Just like in many other healthcare and medical research areas, experts in aging and longevity are now dealing with Big Data which is only growing in size. Scientists are increasingly more often using Artificial Intelligence to analyze this Big Data. There are plenty of large companies and small startups involved but very few manage to succeed.
How did aging fighters come to use AI and why can’t they do without it? Also, why is AI not enough?
The 2030s may become a turning point in the history of humankind. For the first time, the number of people aged over 65 in such developed countries as the United States may outnumber youth under 18.
AN inevitable shift in economic and political models will cause an enormous stress to welfare and social protection systems. Death risks as well as possibility of chronic diseases (i.e. type 2 diabetes, cancer, etc.) and related expenses balloon with age and double every eight years. This means that the result of almost any medical treatment will be very soon negated by an exponential growth of other age-related health risks. Even if tomorrow somebody invents a pill that can cure all types of cancer in one day, life expectancy will only increase by a few years. People who are now dying of cancer will live long enough to get Alzheimer’s or other diseases that are still incurable.
This is why it is important to fight aging altogether rather than individual health conditions. But again, it is an extremely complicated medical and engineering task.
Chronic diseases and aging are very slow processes. To understand the genesis of diseases and identify the risks depending on a wide variety of environmental factors, one would need to monitor a huge number of people over a long period of time.
Only recently has it become possible at all – now that there are government biobanks (for example, UK Biobank) and digital case records, and public health biomedical research results are available. Processing such Big Data is a challenging task as it is; it requires cutting-edge machine learning tools including the so-called deep learning, or AI.
AI can perform many important tasks. It helps search for biomarkers of health risks and of existing diseases. With a sufficiently large dataset supplemented by clinical cases, this approach works best and can be used to identify the molecular mechanisms of disease and aging; it can significantly reduce the development time of new drugs or help rearrange already known compounds for new applications. AI is used to accelerate clinical trials. Attempts are also being made to use AI to monitor human health.
Despite the hype, current AI technology remains weak and is often limited to dealing with narrow tasks. As Andrew Yang put it, if a person requires no more than one second to complete a task, this task can be completed by AI. Artificial intelligence is mostly used to automatically detect numerous complex regularities in noisy or even incomplete data. In any large but finite data array there are many consistent patterns, including random ones. That is why in order to distinguish between true and false associations AI requires much more data than the classic machine learning algorithms. This is especially true for cases when the explanatory effect is weak (such as the influence of one gene on a person’s height or IQ).
The requirement of large data during AI training is a separate obstacle for using it in medicine, because any use of personal data is followed by complex legal issues and is a constant object of interest for regulators.
Statistics knows that correlation does not mean cause-and-effect relationship. Therefore, another weakness of current AI is the difficulty in interpreting the received models. In this case, hybrid techniques that combine AI methods for automated study of the structure of big data arrays and the classic methods of physical modeling could be a solution here.
At Gero, we use machine learning methods, including AI, to analyze age-related biological and medical variables for detecting biomarkers of aging in humans and animals. Recently, we have used a deep convolution network that allows assessing risks of chronic diseases based on the data on physical activity recorded with a mobile phone’s accelerometer. Comprehensive machine learning with the use of blood analysis data obtained from 10,000 mice over their full life cycle allows assessing aging rate in laboratory animals.
Combined with genomic and transcriptomic data, AI helps formulate therapeutic hypotheses and instantly ‘pick’ molecules that can be used for making medications suitable for research in animals. For instance, we have developed glycolysis inhibitors – advanced neuroprotectors that have proved efficient for animal models of ischemic stroke.
Biomarkers of aging developed with AI allow us to promptly identify effective medications that can slow the aging process during animal testing and clinical studies.
AI will be further used extensively; the main thing is to use it wisely and mix it with other advanced technologies. AI is only a technology that is fully available only for those who have access to talents, sustainable finance, and, necessarily, big data.
By Peter Fedichev, Founder and Research Director of Gero, Director of the Biological System Modeling Lab, Moscow Institute of Physics and Technology