image: This is the Aging.AI logo. view more
Credit: Insilico Medicine
Summary:
- Deep learning methods are propagating into biomarker discovery and aging research
- This system may provide insight into the biological age of the person if the person “looks” older or younger to Aging.AI then his/her chronological age- Inspired by Microsoft’s How-Old.net, Insilico Medicine scientists created Aging.AI, which guesses patient’s age using basic and inexpensive blood tests
- An Ensemble of Deep Neural Networks achieved 83.5% accuracy within a 10-year frame (r = 0.91 with R2 = 0.82 and MAE = 5.55 years) when guessing chronological age outperforming many other available markers of aging
- Insilico Medicine’s Pharma.AI division is soon to publish a range of drug and nutraceutical predictions called geroprotectors, where organismal and tissue-specific efficacy is predicted using a system trained on multiple data types
May 19, 2016, Baltimore, MD - Insilico Medicine, Inc announced that a paper titled "Deep Biomarkers of Human Aging: Application of Deep Neural Networks to Biomarker Development” by Putin, et al, was accepted for publication by Aging, one of the highest-impact journals in aging research on 9th of May, 2016 and today became available online as advance publication at http://www.impactaging.com/papers/v8/n5/full/100968.html.
"It is exciting to see the power of deep learning applied to potential aging biomarkers. The availability of such markers is an essential prerequisite for any future clinical trials to try to ameliorate the effects of human aging", said Charles Cantor, PhD, CSO of Agena, Inc, former director of the Human Genome Project (DOE).
The availability of big data coupled with advances in highly-parallel high-performance computing led to a renaissance in artificial neural networks resulting in trained algorithms surpassing human performance in image and voice recognition, autonomous driving and many other tasks. However, the adoption of deep learning in biomedicine and especially in the pharmaceutical industry has been reasonably slow. In order to outperform more traditional machine learning methods, deep neural nets require large amounts of data and expertise with highly-parallel and high-performance graphics processing unit (GPU) computing.
Evgeny Putin, lead author on the paper commented: "While internally we are working on more sophisticated machine learning problems, Aging.AI is a good example, where DNNs outperform other machine learning methods and can be extended into multiple applications".
Insilico Medicine is working on over a dozen different applications of deep learning methods to regenerative medicine, embryonic development, cross-species comparison and drug discovery and repurposing providing contract research services and developing a range of molecules for cancer, metabolic and CNS pathologies.
"I am happy to work in a very dedicated team, which has significant domain expertise in aging research and is working on grand projects, while solving smaller problems and publishing these in peer-reviewed journals. We want to minimize animal testing and simulate many biological processes in silico", said Putin, deep learning lead at Insilico Medicine, Inc.
To develop a data set of blood biochemistry and cell count samples Insilico Medicine collaborated with the largest independent laboratory test service provider in Eastern Europe, Invitro Laboratories. Scientists of both companies went through over a million samples to select a data set, with the optimal number of features from patients that came for routine blood checkups to build a data set of just over sixty thousand samples. Using this data set Insilco Medicine scientists then trained 40 different deep neural networks (DNNs) of different depth with a single neuron output predicting chronological age and optimized using different optimizers and started organizing these DNNs into an ensemble. Experimentally, 21 DNNs providing optimal performance were organized into an ensemble using a stacking model.
Using the best performing DNN in an ensemble scientists identified most important features contributing to the accuracy of predicting human chronological age: albumin, glucose, alkaline phosphatase, urea and erythrocytes. This finding may be relevant for further studies in biomarkers of aging.
"Inspired by Microsoft's How-old.net, which can recognize your age using a photograph, an approach we also employ in projects with skincare collaborators, we decided to train an ensemble of deep neural networks on a very large number of simple inexpensive historical blood tests linked to age and sex and built a predictor, which is scalable and can include many other data types to build more comprehensive biomarkers of aging. Aging.AI can in principle be extended as a biomarker of biological aging that can be used to assess the efficacy of various therapies", said Poly Mamoshina, research scientist at the Pharma.AI department of Insilico Medicine, Inc.
###
About Insilico Medicine
Insilico Medicine, Inc. is a bioinformatics company located at the Emerging Technology Centers at the Johns Hopkins University Eastern campus in Baltimore with R&D resources in Belgium, Russia and Poland hiring talent through hackathons and competitions. It utilizes advances in genomics, big data analysis and deep learning for in silico drug discovery and drug repurposing for aging and age-related diseases. The company pursues internal drug discovery programs in cancer, Parkinson's, Alzheimer's, sarcopenia and geroprotector discovery. Through its Pharma.AI division the company provides advanced machine learning services to biotechnology, pharmaceutical and skin care companies. Brief company video: http://bg-rf.us7.list-manage.com/track/click?u=ea13c3b22fca5092d074cccc6&id=201bc90dc4&e=7b8fcba750