Artificial Intelligence and Machine Learning: More Than Just Hype

Bigfinite Briefing Note

By: Sandra K. Rodriguez, Market Analyst

Axendia was recently briefed by the Executive Management team at Bigfinite following their announcements of their Series B funding and news of their collaboration with Honeywell.  With an executive team that has over thirty years of experience in the Life Science industry, Bigfinite helps biotech and pharmaceutical companies establish a lasting, competitive advantage through an agile and effective use of manufacturing data via their GxP-compliant AI/ML SaaS platform. The platform is offered as a single tenant, virtual private cloud for every one of its clients for security and for confidentiality purposes.

“We are not only tech people doing pharma, we are also pharma people doing tech,” began Pep Gubau, CEO at Bigfinite.  Manufacturing intelligence is not new; yet it remains a challenge.  Pharma, as well as other industries, have dedicated a lot of effort and technology towards this initiative for many years. “We believe manufacturing processes can’t be improved due to an incomplete understanding of operational realities.  There are still too many things that contribute to variations in processes that are not well understood. That leads to repeat deviations and the inability to get the root causes of those deviations,” said Claus Abildgren, VP Business Development at Bigfinite.

Leveraging AI for Root Cause Analysis

For example, Bigfinite’s SaaS platform and GxP data lake are helping companies develop a deeper and more comprehensive root cause analysis process.  The platform allows companies to review historical events (even if very recent) and look beyond any single siloed system.  In doing so, and with the ability to capture a variety of data, organizations are then able to develop a more comprehensive understanding of the operational footprint that was present in advance of those events.  Abildgren continued with the following example, “We offer the ability for companies to understand new relationships; how peak humidity in a media prep room can have a direct impact on new fermentation processes three days later, or next week.” 

Manufacturing intelligence is achieved when companies can turn the deep understanding of relationships into real-time predictive models.  Bigfinite has also set out to predict in real-time the risk of experiencing abnormal or undesired events.  “The combination of better root cause analysis and being able to have more comprehensive predictive models that run in real time, in front of the people closest to the process – I think is core to what we offer and how we differentiate ourselves,” added Abildgren.

Image Courtesy of Bigfinite

Use Cases

In an example offered by Abildgren, “If I’m looking at my PAT in my fermentation process, I’m only able to consider the very specific variables in a given unit operation that I have in my control system.”  With AI/ML     , companies are able to expand the variables, data points, and factors that can drive variation or that can have a direct impact on the performance of any given process – whether that’s the external environment, the conditions that are surrounding it, skills, or training of human beings.  A contributing factor may also be a combination of set points, CPP, CQA, and room conditions.  Bigfinite is helping organizations collectively gain a much deeper understanding of the factors and cumulative patterns that could possibly contribute to how well a process is running.

In another example, Novo Nordisk sought to retrieve and organize data from multiple sources.  The issue was that line was working less than the expected efficiency even though it was an automated system.  There was a lack of confidence in existing tools and data, coupled with manual efforts to manage data and print outs.  By applying the real-time analytics to gain deeper insights and understanding of the underlying issues, the company was able to save 3 hours of work per day, improved OEE stability with up to an 11% increase, as well as eliminated manual reporting.

Bigfinite doesn’t just look at Root Cause Analysis but also takes what is learned in that analysis to help with AI/ML models that will then help predict future failures. Abilgren described how Takeda was looking to understand the impact of variability in upstream processes and critical raw materials on the yield and quality of the final drug substance. Using Bigfinite’s platform, Takeda used the predictive insights to determine which bags of frozen unpurified bulk material to pool into downstream batches by automatically assigning bags for all runs to meet fulfill criteria. This includes a prediction of final protein concentration and yield as an indicator of goodness of each run. It significantly reduced typically manual effort involved in making pooling decisions and reduced the risk of suffering poor runs.

Another global biotech manufacturer had significantly reduced the number of re-circulations of their ultra-filtration step in their downstream purification process. Using an AI model to predict the optimal pump speed (considering a wide range of factors), the target concentration of a critical quality attribute for that unit operation could be achieved during the first pass through run, on average saving 3.4 recirculation runs. Getting it ‘right first time’ has direct operational gains and significantly increased the confidence of the process robustness.

In Brief

Companies have gathered and continue to gather vast amounts of data.  Bigfinite’s powerful GxP AI platform is helping organizations leverage manufacturing and process intelligence to gain operational efficiencies in a compliant manner.  In doing so, Life Science companies whose data has been gathering digital dust in siloed systems, can unlocked their data for actionable manufacturing intelligence.  Everything Bigfinite has built into the platform, on a server-less infrastructure, is aimed at the anticipated end-state of fully validated applications, including the AI/ML predictive algorithms. 

Based on the examples listed above, it can be realized that the platform does not aim to solve just one single problem, or the continuous process of verification, or real time release, or golden batch.  Rather, the platform can address a range of problem statements giving companies the foundation to focus on several of those, as the organizations mature and as they deploy the technology.  We will continue to monitor and report on future developments.

Leave a Reply