21 September 2020

I see right through

Are those who look young young?

Polina Loseva, N+1

Looking at old photos often turns into an optical illusion: parents at the same age as us often seem older to us, and Keanu Reeves, on the contrary, looks eternally young. Is it worth believing your eyes and is there always an equally young insides hiding behind a young exterior? We tell you what the science of aging thinks about this, and explain what benefits a neural network trained on everyday wisdom can bring.

My own biomarker

For those who today decide to ask the eternal question "How long do I have left to live?", modern statistics have no more accurate answers than the forest cuckoo. The average life expectancy in the world or in a single country has little to do with the period that is "measured" to each individual person. This figure does not even mean that most of his neighbors will die at this age – perhaps half will live ten years less, and the other ten years more. In search of a more reliable way to predict how many years a person has ahead, scientists come to the idea of a "real", biological age – as opposed to a "passport", chronological one – which should reflect what is happening inside the body and how long this body will last.

However, it is still unclear by what signs it should be measured. There were many suggestions – from the length of telomeres to a set of intestinal microbes – but none of these markers of biological age turned out to be universal. In other words, each of them works well in his patrimony (for example, it is convenient to predict heart problems and the risk of dying from them by the concentration of a number of substances in the blood), but is blind to other organs and systems.

Meanwhile, a system for detecting biological age is already embedded in the brain of each of us. We can determine how old a person we see for the first time is, literally by eye – and then, comparing our assessment with his passport data, conclude that he looks "unusually young" or "older than his years."


To Ian Reeves does not change with age.

The accuracy of such an assessment (it is called perceived age, perceived age) for each individual may be small. But if you take a few dozen third-party observers, you can achieve good results: the arithmetic mean of their estimates allows you to determine the "passport" age with an accuracy of a year or two and fix the "outliers" – people who do not outwardly correspond to this age. No special training seems to be needed for this: at least in some works, the age of seventy-year-olds was equally well detected by their peers, geriatric nurses and young male students.

What's in your head

The method of determining biological age "by eye" has an important drawback – no one knows how it works. We can argue for a long time about the methods of calculating the length of telomeres, but at least we know exactly what we are measuring. In the case of perceived age, we are dealing with a certain image of a 70- or 50-year-old person, which people construct in their head automatically, scanning hundreds and dozens of faces every day. And if we ever decide to use the "trained eye" of an ordinary person as a biomarker to make medical predictions based on his testimony, it would be good to understand what signs people rely on in their visual assessments.

Apparently, there are a lot of signs. Plastic surgeons, for example, distinguish five main anatomical changes characteristic of an aging face: nasolabial wrinkles and wrinkles on the bridge of the nose, weakening of the zygomatic ligaments, sagging adipose tissue around the eyes and on the chin.


Young and old face, colored circles indicate problem areas from the point of view of a plastic surgeon (Cotofana et al. / Facial Plast Surg).

But ordinary people seem to use completely different signs: in different studies, an independent assessment of age depends not only on the number of wrinkles, but also on age spots, the color and density of hair and the thickness of lips. And in order to influence this assessment, it is enough to remove the contrast in the photo – pale faces get at least an extra year plus. At the same time, sometimes most of these signs turn out to be completely superfluous: even a cropped frame of the left cheek, where neither eyes, lips, nor hair are visible, is quite enough for the assessment to be as accurate as for a full-fledged photo.


Contrast zones on a person's face (on the left). A face with artificially reduced contrast (on the right) appears older than the original (in the middle). (Porcheron et al. / PLOS ONE).

Therefore, despite the abundance of research, the "trained eye" remains a black box for scientists. It is still unclear which facial features we pay attention to and which of them we take into account first of all when assessing age. If we add to this the influence of personal sympathies (attractive faces usually seem younger to people) and the confusing factor of emotions (the stronger they are, the lower the accuracy of the assessment), then we have to admit that the perceived age is not suitable for real biomarkers yet.


Exemplary emotions with which experimenters confuse independent evaluators. People with positive emotions usually seem younger. (Voelkle et al. / Psychology and Aging).

What's behind the facade

In the meantime, some scientists are trying to understand the principle of operation of the "trained eye", others are interested in whether there is any biological meaning in his observations. And some relationships can really be detected. For example, it turned out that by the number of wrinkles one can judge the number of senescent ("old") cells in the skin, and by the perceived age – the concentration of glucose in the blood (it usually grows over time) and insulin-like growth factor-1 (but it often disappears). And every next 0.1 micromole of cortisol (stress hormone) per liter of blood adds 0.42 years to the subjects in the eyes of others. Thus, the appearance of a person in some way reflects his "true age" (and Keanu Reeves from this point of view can really be considered younger than many of his peers).

But with other markers of biological age, the perceived age does not always fit together: its predictions sometimes coincide with the length of telomeres, but not with epigenetic clocks, even if they are measured in the same blood cells as the length of telomeres. This, in general, is not surprising, especially considering that other biomarkers have also not yet been reconciled with each other. But if it is already clear for the length of telomeres or epigenetic clocks what aging processes they display, then the situation with perceived age is much more complicated.

We don't fully know what exactly makes some people look younger or older than their peers. To some extent, genes are certainly to blame for this: at least in identical (genetically identical) twins, the spread in perceived age is usually less than in fraternal (who are only half the same). For each specific trait that independent observers use in their assessments, it is possible to calculate how strongly it is genetically determined. In some cases, the dependence is quite significant – for example, for the thickness of the lips and the appearance of gray hair. However, for sure we know only about one gene that is associated with perceived age – this is MC1R (Melanocortin 1 receptor), the synthesis of melanin in skin cells depends on it. Two copies of the "aging" variant of this gene add up to two years to carriers when viewed from the outside.


Two pairs of twins, in each of which one of the sisters looks younger (left column). On top are identical twins, they are 67 (estimated at 64 and 70), the bottom row is fraternal, they are 69 (estimated at 64 and 76). (Gunn et al. / PLOS ONE).

In addition to genes, the environment, of course, also affects a person's appearance. According to scientists, more than 36.5 percent of the spread in the difference between chronological and perceived age can be attributed to external factors, including not only skin care (cleansing and moisturizing), but also lifestyle (sports, regular brushing of teeth, frequency of visits to the doctor), the number of relatives living in the same apartment and even the level of education.

It turns out that the perceived age reflects everything at once: successfully or unsuccessfully inherited gene variants, successfully or unsuccessfully chosen lifestyle and general health. This can explain the phenomenon of "old parents", which at the end of 2019 attracted the attention of Twitter users: in old photos, parents seem much older than their children at the same age. Even if the children inherited "aging" gene variants, they did not necessarily preserve the parental lifestyle. Researchers who use other biological markers (based on the state of internal organs) also note that the new generation is aging more slowly than the previous one – not least due to the fact that they are increasingly quitting smoking and monitoring the intake of prescribed medications.


Twitter users with the pitch of baseball star Brandon McCarthy share photos of relatives to find out if people really looked older in the past.

In addition, each next generation brings with it new standards of perception. To test this, Danish scientists invited people to evaluate several hundred pictures of elderly people in 2002, and then repeated their request in 2012 (some of the evaluators participated in the experiment, and someone came for the first time). By calculating the average perceived age scores for each image, they found that in 2012, people needed to look 2.3 years younger to be rated the same as 2002. In other words, the standard has shifted: people are used to the fact that others live longer, and those who would have been given 70 years earlier are already given 72.3 in ten years. This figure almost coincides with changes in life expectancy – in ten years in Denmark it has increased by 2.6 years. Shifts in the perception of age reflect the slowing down of aging in the population – thereby showing that biological truth can be hidden in the "worldly wisdom".


Now neural networks have come to the aid of the "trained eye". Recently, an article was published in the journal Nature Metabolism, the authors of which first taught the neural network to predict the chronological age of a person from a photograph, and then "introduced" it to estimates of the perceived age for the same set of images. The average accuracy of predictions in the first and second variants turned out to be similar – about four years (which is comparable to the accuracy of epigenetic clocks), and in most cases deviations from the average coincided – that is, people whom neural networks considered significantly older or younger than their passport age also seemed older or younger to a person.

Next, the researchers decided to see how strongly the deviations of the predicted age from the chronological correlate with different physiological parameters and lifestyle features (smoking, alcohol consumption, and so on). It turned out that the neural network, which is trained on both chronological age and perceived age, works here more accurately than anyone else – that is, the deviations predicted by it from the passport age correlate best with deviations from the average state of health. This neural network overtook not only the one that was trained only on chronological age data, but also the one that took into account only the perceived age. Thus, the authors of the article managed to create an algorithm whose eye is more trained than that of an ordinary person, but operates according to the same principles.


The relationship between the predicted age and lifestyle features, on the left – positive (people look older), on the right – negative. The thicker the line, the stronger the connection. Prediction algorithms, from top to bottom: chronological age (neural network), perceived age (neural network), chronological age, chronological age (other model). (Xia et al. / Nature Metabolism).

This is not the first time that a neural network is being used to predict age. This is usually done when it is impossible to "manually" isolate the only reliable biomarker from a complex system. For example, with the help of a neural network, it was possible to find out that among the huge microbial community of the intestine, some bacteria serve as a more reliable marker than others.

Now the neural network has also reached the wisdom of everyday life – and so far it is coping better than its direct carriers. We can hope that this union will bear fruit and we will finally find out exactly which facial features are worth paying attention to in order to determine for sure how old or young the person standing in front of us is really old or young at heart. In the meantime, it remains only to admire the way gerontologists have traveled in search of biological markers: from complex and incomprehensible like epigenetic clocks to simple, obvious and lying, literally, on the surface.

Portal "Eternal youth" http://vechnayamolodost.ru

Found a typo? Select it and press ctrl + enter Print version