278: Your Bioprocess Data Already Holds 35% More Yield: From End-to-End Models to Digital Twins with Ignasi Bofarull-Manzano - Part 2

How do you take a model that works in process development and get it accepted for use in GMP manufacturing? That question stalls most bioprocess modeling projects before they start. Ignasi Bofarull-Manzano, Senior Data Scientist and CMC Consultant at Körber Pharma, pushes back on the premise: the process you run today is already governed by a mathematical model, fitted once at small scale during process characterization and then left untouched for years, even as the process shifts.Part 1 separated digital models from digital shadows and digital twins, and made the case for starting with the decision rather than the data. Part 2 goes into the plant: what regulators actually require, what the numbers looked like on a real biologics process, and where a team should start on Monday morning.Topics covered:Core differences—and surprising similarities—between modeling in development versus manufacturing (02:35)Regulatory requirements: credibility assessments, model risk, and validation steps for digital twins (05:07)Real-world example: How deploying an end-to-end process model led to 35% yield increase for Takeda, and considerations for ROI in manufacturing (08:34)Advice for startup leaders on when to invest in modeling and how to scale efforts case-by-case (11:42)Steps for scientists new to modeling: identifying bottlenecks, starting simple, and proving value offline before scaling up (12:26)The importance of understanding basic statistics before relying on AI-generated models (15:22)A stepwise summary for deploying digital modeling effectively in biotech (16:01)Smart insight: The digital twin is the last step, not the first. Identify the bottleneck, build the simplest model that supports the decision, and concatenate it end to end so you can see how a parameter moves final drug substance quality rather than one unit operation's output. Prove the value offline. Only then connect interfaces, because that is where the cost and the validation burden live. Teams that lead with the twin arrive at the C-level with a proof of concept and no evidence. Teams that lead with the offline model arrive with a number.Before a digital twin can earn its keep, you need connected data, the right model, and a clear decision for it to support. These four episodes cover that ground — data silos, hybrid and mechanistic modeling, and twins built to survive regulatory scrutiny.Episodes 215 - 216: From Data Silos to Autonomous Biomanufacturing: Digital Twins and AI-Driven Scale-Up with Ilya BurkovEpisodes 05 - 06: Hybrid Modeling: The Key to Smarter Bioprocessing with Michael SokolovEpisodes 17 - 18: How Extracting Gold From Your Data Accelerates Process Development with Ioscani Jiménez del ValEpisodes 263 - 264: Why AI and Automation Tools Won't Deliver Until Your Lab's Data Is Connected with David HardyConnect with Ignasi Bofarull-Manzano:LinkedIn: www.linkedin.com/in/ignasi-bofarullKörber Pharma website: www.koerber-pharma.comSupport the show