How AI is Transforming the Pharmaceutical Industry

Use case 1:

AI in Pharmacovigilance (Adverse Event Identification, Validation and Reporting):

Pharma companies are pivoting towards solutions beyond the pill solutions by harnessing the broad capabilities of digital and data analytics to research, develop, and craft experiences in new ways across the entire value chain. Some pharma companies are using digitally native solutions that integrate with consumer/clinician activities and workflows to improve potential adverse event detection, accelerating potential remediation, and improving outcomes. Identifying and reporting AEs is an extremely labor intensive process for pharma companies, requiring significant manual effort and is prone to errors. An enormous number of potential AEs are reported annually; however, only a small fraction of them (5-20%: source FDA MedWatch System) end up being actually confirmed post investigation as an AE and linked to the use of pharma companies’ drugs. Each potential AE needs to be assessed and adjudicated carefully to assess if it’s a true AE or not. Therefore, it is a perfect use case for using the power of AI to automate the process and make it more efficient/less prone to errors.

Solutions out there include chatbots/apps on social media and other channels which consumers/clinicians can interact with directly and report their potential AEs that are linked to use of pharma companies’ drugs. These inputs are assessed by AI algorithms to ascertain if the symptoms being experienced were indeed linked to the manufacturer’s drug or potentially something else. This assessment needs intricate algorithms and some manual involvement initially by the AE teams to validate what the AI is adjudicating is correct. The system/data from several AE adjudication cases feeds self-learning, whereby the manual verification overrides the AI recommendation and eventually the AI is optimized by being trained on data that becomes more available over time. If classified as an AE, it is sent to the company's Drug Safety database  and eventually reported to the FDA. Interestingly, these clinician/patient facing chatbots can also be used in a compliant fashion to gather valuable patient/caregiver/HCP data that can inform commercial teams with several key metrics

Outcome: Using AI in AE identification, validation and reporting helps differentiate true AEs from potential AEs and filter out false positive AEs in a highly efficient fashion by reducing millions of dollars of overhead costs linked to manually performing these processes. It also helps leverage end user inputs for improved clinical and commercial insights generation and enhanced overall consumer experience.

Use case 2:

AI in drug discovery:

AI-based drug discovery in oncology has made significant advancements in recent years, transforming the way new cancer treatments are being developed. Drug screening, repurposing and target identifications are three common use cases. One of them, which the author intends to share here, was the topic of his PhD dissertation—the use of in silico/AI-based drug design to characterize and find ligands binding to tumor suppressor protein p53—a pioneering approach that is now being followed by multiple pharma companies to identify  oncology therapeutics, one of which is currently in Phase 2 trials and granted FDA Fast Track Designation. p53 is a key protein in cell cycle regulation, the defect in which causes unchecked growth of more 50% of cancers in humans. In many tumors, p53 is inactivated directly by destabilizing mutations. The aim of the project was to rescue the function of p53 by binding of small-molecule compounds. A lot of efforts have been made in the past to target p53; however, none have ended up in the clinic. The strategy used here was to screen and design compounds that could bind to cavities on the mutated p53 protein surface and thus shift the folding-unfolding equilibrium toward the native folded state. It is a very elegant and unique mechanism of action whereby the aim is to re-activate a mutated protein, and thus very different to traditional mechanisms of blocking protein targets, which is more or less the standard in the industry.

In silico/AI-bbased screening of small molecules that might have a stabilizing effect is a viable and attractive strategy to minimize the number of compounds to a few, which can then be experimentally tested. Using this philosophy, over 2.5 million compounds were computationally screened by employing several filters, such as Lipinski's rule of 5, pharmacophore models,  small moleculedocking and manual analysis of the resulting high-scoring small molecules. For the pharmacophore model, the structure of p53 core DNA binding domain with Y200C mutation (reported to affect 75,000 human cancers every year) was used as a starting point to identify small molecule compounds that bind to the Y220C cavity.

Learn more: https://www.pharmafocusamerica.com/information-technology/how-ai-is-transforming-the-pharmaceutical-industry