AI in Pharma Adoption, Part 1: Why is AI Not Broadly Adopted in the Lifescience Industries (Pharma, Biologics and Medical Devices)?

The statistical methods and mathematical procedures used in the pharmaceutical industry are pieces of analysis that provide critical information for decision-making. Assuming mathematics are used as an instrument in manufacturing operations, it must be managed in the same way as equipment. Artificial Intelligence can be understood as the combination of math, algorithms, computation and data. Therefore, the validity of its usage in regulated environments must follow the same validity principles and criteria as systems designed for GxP purposes, regardless of their nature.

The regulatory bodies are showing some signs to support the AI application in the life science industries:

• Pharma: The European Pharmacopoeia has promoted two AI algorithms (neural networks and support vector machines) as valid chemometric methods applied to analytical data in pharma processes. This publication is a turning point in the use of AI for critical processes in pharmaceutical environments that is reinforced by several initiatives promoted by the FDA after publishing a discussion paper to solicit comments on the use of artificial intelligence and machine learning as medical devices.

• Devices: Medical devices are being approved by regulatory agencies that have AI algorithms included in the product itself.

• Biologics: Regulatory agencies are approving biologics processing that use AI as part of the GMP validated system.

Even though regulatory agencies are showing some signs of support, what concerns do you think regulatory agencies still have?

If regulatory agencies are willing to accept AI as a valid tool in the life sciences (pharma, biologics and medical devices), then why is AI not being more broadly adopted across the industry? One of the main reasons could be in the fuel for AI: data. Extracting and making sense of data within the life science industries remains a significant issue. The manual analysis of large datasets to identify potential new drug candidates is particularly time-consuming. Additionally, the data often must be put into a form that can be analyzed and not left ambiguously labeled, structured or defined. For example, existing data silos make the implementation of AI difficult because the datasets are not accessible, remain housed in the equipment, reside in outdated or awkward storage media, or there are technical barriers to transmission imposed by obsolete IT infrastructures. For instance, systems, equipment and PLC raw data are usually inaccessible because the facility data networks are too old or IT security prohibits its usage for feeding into AI algorithms. Importantly, data must also follow the FAIR and ALCOA principles (good practices for Data Science and Pharma, respectively) to be considered reliable in life sciences.

Do you feel the life science industries will begin to more rapidly implement the power of artificial intelligence across their manufacturing and quality operations? Why or why not?

https://www.xavierhealth.org/ai-blog/2021/2/2b