Why Data Quality, Not Regulation, Is the True Barrier to AI in Biologics

In this part 1 of a 2-part interview regarding the presentation “From Science to Scale: Crossing the Hype Chasm with Industrialized GMP AI in API and Drug Manufacturing” at CPHI Europe 2025, held Oct 28-30 in Frankfurt, Germany, Toni Manzano, PhD, co-founder and chief scientific officer at Aizon, outlines the challenges and opportunities for integrating AI into the biopharmaceutical sector. Manzano, who also serves as a subject matter expert for life science at the United Nations and is a PDA board member for regulatory affairs, emphasizes that the industry, particularly when handling biologics, deals with highly complex, multi-variable systems.

Analyzing the variability in these complex processes using traditional statistics is ineffective, according to Manzano, as classical methods typically only control variables one by one. AI, functioning as a scientific and mathematical instrument, brings significant value by efficiently managing numerous variables simultaneously, thereby generating actionable and knowledgeable insights from discovery through manufacturing, he says.

Manzanoa adds that, despite the clear benefits, the primary barrier to adoption remains data quality. He notes that AI relies entirely on data; consequently, robust AI models require equally robust, high-quality digital data collected through appropriate digital systems.

However, the regulatory landscape has cleared considerably; Manzano notes favorable news regarding regulatory preparedness, stating, "For the first time in the history, and this is very important, by the first time in the history regulators, FDA in the US and Ima here in Europe from the first time we have guidelines and good practices to apply AI in GMP before the industry."

This regulatory preparation means the GXP carpet is prepared, removing potential regulatory blockers for CMOs and CDMOs, Manzano states. He adds that the critical operational challenge is scaling the technology. The complexity of drug manufacturing dictates that a single AI model cannot solve a full process. Instead, according to Manzano, multiple specific models are necessary, with at least one model per variable (e.g., pH prediction, anomaly detection, or data drifting detection). Each model must replicate a specific combination of product and equipment. A company producing 50 products might thus require at least 100 dedicated AI models. Managing and industrializing this large volume of models is the key problem facing CMOs today, says Manzano, which companies like Aizon are addressing by providing industrialization platforms.

Check out Part 2 of this interview and access all our CPHI Europe coverage!

Transcript

*Editor’s Note: This transcript is a direct, unedited rendering of the original audio/video content. It may contain errors, informal language, or omissions as spoken in the original recording.

My name is Toni Manzano. I'm Chief Scientist Officer and also co-founder of Aizon. Aizon is a cloud company that is providing AI in the drug manufacturing industry, industrialized ai. I'm also teaching at the university and I'm also a subject matter expert for life science in United Nations. And also I am a PDA board member for the regulatory affairs.

Before to answer the question, it's important to know that the pharmaceutical industry is working with very complex systems. From research and discovery till drug manufacturing, we are always facing complicated problems. Of course classical chemistry with APIs and classical genetics and so on. The processes are really very well known, but we are talking about biologic.

Believe me the complex systems are very complicated to analyze in a traditional way. So if we try to understand the variability on these complex processes using classical statistics that doesn't work. That doesn't work because always biologics are always multi-variable and always complex. A very broad strategy is try to tackle the problems using classical statistics where we have only variable by variable control, not altogether. This is where AI brings a lot of value. AI at the end is a scientific tool, is a scientific instrument that we are now teaching at the university. Actually, AI is a very old scientific discipline and this.

Tool. This instrument and this mathematical tool is very well used when you have to manage multi-variable things. A lot of different variables at the same time, not one by one in a complex processes where you have data. In this case, AI brings a lot of value, bringing a lot of understanding of the process and having the capability to provide insights.

Based on actionable insights and also knowledgeable insights. So this is where AI brings value from discovery to drug manufacturing. Main barrier nowadays is that AI only works with data. Data is everything for ai. The main barrier today is if you don't have good data you won't have good AI models.

So this is the main barrier. So first thing, we need to have digital systems that they are able to collect and to create digital data in order to build AI models.

I think that we have very good news in the pharmaceutical sector. The good news is that by the first time in the history, and this is very important, by the first time in the history regulators, FDA in the US and Ima here in Europe from the first time we have guidelines and good practices to apply AI in GMP before the industry.

Adopt AI, that means a lot because nowadays we have all the regulatory plan ahead, the industry, and this is so good. So that means that in terms of CMOs and CDMOs, we can say that we don't have any blocker in terms of regulatory. The GXP carpet is already prepared with all the good practices and guidelines in order to use AI.

This is very good news. Maybe the main handicap in order to adopt AI in GXP is the one that I already mentioned before. We need good data as basics, but in terms of GXP, we have everything already in place.

Let me share with you a real experience. Imagine that you are able to modelize your physical process, your biotech processes, and you have model that is able to answer and to replicate your process as ongoing as going on. The idea that with one model, you can replicate one variable in the process.

One model is able to predict the pH, for example. You will need another model in order to predict another variable density. You will need another model in order to detect potential anomalies or another model to put, to detect potential drifting data drifting. So at the end, the reality is that with only one AI model, we cannot solve the full process.

We are gonna need one specific model per variable, at least per bio. And every single model is responding is replicating and specific combination of product plus equipment. Which is the complexity behind that. The complexity is that if you want to manage drug manufacturing processes and operations very well in an accurate way, you are gonna need more than one AI model per process and product.

So then the problem now is how to industrialize. That because if we had, if we need at least one model per variable and another model to monetize this model, that means that we are gonna need at least two models per product. If we were producing, I dunno, 2050 products in my company, that means I knew I will need 100 models.

How to manage that. This is the main problem that we have nowadays in CMOs how to neutralize this is why. Companies like Azion are providing this platform in order to industrialize AI. This is the goal.