An artificially intelligent computer has taught itself how to optimally interpret the vast quantities of DNA data microbiologists extract from billions of gut bacteria.
Within the past 10 years, researchers have been intrigued to discover how strongly gut bacteria affect our health.
The gut microbes influence our physical and mental health in many ways, and an unhealthy composition of gut bacteria is associated with the development of diseases such as type 2 diabetes, chronic intestinal inflammation, obesity, depression, Parkinson’s and Alzheimer’s.
Within the same 10 years, researchers have also become more adept at extracting DNA from gut bacteria so that they can painstakingly assemble the genetic codes one by one and understand them better. However, in the process, they have also become painfully aware that mathematical models are lacking to process and interpret the information in billions and billions of tiny fragments of DNA to identify the intestinal bacteria of a given individual.
However, Danish researchers have begun to solve this problem, enabling much better insight into the genomes of the bacteria that can promote either health or illness.
“Some mathematical models have been useful in providing insight into the thousands of bacterial species living in our gut, but this has only been the tip of the iceberg. We have determined a fraction of the bacteria and obtained insight into only part of their genomes. Our new mathematical model will enable us to obtain insight into far more bacteria and their DNA,” explains a researcher behind the new model, Simon Rasmussen, Associate Professor, Novo Nordisk Foundation Center for Protein Research, University of Copenhagen.
The model and the research have been published in Nature Biotechnology.
Assembling 1,000 jigsaw puzzles from a million random pieces
The new research tried to determine which bacteria live in our gut and how they affect our health.
Our gut microbiome comprises about 1,000 species, and 1 gram of faeces contains 100 billion bacteria.
To determine how these bacteria affect our health, the researchers sequenced their genomes and studied their DNA to identify which proteins they can make and thereby better understand their function in the gut.
However, an individual species cannot be cultured separately in the laboratory. Instead, the research team had to sequence the total DNA of the gut microbiome and then try to determine which DNA fragments belong to which bacteria.
This is like taking apart one thousand 1,000-piece jigsaw puzzles, throwing the million pieces randomly into a box and then trying to assemble all the puzzles without knowing what they should look like. Some pieces can be assembled into parts of some jigsaw puzzles, but many pieces inevitably cannot be placed.
“The ecosystem of the gut microbiome is extremely varied and complex, like walking through a forest and meeting 1,000 animals. We have to use machine learning to analyse data from the ecosystem’s total genome to obtain better insight, but this is easier said than done,” says Simon Rasmussen.
Computer teaching itself how to assemble the DNA puzzle
To carry out the task, Simon Rasmussen and colleagues resorted to artificial intelligence, which can learn to find patterns in massive data sets in two ways.
· Supervised learning provides examples of solutions the artificial intelligence program can use to learn.
· Unsupervised learning involves the computer program learning to interpret data and design its own rules.
The new model Simon Rasmussen and colleagues developed uses unsupervised learning to navigate the map of the gut microbiome.
“We feed the data into the artificial intelligence program and then it determines how to interpret the data so they make the most sense. The programmer does not limit the model, and it can detect associations that people cannot. This makes assembling the DNA jigsaw puzzles easier. The new model has enabled us to obtain insight into about 400 of the 1,000 species of gut bacteria versus 100 previously. We can also analyse 90% of the bacterial DNA versus only 50–60% previously,” explains Simon Rasmussen.
Cancer treatment requires specific gut bacteria
Simon Rasmussen’s research group not only works with data scientists but also tries to use knowledge from the artificial intelligence programs to better understand how gut bacteria are associated with the development of diseases such as type 2 diabetes and chronic intestinal inflammation.
People with these diseases often lack several bacterial species in their gut, and these species might be identified and supplemented in the future to cure or alleviate the diseases.
This also applies to other diseases. Simon Rasmussen emphasizes research published in August 2020 showing that anticancer immunotherapy requires specific gut bacteria that train the immunotherapy to recognize cancer cells.
“We will also explore whether we need to ensure that people with cancer have the required gut bacteria so that immunotherapy is effective. There are also perspectives in helping people with mental disorders. Hundreds of researchers have already used our new model to learn more about the bacteria on which they focus,” says Simon Rasmussen.