Active Machine learning for formulation of precision probiotics
It is becoming clear that the human gut microbiome is critical to health and well-being, with increasing evidence demonstrating that dysbiosis can promote disease. Increasingly, precision probiotics are being investigated as investigational drug products for restoration of healthy microbiome balance.
Highlights
• Active machine learning was applied to formulation and microbiome science for the first time.
• Leveraging a small dataset of 6 probiotic-excipient interactions, the effects of a further 111 pharmaceutical excipients on probiotic proliferation were predicted.
• Uncertainty sampling was used to obtain a final machine learning model certainty of 67.70%
• Experimental validation found that the effects of 3/4 tested excipients could be correctly predicted.
• Feature importance analysis via random forest and principal component analysis found the most influential chemical features determining excipients’ effects on probiotic proliferation.
To reach the distal gut alive where the density of microbiota is highest, oral probiotics should be protected from harsh conditions during transit through the stomach and small intestines. At present, few probiotic formulations are designed with this delivery strategy in mind. This study employs an emerging machine learning (ML) technique, known as active ML, to predict how excipients at pharmaceutically relevant concentrations affect the intestinal proliferation of a common probiotic,Lactobacillus paracasei.
Starting with a labelled dataset of just 6 bacteria-excipient interactions, active ML was able to predict the effects of a further 111 excipients using uncertainty sampling. The average certainty of the final model was 67.70% and experimental validation demonstrated that 3/4 excipient-probiotic interactions could be correctly predicted. The model can be used to enable superior probiotic delivery to maximise proliferation in vivo and marks the first use of active ML in microbiome science.
About this article: Laura E. McCoubrey, Nidhi Seegobin, Moe Elbadawi, Yiling Hu, Mine Orlu, Simon Gaisford, Abdul W. Basit, Active Machine learning for formulation of precision probiotics, International Journal of Pharmaceutics, Volume 616, 2022, 121568, ISSN 0378-5173, https://doi.org/10.1016/j.ijpharm.2022.121568. (https://www.sciencedirect.com/science/article/pii/S0378517322001223)