities, route of infection, baseline pathogen load, treatment record, current treatment regimen, and pharmacokinetic variables, and single nucleotide polymorphism), laboratory test results (e.g., levels of systemic inflammatory mediators), and radiological examinations (e.g., X-rays and CT scans).
Enormous amounts of time and costs in drug research and development necessitate the application of more innovative techniques and strategies. AI technologies offer tremendous opportunities for analyzing the massive amounts of multivariate data, solving the complex problems associated with designing functional drug delivery systems, making more accurate decisions, classification and modeling diseases, accelerated drug discovery, identifying biomarkers, drug targets, potential drug candidates and their pharmacological properties, novel indications for existing therapeutics, relationships between the formulations and processing variables, and physiological or pathophysiological pathways, optimizing dose ratio, and predicting the bioactivities and interactions of drugs, molecular behavior, disease status, cellular response, the efficiency of drug combinations, and treatment outcomes.
References
- Alshawwa SZ, Kassem AA, Farid RM, Mostafa SK, Labib GS. Nanocarrier Drug Delivery Systems: Characterization, Limitations, Future Perspectives and Implementation of Artificial Intelligence. Pharmaceutics. 2022; 14(4):883. https://doi.org/10.3390/pharmaceutics14040883
- Colombo, S. (2020). Applications of artificial intelligence in drug delivery and pharmaceutical development. Artificial Intelligence in Healthcare, 85–116. https://doi.org/10.1016/B978-0-12-818438-7.00004-6
- He, S., Leanse, L. G., & Feng, Y. (2021). Artificial intelligence and. machine learning assisted drug delivery for effective treatment of infectious diseases. Advanced Drug Delivery Reviews, 178, 113922. https://doi.org/10.1016/j.addr.2021.113922
- Kolluru, L.; Atre, P.; Rizvi, S. Characterization and Applications of Colloidal Systems as Versatile Drug Delivery Carriers for Parenteral Formulations. Pharmaceuticals 2021, 14, 108.
- Majumder, J.; Taratula, O.; Minko, T. Nanocarrier-based systems for targeted and site-specific therapeutic delivery. Adv. Drug Deliv. Rev. 2019, 144, 57–77.
