Pre-packed columns have been increasingly used in process development due to their ease of use in biomanufacturing compared with in-house packing procedures. The packing quality of 25000 pre-packed columns manufactured over a period of ten years was analyzed. Rather than using traditional statistical methods, machine learning was employed thanks to its superiority in handling complex problems with multiples features. In particular, XGBoost was executed in a Python platform to predict packing quality (HETP and asymmetry) as a function of quantitative parameters (column length, column diameter, particle size) and qualitative attributes (functional mode, and backbone). The ML model offered excellent predictive capabilities for HETP and asymmetry (90% and 93%, respectively), with packing quality strongly correlated with column backbone and functional mode. The results highlight machine learning’s ability to provide reliable predictions of column performance from simple column parameters, offering a quantitative alternative to first principal models in chromatography. In addition, our results will guide further efforts in column optimization, e.g. by focusing on improvements of backbone material and functional mode to lead to optimal packing.

Analyzing pre-packed columns with machine learning: influence of column backbone and functional mode on packing quality

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.