AI Model Improves Cell Culture Forecasting with Raman Data Fusion.

Biopharmaceutical manufacturing relies heavily on mammalian cell cultures, but maintaining optimal conditions is challenging due to parameter drift and sparse, irregular measurements. Traditional forecasting often fails to provide timely insights, leading to potential production issues. This new research introduces an adaptive framework designed to overcome these limitations. The framework combines a Gated Bottleneck Latent Ordinary Differential Equation (GB-Latent ODE) with Multi-Path Just-In-Time Fine Tuning (MP-JIT-FT). The GB-Latent ODE efficiently processes high-dimensional, sparse inputs, while MP-JIT-FT retrieves similar historical data to generate multiple plausible future scenarios, each with a confidence score, rather than a single average forecast. A key innovation is the fusion of Raman spectroscopy data, which acts as a machine-learning soft sensor to enrich sparse offline measurements. This data enrichment significantly improves training robustness. Experiments on bioreactor runs demonstrated that this combined approach outperforms baseline models, especially when early process behaviors diverge into different outcomes.