The Problem
At a leading global biotechnology manufacturing site, renowned for producing a high-value rare disease medicine, the team faced a significant challenge in optimizing the yield from their downstream bioprocess using historical data. The site produces a rare disease medicine where the value per gram can be hundreds of thousands of dollars, making it important to optimize for every gram per batch. Despite the site's accolade as a 'Facility of the Year' with advanced data infrastructure, their reliance on manual data collection into spreadsheets presented a major bottleneck. This labor-intensive process required extensive effort to organize information for downstream purification production runs, undermining efficiency and confidence in batch quality.
Using a cutting-edge perfusion (continuous) upstream process, the team harvested unpurified bulk material daily into bags, storing them in deep freeze for later pooling into 5-6 run campaigns for downstream purification. The challenge was to optimally select bags from inventory for each run to maximize protein yield without depleting high-quality bags early, which could affect later yields. The goal was to optimize predicted yield per run and in aggregate for the campaign while fulfilling the ‘hard requirements’ per the control strategy.
The Solution
Our client leveraged Aizon’s award-winning and cloud-based platform to seamlessly integrate and securely store data from five critical sources, including material genealogy, electronic batch records, process data (CPPs), ERP planning data, and LIMS data (CQAs); the pooling solution provided web-based access to all the relevant information in a single spot, including the ability to simulate various 'what-if' scenarios.
Utilizing principal component analysis, Aizon distilled thousands of variables down to 36 key factors that significantly influenced variations in bag content. This refined data set served as the foundation for an advanced predictive AI model, trained on LIMS data from 104 different downstream runs, each verified for its relevance and impact on the predictive accuracy.
The deployment of this model in the AI-powered, GxP-qualified platform marked a pivotal shift towards precision and efficiency. The model's ability to predict expected yields for individual runs, as well as for entire campaigns, transformed inventory management. Operators gained real-time insights into inventory status with the flexibility to adjust bag allocations on the fly.
The Results
The result was a staggering 93% reduction in the time required for pooling strategies, shortening the process of pulling data together and trying to extract how to combine the bags from multiple days to a matter of minutes. This efficiency leap empowered the operations team to approach pooling with newfound confidence and significantly less manual effort in their pooling strategies. Quality assurance teams also benefited from this streamlined process, as the heightened predictability and reliability of the pooling strategies facilitated quicker, more confident approvals.
