This AI-based tool called "GutBug" provides information about the specific bacterial enzymes, reactions, and bacteria involved in the process of digestion and absorption of nutrients by the human gut.
The human gut microbiome is made of over a thousand different species of bacteria, which have more than 3.3 million unique genes. These bacteria secrete enzymes to process what a human eats and provide the body with various metabolites that are essential for health and body functions.
However, studying the complex host-microbial relationships is a challenge because of the vastness of the microbiome and the fact that the collection of bacteria varies among individuals.
But the tool GutBug, detailed in the Journal of Molecular Biology, can potentially predict all possible bacterial enzymes that act on bioactive dietary molecules as well as oral drugs, the researchers said.
"GutBug uses a combination of machine learning, neural networks, and chemoinformatic methods. We used a curated database of 363,872 enzymes from about 700 human gut bacterial strains and a substrate database consisting of 3,457 enzymes to train the AI model," said Dr Vineet K. Sharma, IISER Bhopal, in a statement.
"GutBug can help us better understand how the food we eat or medicines we consume orally are processed by our gut bacteria and how this affects our health. Such understanding can be useful in designing diets, developing new prebiotics, creating nutraceutical products, and improving drug design depending on the nature of the gut bacteria each individual has, leading to personalised medicine," Sharma said.
To train the AI model the study used the largest enzyme substrate database. It contained information on the gut bacterial species from eight populations to understand how drug and food metabolism varies across populations due to gut bacteria activity.
GutBug was able to identify the specific enzymes involved in the breakdown of various molecules and the bacterial strains that carry them.
The tool was tested with 27 different molecules, including complex carbohydrates, flavonoids, and drugs, and was shown to be highly accurate, with success rates ranging from 0.78 to 0.97.
Also read: