The biopharmaceutical industry is increasingly focused on shortening development timelines to accelerate progression into clinical stages, thereby improving patient outcomes and boosting market competitiveness. Dielectric spectroscopy, a promising process analytical technology (PAT), offers enhanced monitoring, and control of mammalian cell culture bioprocesses, but its practical application is constrained by the limited accuracy of existing models due to sparse historical data within compressed development periods. Just-in-time modeling (JITM) has recently emerged as a soft sensing approach balancing model accuracy and generalizability. However, its application in dielectric spectroscopy remains unreported, mainly due to the absence of suitable similarity metrics. This study firstly proposes a novel dual-similarity driven JITM framework tailored for dielectric spectroscopy-based modeling. Various similarity strategies were evaluated through validation study, then the generalizability of JITM was demonstrated using multiple local model algorithms, while its effectiveness was further confirmed using a testing application involving real-time predictions of VCD. Compared to traditional techniques, the proposed JITM framework exhibited reliable and accurate VCD monitoring across diverse cell lines and culture conditions, showing great potential for industrial applications, especially in agile biomanufacturing where historical data and development time are limited.
Chen et al. (2026) studied this question.