Artificial Intelligence transforms search for COVID-19 vaccines, cure
Click Here to Manage Email Alerts
Developers are using artificial intelligence to help researchers comb through thousands of studies and find relevant COVID-19 information for the development of potential treatments and vaccines.
For example, a team at Lawrence Berkeley National Laboratory in California built covidscholar.org, a publicly available model that provides “research that's relevant to the work researchers are doing and helps them discover hidden connections between thousands of research results from the past and present,” according to Amalie Trewartha, postdoctoral fellow and one of the model’s developers.
She said that within 15 minutes of a COVID-19 paper appearing online, it will be on covidscholar.org.
Brian Uzzi leads efforts at Northwestern University to develop another AI model that predicts which COVID-19 studies' results are most likely to be replicable.
“If study results don’t replicate, expensive drug tests fail and the pace of breakthrough slows down, which are particular issues now that the COVID crisis has revealed the need to create therapies and vaccines quickly,” he told Healio Primary Care. “Our analysis provides the backbone for creating new drug therapies and vaccines as fast as possible.”
Thusfar, Northwestern University’s AI model showed 78% accuracy in ascertaining which study findings could be replicated, Uzzi added.
Shaun Grannis, MD, vice president of data and analytics at Regenstrief Institute in Indiana, told Healio Primary Care that his institution is gearing up to use AI to “sift through the large volume of data flowing from hundreds of sources to uncover emerging patters beyond what we already know.”
As states begin to relax travel restrictions and social distancing, Grannis said that he wants to use AI “to monitor for emerging populations and subcohorts at risk” for COVID-19.
Quality of data
Edmondo Robinson, MD, chief digital innovation officer at Moffitt Cancer Center in Florida, applauded Berkeley’s effort but cautioned that there may be limitations to using their AI model for COVID-19 research.
“This approach may not stand up to the rigor of more established, albeit slower, scientific methods for assimilating data across multiple published studies, namely systematic review and when possible, meta-analysis,” he told Healio Primary Care.
Andrew Boyd, MD, who oversees AI models at the University of Illinois at Chicago that help determine the severity of a patient’s condition and their risk for intubation, said that using AI models like Northwestern’s present other possible challenges.
“Will it help us correct concepts faster or will they help us perpetuate old ideas for a longer period of time?” he said.
Trewartha, Uzzi and Grannis said they have taken steps to ensure that their respective AI tools gather sound, valid data.
“We have machine learning models that are classifying new research as it comes in for quality, relevance and to sort into topics,” Trewartha said. “We also have built-in tools on our site to let users easily flag papers that have errors or are not relevant.”
Uzzi added, “we examined [our] model for baked-in human biases in the data that were used to train the model and we found no biases were operating.”
Grannis said researchers at the Regenstrief Institute use “a variety of techniques” to ensure the quality of the data, including testing algorithms using cross validation, controlling for confounders like race, ethnicity, age and gender, and asking experts with clinical and biomedical expertise to review the algorithm output. – by Janel Miller
References:
Berkeley Lab. https://newscenter.lbl.gov/2020/04/28/machine-learning-tool-could-provide-unexpected-scientific-insights-into-covid-19/. Machine learning tool could provide unexpected scientific insights into COVID-19. Accessed May 6, 2020.
Northwestern University. AI speeds up search for COVID-19 treatments and vaccines. https://news.northwestern.edu/stories/2020/05/ai-tool-speeds-up-search-for-covid-19-treatments-and-vaccines/. Accessed May 6, 2020.
Disclosures: Boyd reports receiving funding from the Agency for Healthcare Research and Quality, the NIH and the National Science Foundation, and speaking fees from several organizations. Grannis reports receiving research funding from the Agency for Healthcare Research and Quality, the CDC and the NIH. Robinson, Trewartha and Uzzi report no relevant financial disclosures.