Issue link: https://www.e-digitaleditions.com/i/1509086
40 Pharmaceutical Technology ® Trends in Formulation 2023 eBook PharmTech.com CPHI BarCelona Coverage ministration, dosage form, and even the primary con- tainer design can be added as constraints within the analysis framework to optimize the physicochemical, therapeutic, and compliance considerations. Advanced surrogates to human testing, such as physiologically based pharmacokinetic (PBPK) mod- eling require the collection of species–specific physio- logical and chemical-specific absorption, distribution, metabolism, and excretion (ADME) parameters, which can be a time-consuming and expensive process. AI can be used to create computational models capable of predicting input parameter values for PBPK models, especially for new compounds. FDA has stated that it no longer will require animal testing for establishing toxicological safety. The mo- tivation behind this is the poor correlation between animal models and human models. Replacing animal testing requires models which can demonstrate first that they are as at least as good as animal testing es- pecially for predicting the ADME behavior in the body. This may seem like a simple undertaking given the poor correlation between animal and human models, but drug sponsors will have to address the perceived value that animal testing provided rather than the actual cor- relation between models, which is a much more diffi- cult paradigm shift. For example, a sacrificed animal post-testing may have an accumulation of a drug or im- purity in the liver, and, historically, that knowledge was perceived to be of value prior to human testing, even though the actual correlation was not strong between models. Satisfying this thinking will make a surrogate model more difficult to validate. FDA proposed an al- ternative to animal testing utilizing toxicogenomics (TGx) that incorporates emerging genomic technologies into the conventional animal models (8). This offers an unprecedented opportunity to move away from animal testing by inferring toxicity mechanisms based on indi- vidual gene activities and developing safety biomarkers based on gene expression profiles. The TGx model uti- lizes a GAN architecture to generate both gene activi- ties and expression profiles in TGx involving multiple doses and treatment durations. AI in process development. There are myriad ap- plications of AI in large- and small-molecule manufac- turing. Discrete element analysis (DEM) has been used successfully to determine the design space for high- shear granulation processes. The DEM model is used to predict agglomeration as a function of impeller speed and liquid addition rate in a high-shear wet granulator. The model tracks dynamic formation and breakage of liquid bridges between particles as liquid binder in the system is added and corrects for the change in material properties as a function of the binder content. DEM has also been successfully applied to understand the segregation of powders in a binary mixture as well as the effects of varying blade speed and shape in the granulation process. DEM has also been applied to downstream processes predicting the possible path of the tablets in the coating process, along with analysis of time spent by tablets under the spray zone. Continuous processing generates significant data as part of the characterization of material properties and their behavior within each unit operation. For process analytical technology (PAT) applications, utilizing AI can greatly accelerate the intensive statistical analyses typically applied as part of this process development exercise. For example, combining ANN AI with partial least square modeling (PLS) as part of a chemometric analysis strategy has shown to provide better results than either model by themselves. In quality assurance, some drug sponsors are apply- ing AI to critical quality systems such as their devi- ation program. Using historical data AI can quickly compare an event observation with historical devi- ation to identify the appropriate response. This not only reinforces a more consistent application of the organization's quality management system (QMS) it ensures, but past solutions are being applied to avoid failed corrective and preventive actions (CAPA) imple- mentation and recurring CAPAs, all of which impact both productivity and product quality. AI is being applied to the continued process verifi- cation (CPV) program to trend and compare process behavior against predefined criteria and historical behavior, automating control chart generation and making recommendations from a process and quality perspective that is built on, not only the underlying Stage 1 development data but subsequent manufactur- ing data. Together these continuously feed the model design and data set. AI in clinical studies. Clinical trials make up a signif- icant portion of the overall drug development timeline with patient selection and recruitment being the pri- mary challenge. In addition to the time invested, ensur- ing the proper cohorts are identified can make or break a clinical study. AI technology can help sift through large amounts of medical health records and generate data to help find eligible populations for a clinical trial. The technology can also simplify complex entry criteria and make it more presentable to potential candidates. Once a population is selected, AI can help with re- cruitment. Traditionally, eligible patients are found though hospitals or clinics, but often, when recruiting large numbers of people, only a few will be suitable for a trial. When provided with eligibility criteria, an automated system can generate a list of potential participants by examining a database. This list can be used by clinicians to inform their patients about their eligibility, or the patients themselves can be con- tacted directly by the system. NLP can help to stratify patients and, during trials, can quickly identify patient safety events. It is easy to conclude that AI will have a