Predicting drug-microbiome interaction with machine learning

Written by Laura McCoubrey

In a recent review published in Biotechnology Advances, we reveal how machine learning can be best applied to predict interactions occurring between drugs and the microbiome.

The human microbiome

The human microbiome is becoming notorious for its wide-reaching interactions with medicines. Possible interactions include metabolism of drugs by microbiota, resulting in changes to drug potency or pharmacological activity, and drugs affecting the growth of microbiota, potentially increasing patients’ risk of developing diseases. Predicting such interactions can be very difficult, as not all drugs have been experimentally tested in microbiome studies and microbiome composition varies widely between individuals. Drug-microorganism interactions that occur in one individual may occur to a significantly lesser extent in another individual, or may not occur at all. To add another layer of complexity, microbiome composition is known to naturally change throughout life, meaning that drug-microbiome interactions could vary over time even within individuals.

Harnessing machine learning

Predicting and assessing drug-microbiome interactions is important, as their presence can affect the efficacy of therapeutics, increase patients’ susceptibility to certain diseases, and account for significant inter- and intra-individual pharmacokinetic variability. In our review, we present how machine learning (a powerful form of artificial intelligence) can be used to accurately predict drug-microbiome interactions. We highlight the typical machine learning workflow, specific techniques, and existing applications within the field. We also examine the challenges of utilizing machine learning for such tasks and the steps the technology should take to achieve validation for use in clinical and industrial settings.

Author insights

Laura McCoubrey, the first co-author of the review, describes the publication as “... a useful guide for those interested in predicting all forms of drug-microbiome interactions. As medicine becomes more and more personalized, it is imperative that we know how medicines behave within individuals. The mechanisms behind drug-microbiota interactions can be exceedingly complex and thus very difficult to predict on a case-by-case basis. The time and resources needed to experimentally validate such interactions prohibit their everyday use in drug development pipelines or the clinic. Here, machine learning is an ideal technology to take the big data associated with interactions and apply it to individual cases. How amazing would it be if healthcare professionals could one day generate personalized predictions for drug-microbiome interactions at the bedside?”

The review is available to read here.