By John P. Desmond, AI Trends Editor
AI is being used on multiple fronts to combat the coronavirus (COVID-19), including for monitoring the spread, finding effective drugs, developing therapies and customizing clothing wearables embedded with useful intelligence. Here is a survey of ongoing efforts:
A team at Boston Children’s Hospital is using machine learning to scour social posts, news reports, data from public health channels and information supplied by doctors, for warning signs of how the virus is moving, according to an account from abcNews.
“Incredible data is locked away in various tools like online news sites, social media, crowdsourcing, data sources, that you wouldn’t think of that would be used for public health,” stated Dr. John Brownstein, chief innovation officer at Boston Children’s Hospital. “But actually they have incredible amounts of information that you wouldn’t find in any sort of traditional government system.”
Brownstein’s HealthMap team at BCS is looking for social posts that mention specific symptoms indicating respiratory problems and fever, from geographic areas where doctors have reported potential cases. The result is a map that represents a live tracking of progression of the virus.
“Whether it’s social media, online news reports, blogs, chat rooms—we’re looking for clues about symptoms, reports of disease, that tell us something unique is happening,” stated Brownstein.
Australia Team Using Predictive Analytics
A team in Australia is using AI with predictive analytics to predict the number of deaths and confirmed cases in weeks going forward.
“To develop contingency plans and hopefully head off the worst effects of the coronavirus, governments need to be able to anticipate the future course of the outbreak,” stated the team, which includes Belal Alsinglawi, a PhD candidate in data science at Western Sydney University; Mahmoud Elkhodr, a lecturer in Information and Communication Technologies at CQ University Australia; and Omar Mubin, a senior lecturer in human-centered computing and human-computer interaction at Western Sydney University.
The team is predicting that by March 31, the number of deaths worldwide will surpass 4,500 and confirmed cases will reach 150,000. They noted the model is best suited for short-term forecasting, given the limited (but building) historical data and limited understanding of the variables determining the rate of COVID-19 spread. The work was reported in NewsHub of New Zealand.
To create its simulations, the team extracted coronavirus data going back to Jan. 22, from an online repository provided by the Johns Hopkins University Center for Systems Science and Engineering. The data is time-stamped and indicates the number and locations of confirmed cases of COVID-19, including people who recovered and those who died.
The team used time-series forecasting to predict future values based on previously-observed values, a technique proven to be suitable to predict the outbreak of disease. Simulations were run via Prophet, a time series forecasting model; data was input using the Python programming language. The predictions were combined with location-based services such as GPS tracking, to provide insight to help governments implement contingency plans to slow the virus’s spread.
“The virus’ spread is influenced by many factors, including speed of diagnoses, government response, population density, quality of public healthcare and local climate,” the team stated. They expect different regions will see different growth rates of COVID-19 as a result.
Private Industry Responding Too
Private industry is responding to the threat in force as well. BlueDot maintains it recognized the significance of higher rates of pneumonia in China nine days before the World Health Organization, according to an account in VentureBeat. BlueDot was founded in response to the SARS epidemic. It uses natural language processing to skim thousands of sources making statements about the health of humans or animals.
A deep learning model created by researchers from Renmin Hospital of Wuhan University, China, Wuhan EndoAngel Medical Technology Company, and China University of Geosciences, is detecting COVID-19 with what they claim is 95% accuracy. The model is trained with CT scans of 51 patients with laboratory-confirmed COVID-19 pneumonia and more than 45,000 anonymized CT scan images.
The deep learning model showed a performance comparable to expert radiologists. “It holds great potential to relieve the pressure on front line radiologists, improve early diagnosis, isolation, and treatment, and thus contribute to the control of the epidemic,” the team stated in a “preprint paper” (not yet reviewed) about the model, published in medrxiv.org. The team reports the model can speed confirmation of cases from CT scans by 65%. In another effort at Zhongnan Hospital in Wuhan, machine learning from Infervision is training on hundreds of thousands of CT scans to help in detection.
The prediction of survival rates of those with confirmed cases of COVID-19 is happening using clinical data from Tongji Hospital in Wuhan with more than 90% accuracy, according to researchers with the School of Artificial Intelligence and Automation, and other departments from Huazhong University of Science and Technology in China. The team stated that it can draw from more than 300 lab or clinical results.
In a paper entitled “Deep Learning for Coronavirus Screening,” released last month on arXiv by collaborators working with the Chinese government, researchers describe a model that uses multiple convolutional neural net (CNN) models to classify CT image datasets in calculating infection probability. Preliminary results suggest the model is able to predict the difference between COVID-19, influenza-A viral pneumonia, and healthy cases with 86.7% accuracy.
This deep learning model is trained with CT scans of influenza patients, COVID-19 patients, and healthy people from three hospitals in Wuhan, including 219 images from 110 patients with COVID-19.
Efforts to Find a Cure Include a Drug Discovery Competition
To help find a cure, a Coronavirus Drug Discovery Competition was launched by Sage Health, a healthcare company in San Francisco, offering $3,500 in prizes to the top three finishers. They were: Matt O’Connor (Hong Kong, China), Thomas MacDougall (Montreal, Canada), and Tinka Vedovic (Zagreb, Croatia).
O’Connor’s winning AI algorithm identified a compound called Remdesivir as the most promising treatment for COVID-2019, due to its high-scoring inhibitory potential when docked against the Coronavirus main protease enzyme. Remdesivir was recently shown to be effective in treating the first US patient infected with COVID-2019; it is currently undergoing clinical trials to gain FDA approval.
Sage Health Co-Founder Dr. John Billings stated, “These results have shown that this process is no longer limited to just multi-billion dollar pharmaceuticals since many of the competitors were independent developers. The rise of free and public data, algorithms, and education is democratizing Drug Discovery like never before.”
Sage Health said it would donate the top-scoring compound Remdesivir to the Wuhan Institute of Virology for further analysis.
Another collaboration to accelerate the discovery and development of novel antiviral therapies, including for COVID-19 and influenza, was announced by SRI International of Menlo Park, Calif., and Iktos of Paris. The companies plan to combine SynFini, SRI’s automated synthetic chemistry platform, with AI technology from Iktos to accelerate design and production of target molecules.
The Iktos AI technology is based on deep generative models, helping to automatically design virtual novel molecules that have all the desirable characteristics of a novel drug candidate. This is said to allow for rapid and iterative identification of molecules which validate multiple bioactive attributes and drug-like criteria for clinical testing.
“Iktos’ generative AI technology has proven its value and potential to accelerate drug discovery programs in multiple collaborations with renowned pharmaceutical companies. We are eager to apply it to SRI’s endonuclease (enzyme) program, and hope our collaboration can make a difference and speed up the identification of promising new therapeutic options for the treatment of COVID-19,” stated Yann Gaston-Mathé, co-founder and CEO of Iktos.
Yann Gaston-Mathé, co-founder and CEO of Iktos
The SRI SynFini platform is designed to accelerate chemical discovery and development, with the aim of bringing new drugs to the clinic more quickly and affordably. The closed-loop SynFini system automates the design, reaction screening and optimization (RSO), and production of target molecules.
“The SynFini system has the potential to dramatically expedite small molecule drug discovery,” stated Nathan Collins, Ph.D., chief strategy officer of SRI’s Biosciences Division and head of the SynFini program. “We look forward to exploring how the integration of Iktos’ AI-driven generative molecule combined with SynFini supports the rapid and efficient discovery of new drugs to treat emerging infectious diseases.”
Biofourmis of Boston at Work in Hong Kong
A remote monitoring and disease surveillance program is being instituted in Hong Kong using wearable AI technology from Biofourmis of Boston, on patients with diagnosed or suspected COVID-19. The program is being administered by The University of Hong Kong and includes Hong Kong-based Harmony Medical, a Biofourmis’ joint venture partner for the China region.
Diagnosed and potentially COVID-19-infected patients are being monitored with the Biofourmis’ Biovitals Sentinel platform, a solution the company built specifically for this initiative. The product customized the FDA-cleared, AI-powered Biovitals Analytics platform. Biovitals Sentinel’s 24/7 remote monitoring technology and analytics are providing clinicians involved in the COVID-19 program with clinical decision support for early identification of any physiological changes that could indicate deterioration, and to enable earlier interventions for better outcomes.
“The goal of this program is to remotely monitor patients and identify COVID-19-related physiological biomarkers that indicate deterioration in patients,” stated Prof. David Chung Wah Siu, MD, Department of Medicine, The University of Hong Kong. “We hope our combined efforts also will rapidly lead to a better epidemiological understanding of COVID-19 so we can improve the outcomes of our citizens—as well as the global community—as more people become infected.”
The company’s Everion sensor is worn on the arm of patients quarantined in homes or clinical settings. Analytics will derive more than 20 physiological signals from the sensor data, including temperature, heart rate, blood pulse wave, heart rate variability, respiration rate, inter-beat-interval and others. These signals are then fed through advanced AI and machine-learning techniques to flag key physiological changes that could indicate disease progression.
Kuldeep Singh Rajput, CEO of Biofourmis
“The sooner these biomarkers associated with COVID-19 deterioration are identified, the sooner healthcare providers can intervene and prevent a serious medical issue,” said Kuldeep Singh Rajput, CEO of Biofourmis. “We currently know the common symptoms, but we are still learning how this strain of the coronavirus affects the body. This program will be a key step in achieving this important goal. When a pandemic such as COVID-19 spreads and so much is unknown, every second counts.”
Read the source articles and view references at abcNews, NewsHub, VentureBeat, medrxiv.org, Infervision, Deep Learning for Coronavirus Screening on arXiv, Sage Health, SRI International, Iktos and Biofourmis.