Researchers claim cutting-edge algorithm can predict biological age to better measure interventions that extend lifespan.
The observable differences between the old and the young are as intuitive to spot as they are rude to detail. But the biomarkers of aging, those that give true “biological age” distinct from chronological age, are much trickier to find. Scientists have searched for appropriate biomarkers for decades, hoping to use them to better measure interventions that extend lifespan, but it has only been in recent years that their efforts have begun to bear fruit. The latest in these have come from the AI healthcare company Insilico who, working at John Hopkins University, used their machine-learning algorithm to accurately predict the age of individuals to within a few years. All they need is a quick peak inside your gut.
Revealing their findings in preprint form on the site bioRxiv [1], the researchers are looking for feedback before they enter the peer review process. Writing in the paper they claim that they have “developed a method of predicting [the] biological age of the host based on the microbiological profiles of gut microbiota” as well an “approach [that] has allowed us to define two lists of 95 intestinal biomarkers of human aging.”
So how did they do it? First they collected 3,663 gut bacteria samples from 10 publicly available data sets, containing age metadata, before analyzing their samples using a machine-learning algorithm. The samples originated from 1,165 individuals between the ages of 20 and 90, and came from nine distinct countries, including the US, Kazakhstan, China and Germany. After dividing their samples into three sets according to age (‘young’, ‘middle-aged’ and ‘old’) they separated them further by randomly categorising 90% into a training set and 10% into a validation set. They first refined their deep learning regression algorithm on the data in the training set, pulling in markers from 95 different types of bacteria in the gut microbiome.
After refining their algorithm, they used it to predict the ages of the 10% in the validation set. They found that their program could accurately place an individual’s age to within four years of the chronological figure, as well as determining the best 39 of the 95 bacteria types to predict it with. They found that some of these bacteria, like Eubacterium halli, increase with age, while others, like Bacteroides vulgatus, decrease.
Their method, if it passes peer-review, will represent a significant advance in our understanding of the gut microbiome, a part of anatomy of increasing interest to scientists studying inflammatory bowel disease, arthritis, autism, and obesity. It would also be a welcome addition to a growing roster of nascent biomarker approaches that could measure aging, like the one at MouseAge.org, who are currently devising a pilot algorithm to accurately predict the age of mice from photographic evidence alone. Insilico are closely involved in this project, alongside Aubrey De Grey, Jim Mellon and Professor Steve Horvath (the discoverer of the methylation-based biomarker tool dubbed the “Horvath clock”).
Further research will be needed before it can be developed into a diagnostic method robust enough for use in patients, but if successful it could be used to more accurately determine how the microbiomes of sick individuals differ from a healthy average, as well as indicating personalised interventions and drugs to make them better. “Age is such an important parameter in all kinds of diseases. Every second we change,” Dr Zhavoronkov of Insilico told Science [2]. “You don’t need to wait until people die to conduct longevity experiments.”
[1] https://www.biorxiv.org/content/10.1101/507780v1
[2] https://bit.ly/31LFIOX
