# Targeted probiotic tabletting: A hybrid active learning and finite element modelling approach for process optimisation

## TL;DR

This study combines machine learning and modeling to optimize probiotic tabletting processes, improving survival rates efficiently.

## Contribution

A hybrid approach using active learning and finite element modeling for rapid optimization of probiotic tabletting.

## Key findings

- The integrated approach achieved high prediction accuracy (R2 of 0.96) in 78 iterations.
- Survival rate maps revealed the relationship between process parameters and tablet performance.
- Optimal conditions were identified using global sampling and threshold filtering strategies.

## Abstract

Tablets are an efficient dosage form for delivering probiotics. Prior studies have identified compression pressure, compression speed, and precompression pressure as critical process parameters determining probiotic survival during tabletting. However, due to the labour-intensive and time-consuming nature of experimental investigations, most previous studies focused on evaluating the impact of individual parameters in isolation. Consequently, the rapid and systematic identification of optimal process parameters to maximise probiotic survival remains a significant and unresolved challenge in pharmaceutical formulation research. To address this gap, an integrated approach combining active learning (AL) based Gaussian process regression (GPR) with finite element (FE) modelling was developed to systematically explore the compaction parameter space and identify optimal process conditions. All data utilised in AL were generated using an FE model that was specifically developed to predict viability of probiotics during tabletting. Remarkably, the integrated approach achieved high prediction performance after only 78 iterations, demonstrating a coefficient of determination (R2) of 0.96 across the entire design space for predicting probiotic survival rate during tabletting. Using the well-trained model, a global random sampling strategy combined with threshold filtering was employed to identify regions of the design space likely to yield near-optimal survival rates. Furthermore, the exploration of compression speed and precompression pressure at selected fixed main compression pressures enabled the generation of survival rate maps, providing insights into the interplay between probiotic survival rate and tablet mechanical performance. This study demonstrated the potential of hybrid data-driven and first-principles modelling approaches as a robust strategy for optimising probiotic tabletting processes and accelerating pharmaceutical development.

Unlabelled Image