Repetition is a basic concept in machine learning and other forms of AI, the notion that something that happens now is a function of something similar that happened a moment ago, or even a long time ago.
Just so, what you are now is likely to be a function of the things you consumed, such as nutrients, yesterday, the day before, and for months or years, a fact that the Twin Health startup is using to learn machinery to try to break the code in medical terms including diabetes.
“We are tackling the root cause of chronic metabolic diseases, including diabetes, hypertension, obesity, a group that affects one billion people and causes twenty-five million deaths a year,” CEO and co-founder Jahangir Mohammed told ZDNet in an interview through Zoom. “Hasht an incredible carnage.”
Twin Health, a three-year startup with offices in Mountain View, California and Bangalore, India, announced Wednesday that it has received $ 140 million in Round C funding led by investment firm Iconiq Growth to further commercialize its approach of learning machinery to recommend to physicians and patients regimens that improve diabetes.
The new money brings Twin Health’s total shipping to $ 186 million. Other investors in this round were venture capital firm Sequoia Capital, Perceptive Advisors, Corner Ventures, LTS Investments, Helena and Sofina. Twin Health has a cash estimate of $ 740 million.
Also: The subtle art of really big data: Recursion Pharma designs the body
Twin Health, Mohammed said, is using a simulation, combined with repetitive neural network, to make predictions that lead to dietary recommendations for patients with diabetes.
“It was not really possible to solve it programmatically [metabolic disease] the root cause, “said Mohammed.” What we have invented is a new technology called a full body digital twin that enables us to help solve the root cause. ”
The digital twin is a common term in engineering these days, where a computer simulation of the features of an object, such as a gas pipeline or an airplane, is created. Used to perform experiments on that simulated object, noting its features, before trying those actions on the real world object.
In the case of humans, Twin Health is building digital metabolism twins of each subject that can then be the subject of tests by neural networks.
Subjects using Twin Health technology have wearable sensors that send information to Twin Health, along with an app they use for food diaries and quarterly blood tests. All of those data points, 3,000 signals per day, are used to build the digital twin of the entire body of the individual.
A neural network approach is then applied to the digital twin as an optimization problem, the goal, what is known in machine learning as an objective function, is to predict the absolute mean error of blood sugar as an output of all signals.
“We are using a number of different machine learning algorithms, including, in particular, repetitive neural networks, RNN, because the data we use is data from high-resolution time series,” said the chief technology officer. and co-founder Terry Poon, in the same interview with Muhammad.
Also: MIT deep learning found an antibiotic for a microbe that nothing else can kill
“What we learn using patterns are really patterns in all of these different signals,” Poon said. For example, heart rate data and activity data can be combined with blood glucose levels.
“What we see is that the pattern of heartbeats is different depending on the blood glucose level.” Someone who is diabetic will have a “troubled” pattern of basal recovery after exercise, versus a person without diabetes, whose heart rate returns to normal very soon after exercise.
With the mean absolute error of blood sugar as the objective function, RNNs allow Twin Health to simulate interventions, such as changing diet or changing sleep patterns, and see how those interventions affect blood sugar levels. . A key, Mohammed said, is modeling in that twin the kind of interventions that matter to the specific individual, a kind of personalized medical regimen, in other words.
“It’s not just if you eat this food, what is the result,” Poon said. “More more given your metabolic background and everything you have done so far, then, if you eat this food, what will be the answer.”
The sensors provide a feedback loop for the digital twin as the person receives recommendations, in consultation with an attending physician, such as changing diet or changing sleep or exercise.
Understanding what the objective function should be and what to measure in neural network development has been a process of working with scientists to do by displaying engineering based on what is already known. However, several factors and factor relationships are found during neural network development, Poon said.
Repetition means there is a more complex signal to be processed than some other machine learning subjects, he said.
“Especially a lot of time series features: eat a food, get a blood sugar response, but the relationship is very complex it is very complex in that it is not just the meal you just ate, it is really affected by what has happened before “Such as if you had a big meal last night that caused a very large rise in blood sugar, even when you eat breakfast today, it will have an impact on that.”
“It’s really a really complex relationship of time series actions and behaviors over long periods of time.”
The technology has already been used by thousands of subjects over the past two and a half years, collecting data that has helped refine neural network patterns, Poon said. Those thousands of people become the context for every young person whose twin is created.
Because the interventions that are promoted are regimens, such as diet and exercise, and not chemicals, the approach does not require regulatory approval and has already been undertaken in clinical settings.
“Our early clinical results are very good,” Mohammed said. “We are seeing from the early results 90% of people with diabetes change” and “90% of people give up diabetes medications”. In addition, he said, people who use the regimen have lost 9 pounds of body weight and have seen improvement in liver function, with liver enzymes “ALT” showing measurable improvement.
The company has published controlled clinical trial results in numerous publications this year, including the 2021 Annual Meeting of the American Society of Endocrinology; ISPOR 2021 meeting of the International Society for Pharmacoeconomics; European and International Congress on Obesity; Endocrine Society Magazine, Vol.5, April-May of this year; and in the June issue of the American Diabetes Association magazine dIABETESwith
Also: AI Ethics: Benefits and Risks of Artificial Intelligence
In those studies with patients using the regimen, the company found statistically significant reductions in HbA1c, the amount of hemoglobin-related blood sugar, which it said served as “an early indicator for translating scientific rationale for technological intervention through digitalization.” twin technology, enabled by the Internet of Things and Artificial Intelligence, as a modality to enable the reversal of diabetes. ”
“A very interesting discovery for us is how beautifully elastic our body is,” Mohammed told ZDNet, “and how we do these natural things, if you do the right things, how you heal.”
One implication may be that just as solutions are deciphered by the twin working back over the time series, the medicine for metabolic disorders can be thought of in itself as a sequence of making the right movements in time.
“We have been really surprised over the last three years how beautifully elastic our body is if given just one chance to do these natural things in the right pattern and the right combination, it has an extraordinary ability to heal itself. . ”
Twin Health is already operating commercially, selling its services to patients using the regimen.
Twin Health is doing a business outside of technology by getting paid for the results, Mohammed said. “We are demonstrating a significant cost savings,” Mohammed said. The cost of diabetes amounts to $ 3 trillion a year, including a significant cost of medication, Mohammed noted.
“The beauty of improving health and reversing disease is that it saves such a cost savings in terms of medicines as well as in terms of hospitalizations and visits.”
“Twin Health gets paid for the performance,” he said. “We agree on the results we will give to the patient, when they achieve it, you understand the benefit and pay us for what we have contributed.”
Consumers appreciate that approach because “they are difficult results,” he said, “you do not have to speculate.”
Also: AI encounters a major data problem in the diagnosis of COVID-19
The new money will allow the company to expand the clinical use of the technology globally, Mohammed said. The company declined to say how many current members it has, but noted that they are excited by their continued growth.
“We are really ready to bring this to a wider population.” This will include the staff of the teams responsible for going to market, training and caring for patients.
Money also includes constantly improving the digital twin in order to go beyond diabetes to approach other metabolic disorders. This includes arthritis and abnormal kidney function.
“This is an ongoing optimization approach,” Mohammed said. “Our body is an ocean, it never ends up about how much we can understand about it.”