Neural Networks Can Identify Any Person’s Age
An international group of scientists which included researchers from ITMO University has developed an algorithm that can determine a person’s age, based on their nationality and sex, using an online blood test. Data from more than 120,000 people from Canada, South Korea, and Eastern Europe has been used to identify the key ageing indicators of these populations. Then, they taught neural networks to account for the relevance of particular indicators. This contributed to improving the technique’s precision: for now, the standard deviation is less than six years.
Science still can’t offer any solution against ageing. Still, some researchers believe that this would become possible in the near future. One of the important steps to approaching this issue is to find a reliable technique for determining one’s biological age. In the absence of such an instrument, telling whether some solution is effective becomes impossible. Also, the instrument has to be affordable and not too complex. Scientists from ITMO’s Computer Technologies international laboratory decided to apply neural networks to solve this problem.
They’ve used machine learning to create an algorithm that makes it possible to determine one’s age based on a set of blood tests. For now, the system can only tell one’s chronological age; still, being able to make connections between results of certain blood tests and age is an important step to identifying biological age. The algorithm makes use of 20 biomarkers: from glucose or haemoglobin concentration to red blood cell count. All of them are easy to get and are included into standard blood tests in most countries. The Aging.Ai service is already available online and can be used by just anyone with Internet access and recent blood test results.
“The distribution of biomarkers for determining age differs in every population, shares Kirill Kochetov, one of the algorithm’s developers. We’ve created a neural network that cancels them out, and can be used for any population. At the same time, it is a lot more stable, and makes fewer mistakes. We’ve also identified the key biomarkers that have a greater effect on ageing. Albumin and glucose are among those. This data correlates with what has already been known about ageing-induced change in these indications”
“For the most part of our work, we’ve been translating the data for different populations to a common format, adds Evgeny Putin, Kirill’s co-author. We had to give up on using some of the biomarkers, and restore others. To teach the neural networks, we used overall data on all of the populations, and also data on each population in particular. Then we checked the precision of our solution on external, independent data. It turned out that the network that learned from the overall data on all populations made 10% less mistakes than those that learned from data on specific populations”
This research was conducted in collaboration with the Insilico Medicine company which focuses on using artificial intelligence for studying ageing and the development of new drugs. In future, the scientists plan to use a larger variety of biomarkers so as to increase the technique’s precision.
Reference: «Population specific biomarkers of human aging: a big data study using South Korean, Canadian and Eastern European patient population». P. Mamoshina, K. Kochetov, E. Putin et al. The Journals of Gerontology: Series A Jan. 11, 2018