The latest crowdfunding project now running at Lifespan.io is an interesting initiative to apply machine learning methods to generate a cost-effective biomarker of aging in mice based on image analysis rather than physical samples. The biomarker will then be released for free and open use by research groups. There is a need for ways to easily assess the biological rather than chronological age of subjects in laboratory studies, where biological age reflects the current level of damage, dysfunction, and mortality risk. Aging is a process based on the accumulation of molecular damage and its consequences, but until fairly recently there were no useful tests to reliably assess the current state of aging - to obtain a better, more comprehensive, and repeatable measure than just looking at skin condition, or grip strength, or any of the other simple assessments used in the past. Such simple assessments show good correlations with mortality when assessed over a sizable study population, but have too much variability to be useful for the assessment of one individual.
Why does this matter? It matters because we are now entering the era of rejuvenation therapies, and there are a range of candidate treatments under development. Sooner is better when it comes to assessing the quality of various potential therapies, so as to discard less productive paths in favor of more productive paths. Unfortunately the only truly reliable test at the moment is to wait and see what happens following treatment, for as long as it takes to measure the resulting gains in life span. This requires years in mouse studies, and is clearly impractical in human trials. What if there was a simple, reliable test that researchers generally agree accurately reflects physical age, however? Then scientists could apply the test shortly before and shortly after a treatment, quickly obtain a result, and research and development would proceed much more rapidly.
A range of candidate biomarkers of aging are at various stages of development. Out in front are the DNA methlyation metrics, assessing characteristic changes in epigenetic markers that occur with age. A number of groups are taking a more algorithmic approach instead, attempting to find combinations of simple measures such as grip strength that when processed together can produce a more accurate result. There has also been some experimentation in visual identification of age in humans, proceeding alongside the increased interest in facial recognition technologies that characterizes this unpleasantly surveillance-fixated era of ours. It is plausible that this line of work might achieve the accuracy of other items in the present crop of biomarkers, say a margin of error of 5-10 years of biological age, given sufficient interest and investment. If you go digging through the literature, there is supporting evidence to suggest that facial appearance is, on balance, a decent reflection of age.
So if you can do this for humans, why not for mice? The potential payoff here, if it can be made to work, is the ability to skip over all of the equipment and work needed for physical biomarkers in favor of a hands-off camera and computer system. This might be the case for most of the biomarker assessment needed in exploratory studies in mice as they are currently carried out. Thus this is a potential road to greater automation and lower cost in studies of candidate rejuvenation therapies, though how that cost profile works out in practice is of course very dependent on the details. It is certainly the case the proving out the system, finding whether or not it can be made practical, is a comparatively cheap endeavor. The developers have experience with human facial assessment, computational power is cheap, and visual machine learning is a maturing field of software development. This seems worth a try to me.
Lifespan.io is launching a crowdfunding campaign to support MouseAGE, an application to assess visual biomarkers of aging in laboratory animals. This Artificial Intelligence-powered research tool, which is being developed by Youth Laboratories, will help scientists accurately determine the biological age of mice during experiments using advanced visual recognition and machine learning techniques. The project will help speed up research on rejuvenation therapeutics while collecting useful data in a more humane way.
When we are looking at other people, we can easily determine their ages and even get a rough idea of their health by looking at their skin tone, pigmentation and elasticity, their hair color, and their other characteristics. However, the human eye cannot accurately determine subtle changes in the appearance of such tiny animals as mice, and this is where MouseAGE can help. To rapidly collect data, commercial mouse breeders, research labs, and application beta testers all over the world will take and upload many mouse photos to the database. By using machine learning combined with visual recognition, MouseAGE will learn to recognize mice from images, to define their body parts, and finally to detect the subtle visual biomarkers of aging.
If successfully funded, the MouseAGE image collection tool will be available as a free mobile application by mid-October 2017. This will allow breeding houses and research institutions to begin collecting images and send them to the database. The project team hopes to collect enough data by February 2018 and will implement the algorithm for mouse age prediction by April 2018. This biomarker system will be made available as a free application shortly afterwards.
Here at MouseAGE we are aiming to create an artificial intelligence-powered research tool to help scientists accurately determine the biological age of mice and test longevity interventions based on photographic images of mice. This will introduce the first visual biomarker for aging in mice, and will help validate potential anti-aging interventions, save animal lives, and greatly speed up the pace of longevity research. By using machine learning combined with visual recognition, MouseAGE will learn to recognize mice from images, to define their body parts, and finally to detect the subtle visual biomarkers of aging.
We have chosen to start with the C57BL/6 (the black 6) mouse strain. This is the most common lab mouse globally, so it makes sense to begin here. Collected images at this stage will total approximately 10,000, including a wide age range of the black 6 mouse - this estimate based on our earlier experience with human faces. Once we have enough photographic material, artificial intelligence training will begin.
Our primary goal is to develop the MouseAGE system so that researchers can benefit from it. This will be an application that can be installed on a personal smartphone to make, annotate, and upload images to our cloud-based system for analysis, as well as be able to perform age assessment on newly taken images. The cost includes this data collection tool for researchers, mouse recognition software and the creation of an accurate, deep-learned, mouse age assessment algorithm. This will utilize feature extraction techniques to identify visual biomarkers of mouse aging, which we want to have thoroughly tested and made available for widespread use in common lab practice.