Novel AI-Based Aging Clock Predicts Human Age with Unprecedented Accuracy
We’ve arrived at the point in history where artificial intelligence (AI) can imitate the inner workings of the human brain. Algorithms known as deep learning can process data for detecting objects, recognizing speech, translating languages, and making decisions. This AI has recently been in the mainstream limelight for its role in fueling self-driving cars and generating viral deep fake videos.
Now, deep learning is being used to approximate biological age — how old your cells are, which is looking to be a better predictor of your real age than your chronological one. The Hong Kong-based company Deep Longevity recently announced the development of DeepMAge, which may be the most accurate aging clock to date. DeepMAge can help develop anti-aging interventions or to help aging-conscious people understand and, potentially, affect their pace of aging.
The disruptive potential of AI solutions in science
Deep learning algorithms are best known for their achievements in text, sound, and image processing. Meanwhile, the AI instances developed for research purposes are much less known to the public. This, however, does not indicate a lack of progress or relevance. On the contrary, state-of-the-art deep learning models in biology, chemistry, and medicine could potentially disrupt the healthcare and pharmaceutical industries.
For example, in pharmacology, deep learning methods can be employed to streamline drug design. One deep neural network generator was used to discover a small molecule inhibitor whose beneficial properties in live animals were established within three weeks after the computer-based experiment was launched. Using traditional iterative design methods, the same target-to-hit stage of drug design would take months or even years.
Deep learning models already perform similarly to trained professionals in the differential diagnosis of brain diseases or to measure blood flow parameters based on magnetic resonance imaging (MRI) scans. Clinical AI systems such as these can be used to significantly reduce exam times and to quantify health risks, ultimately increasing the throughput and cost-effectiveness of a healthcare system.
What do blood, microbiomes, and DNA modifications have in common?
In the study of biological aging, all sorts of data types have been analyzed to create a number of novel age predictors. For example, age predictions made by looking at measurements related to blood have been shown to be associated with mortality risk. This particular aging clock uses standard blood parameters measured during a typical check-up, such as glucose, cholesterol, and platelet count.
Today, a variety of machine learning approaches have been tested for the purpose of predicting human age based on molecular-level features. Among these, deep learning is an especially promising approach that has been used to construct accurate clocks using data on blood biochemistry or the microbiome—feats unachieved by other algorithms.
Many researchers have built age prediction models with high accuracy based on age-dependent changes in certain DNA locations using sequencing data. A growing number of studies have shown that changes in the patterns of modifications to a person’s DNA throughout their lifetime exhibit a strong correlation with age and age-related outcomes.
The first aging clocks based on sequencing data obtained from modifications at specific DNA locations date back to 2013. These first studies looked at the levels of a modification called DNA methylation at tens or hundreds of DNA locations called CpG sites. There have been other implementations using the same concept. Together, they show that there are multiple sets of DNA locations that can be used to achieve comparable accuracy to predict biological age.
DeepMAge: A DNA methylation aging clock developed with deep learning
Deep Longevity, a Hong-Kong based longevity startup, recently published its take on the aging clock in the Aging and Disease journal. Their algorithm DeepMAge is the first deep learning aging clock based on DNA methylation. It was trained to predict human age on more than 6,000 DNA methylation sites. By analyzing the methylation patterns, DeepMAge can estimate human age within a 3-year error margin, which is more accurate than any other human aging clock.
DeepMAge can also assign a higher predicted age to people with various health-related conditions, such as ovarian cancer, irritable bowel diseases, and multiple sclerosis. For example, DeepMAge produced significantly higher age predictions (by 1.23 years on average) for people with irritable bowel diseases compared to healthy people.
The genes where the DeepMAge CpGs are located are enriched with those taking part in developmental processes, especially for the heart and nervous system. Understanding the interactions in which the identified genes take part during the aging process can be useful for rejuvenation research.
Gazing into the biological aging crystal ball
This study shows that deep learning algorithms can be used to explore individual DNA methylation landscapes in the context of aging, and they can potentially be used to estimate the risk of certain age-related events in the future. Further research is required to study the reproducibility and robustness of DeepMAge in independent longitudinal studies.
Yet, no matter how fancy or futuristic the algorithm, it likely will never be foolproof and capable of pinpointing exactly how much time we have left to live. So, in the words of the prolific Danish writer Hans Christian Andersen, best remembered for his fairy tales, “Enjoy life. There's plenty of time to be dead.”