An algorithm trained to predict age based on blood tests
February 18, 2016 – BALTIMORE. Insilico Medicine launched aging.AI, a system allowing users to guess their age and gender by entering the results of their blood test. The system will not be used for medical purposes at this time and is focused on gamification of consumer blood testing and attracting the attention of the general public to the importance of periodic blood tests.
The renaissance in deep learning, led by the teams of Geoffrey Hinton, Yann LeCun, Andrew Ng and several other thought leaders, is revolutionizing many areas of information technology in the same way that CRISPR/CAS9 is revolutionizing biomedicine. However, the propagation of deep learning into biomedicine has been slow. Since 2014, Insilico Medicine has been hard at work harmonizing millions of gene expression samples and developing deep-learned transcriptomic biomarkers of cancer, aging, and other diseases for clients worldwide, in many cases achieving performance superior to other machine learning methods. Aging.AI is not a transcriptomic marker, it is a side project of the company’s collaboration with leading diagnostic centers providing fully-anonymized blood biochemistry and cell-count test data linked to age and gender of patients that are presumably free of life-threatening diseases other than aging. Almost a million samples were used to train an ensemble of deep neural networks to predict age and gender of the patient and deployed as a web-based tool, which can be used for entertainment purposes by the customers of diagnostic clinics and make blood testing more fun.
“Deep learning is revolutionizing machine vision and many other fields, but very few groups are exploring its power to extend healthy productive longevity. Aging research is the most altruistic cause resulting in the largest number of quality-adjusted life years (QALY) per dollar spent and maximizing the net present value (NPV) of human life. What social networks, software companies and banks may not understand is that their value is the equation of the NPV of each individual user and now it is possible to apply artificial intelligence expertise to extending productive longevity of the user base. We encourage experts in machine learning to work with our team to significantly accelerate progress in applied human aging research,” said Alex Zhavoronkov, PhD, CEO of Insilico Medicine.
Some of the highlights of 2015 included the development of deep learning systems trained on the NVIDIAR DIGITS™DevBox achieving high levels of accuracy in recognizing images, translating speech, autonomous driving and several other fields.
“We hope that via machine learning, discoveries with biological and cheminformatics data in drug discovery may sprout fruit over the next few years to come. The amount of annotated genomic, transcriptomic, metabolomic and other human data is reaching the levels sufficient for deep neural networks to possibly outperform other machine learning methods. These methods may be useful to move into basic classification tasks into drug repurposing, drug discovery, biomarker development and possibly even aging research,” said Mark Berger, Senior Alliance Manager, Life & Material Sciences, NVIDIA Corporation.
Aging is a disease and our target is to find ways to treat it or even cure it. Advances in deep learning and multi-omics integration will help us find actionable markers of aging in humans and develop novel interventions to extend healthy longevity.
“The launch of Microsoft’s How-Old.net, which can predict the age of the person by the photograph, inspired us to develop a consumer-friendly system to guess the patient’s age by simple blood biochemistry. Our company is quite good at developing similar tissue-specific biomarkers trained on large number of transcriptomic data sets in order to predict the geroprotective efficacy of multiple anti-aging, CNS, metabolic and anti-cancer drugs. But here we stepped outside our core competence. Users may find it interesting to see if the predicted age changes after a certain diet, exercise routine or drug regimen prescribed by their physician. It will be interesting to combine this marker with Beauty.AI, RYNKL and other projects being developed to analyze age-related changes and the effectiveness of anti-aging interventions in the future. We now have many collaborations with IVF clinics and cosmetic and nutrition companies that may result in more comprehensive biomarkers of aging, longevity, attractiveness and mortality,” said Polina Mamoshina, research scientist, Insilico Medicine, Inc, involved in the project.
One of the main collaborators on the Aging.AI project is Invitro Laboratories, Inc., the largest independent diagnostic company in Eastern Europe. It offers a broad range of diagnostic services and has one of the most advanced electronic record management systems in the region.
“Blood tests can help detect problems before these problems turn into pathologies. Blood can tell a lot about the person and it is a true science turning the anonymized statistical data into life-saving interventions and to encourage people to learn about the many components of their blood. INVITRO supports many medical scientific projects, from regular students’ research to breakthrough technologies of 3D-Bioprinting solutions. We also expect that our anonymized data will become a good base for Aging.AI success. We share very much the Aging.AI approach to make blood testing both educational and fun,” said Alexander Ostrovskiy, Chairman of the Board of INVITRO Laboratories.
Source: Insilico Medicine