Overcoming Challenges in Bioanalysis With Artificial Intelligence
AI is set to transform bioanalysis, enabling researchers to be more proactive when carrying out routine work.
Complete the form below to unlock access to ALL audio articles.
Artificial intelligence (AI) is transforming the pharmaceutical industry, with many companies already adopting new strategies to harness its potential. The benefits of AI are vast, with applications in drug design, drug repurposing and clinical trial optimization.
In bioanalysis, AI could help streamline routine processes and alleviate the processing and analyzing of preclinical and clinical bioanalytical data. While promising, AI comes with unique challenges ranging from data quality issues to ethical considerations and regulatory compliance. As highlighted by Elsevier’s “Insights 2024: Attitudes toward AI” report, appetites for AI among corporate researchers are strong but concerns remain around misinformation and inaccuracy.
Technology Networks spoke to Mary Belfast, associate director of Automation, Specialty Bioanalytics at Teva Branded Pharmaceuticals, to learn more about some of the roadblocks to the widespread adoption of AI in bioanalysis. In this interview, Belfast also discusses some of the latest guidelines on AI implementation and validation.
Belfast will present a keynote presentation titled “Applying AI/ML Methodologies Within Regulated Bioanalysis” on Monday, October 21, 2024, at PharmSci 360.
Blake Forman (BF): Can you tell us more about the potential and current challenges associated with AI use in bioanalysis?
Mary Belfast (MB): The incorporation of AI in bioanalysis is still in its infancy. AI is a powerful tool but can get us into trouble if it’s not implemented judiciously. We still don’t have rigid guidelines on how to best deploy AI, as we’re still working out where it will best fit in a bioanalysis workflow. Integrating advanced technologies into existing workflows requires careful planning and execution.
One of the potential uses of AI is to help streamline routine processes, such as writing a standard operating procedure (SOP). When writing SOPs, it is a challenge to ensure they don’t conflict with corporate SOPs or regulatory guidelines. It may be possible to use AI to help gather intelligence on what should be included in an SOP.
AI is a powerful tool, but we must be careful how we implement it to ensure data security. For example, we don’t want our data like SOPs to become public information. At the same time, if we put Firewalls in place to protect this information, it will be harder to promptly incorporate updated guidelines and regulations.
BF: What are some of the existing obstacles that are stopping pharmaceutical companies from adopting AI?
MB: AI is written by software engineers who are not typically employed in any large capacity in a pharmaceutical company, so there is often a lack of relevant expertise. Companies need to start seeking out this expertise to help support any AI projects being undertaken. There is also a need to improve the overall workforce’s understanding of these technologies.
AI comes with change. Managing that change and getting employees on board is important. When you eventually have your AI in place, you want employees to be excited about it not throwing up roadblocks, so it's important to foster a culture that embraces change.
Cost is another obstacle. Companies want to know how to optimize the use of AI without having to spend millions of dollars implementing it.
BF: Do you have any advice for other researchers looking to implement AI technologies in their bioanalysis workflows?
MB: It is important to be open to learning more about the various AI tools available. Talking to vendors and going to trade shows is a great way to find out how manufacturers are incorporating AI into their products. Alternatively, look online at some of the available software. This is important to make sure you make an informed purchasing decision so that you can align any AI initiatives with your organization’s goals.
As an example, I remember when I first came into the workforce, in a hospital performing mouth pipetting was commonplace. Now we have these wonderful robots that can do all the liquid transfer for us. You would go about investigating AI tools in the same way you would go about deciding which robot will be best for your liquid transfers. Unfortunately, unlike with robots where there are a select few vendors, there are a lot of AI databases out there to choose from.
BF: Can you tell us more about the latest guidelines related to the use of AI in pharmaceutical R&D and how researchers can ensure they adhere to these guidelines?
MB: Bioanalysis is facing unique AI challenges of its own. If you have a true, unstructured AI, it will teach itself. Bioanalysis requires control of change. So how do you control a change that changes without you knowing? This is a huge roadblock to implementing wide-scale AI in the bioanalysis field.
The International Society for Pharmaceutical Engineering (ISPE) recently launched the second edition of GAMP® 5 (Good Automated Manufacturing Practice). This framework guides the application of new and developing technological areas such as AI and machine learning. GAMP 5 contains suggestions on how to address the issue of unstructured AI. The guide suggests you need to think about data management before you even start your project. First, set aside a portion of your data exclusively for final validation. Next, create a separate dataset for testing and retesting to ensure everything runs as expected and the desired outcome is achieved.
Once you have the desired outcome, test the original dataset you put aside and make sure that nothing has been inadvertently modified. This ensures the robustness of your results.
After implementation, periodically check your data to ensure it hasn't shifted or introduced bias. For example, with AI models like ChatGPT, biases can be introduced from users' responses, aligning the model with popular opinions of the time. Likewise, you must avoid showing a preference for data that gives the best results, as this introduces bias.
BF: How do you envision AI changing the pharmaceutical industry in the next couple of years?
MB: AI is going to change our workforce and the skill sets that are required. We're going to need scientists who possess a reasonable understanding of computers. They don't have to be computer scientists, per se, but they must be willing to adopt technology.
A lot of the routine workflows that we have, such as reviewing instrument or data reports, will become more automated. AI could be used to identify trends in the lab. For example, it could flag an instrument that is aging and should be considered for retirement by a decreasing trend in its performance. Overall, we will end up being more proactive in our routine work, as opposed to being reactive.
Mary Belfast was speaking to Blake Forman, Senior Science Writer for Technology Networks.
About the interviewee:
Credit: Andy Shelter Photography
Mary Belfast is the associate director of Automation, Specialty Bioanalytics at Teva Branded Pharmaceuticals. Belfast has experience in bioassay method development, R&D and process development, advancing many pharmaceutical small molecules, vaccines and biologics from discovery through clinical development. In her current role, she is focused on the seamless transition from manual assay processes to automated platforms and automating data processing.
