The Watson AI Lab Funds 10 COVID-19 Research Projects

May 21, 2020 by Robin Mitchell

MIT has taken a proactive approach to the COVID-19 crisis, developing medical devices, contact tracing apps, and now, AI-based solutions to provide relief.

To help meet the manifold challenges of the COVID-19 pandemic, the MIT-IBM team responsible for developing the Watson AI system is funding ten research projects that will utilize Watson to find data-driven solutions to the pandemic.

Developing solutions against viruses, in general, may no longer require human ingenuity alone; AI can develop solutions faster than ever before. Providing that there is a large amount of data for the AI to learn from, the results from such systems are often very accurate.


Rendering of MIT-IBM Watson AI Lab

Rendering of MIT-IBM Watson AI Lab in Cambridge, Massachusetts. Image used courtesy of the MIT-IBM Watson AI Lab

That is not to say that AI can't fail at times. In fact, in IBM's Watson for Oncology project, the supercomputer was reported to recommend "unsafe and incorrect" cancer treatments. However, according to IBM, the fault of the failure was IBM engineers using hypothetical patient data as opposed to real data.

If these errors can be corrected, AI can be a potent tool. That's why IBM is funding ten projects utilizing the Watson AI to help find COVID-19 solutions.


Watson AI Lab Supports 10 COVID-19 Research Projects

The ten projects MIT-IBM Watson AI Lab is funding include:

  1. Early detection of sepsis in COVID-19 patients
  2. Designing proteins to block the SARS-CoV-2 virus
  3. Saving lives while helping the economy
  4. Best material for facemasks
  5. Using repurposed drugs for fighting COVID-19
  6. Automated contact tracing with privacy
  7. Overcoming manufacturing and supply hurdles to create a global vaccine
  8. Using electronic medical records to find a treatment
  9. Better use of ventilators to free up resources
  10. Returning to normal with the use of targeted lockdowns, personalized treatment, and mass testing

While all of the projects above utilize AI to some degree in an attempt to find a solution, four of the projects, in particular, stand out: sepsis detection, protein design, facemask materials, and electronic medical records.


Sepsis Detection

Sepsis detection, as its name suggests, uses AI to identify patients that may be at risk of developing sepsis. One of the biggest issues with sepsis is that when the symptoms begin to show, the mortality rate jumps to 50%.

Therefore, if those who may develop it can be identified (with the use of data already available), then those individuals can receive treatment to prevent it from worsening. Interestingly, while the virus can only currently be cured by the immune system, secondary effects such as sepsis are treatable with antibiotics—but only if administered at an early stage. Thus, if the infection can be fought off long enough, then the body can develop antibodies to combat the COVID-19 virus. 


Protein Design

Protein design is another interesting path that may bolster the fight against the virus. The idea behind the project is that a bio-engineered protein may envelop a virus and prevent it from attacking human cells.



A deep neural network (center) assesses the molecules (left) and predicts whether they will hit the coronavirus protein (right). Image used courtesy of the Rafael Gomez-Bombarelli Lab

A team of MIT scientists have already used AI to discover a silk protein produced by honeybees that can coat perishable foods and extend their shelf life. Because protein folding techniques are computationally heavy, AI has been instrumental in finding fast solutions. 


Facemask Materials

Facemasks are a critical PPE used to protect both the wearer and those around them. But their current shortage means any available PPE must first be allocated to medical staff. While this may keep medical staff safe, the general public may be at risk of more exposure and spread of the virus. 

As a result, there has been a surge in individuals creating masks at home, but exactly how effective these are is currently unclear. Therefore, the Watson AI Lab is supporting a project that will explore the effectiveness of various homemade and medical-grade masks during normal breathing, coughing, and sneezing.

The project will test masks by many factors, including the environment the mask is in, what it is made of, and the size of the gaps between fibers in the mask.


Electronic Medical Records

One project will input millions of patient medical records into Watson AI in the hope of identifying potential treatments for the COVID-19 virus. The researchers will combine statistics with machine learning and simulated clinical drug trials to find and test drugs that have already been approved.

The researchers will target chronic illnesses such as hypertension, diabetes, and gastric influx, which are common causes of death in COVID-19 patients. This is a classic example of how AI can be used to correlate data and produce results that would often not be formulated by individuals when two sets of data appear to be unrelated or that there is far too much data to collate.


How Else Is MIT Addressing COVID-19 So Far?

MIT has taken a proactive approach to the COVID-19 crisis, developing medical devices, contact tracing apps, and now, AI-based solutions to provide relief.


Low-Cost Ventilators

Many engineering companies and individuals have been developing innovative solutions to help produce low-cost ventilators to address the shortage of such equipment. MIT is also joining the effort of doctors in diagnosing, disinfecting, and treating COVID-19.


MIT's E-Vent

MIT's E-Vent. Image used courtesy of MIT

Electrical engineering researchers at the institution raced to produce a low-cost ventilator design, which they made open source.


Contact Tracing App

Another solution that MIT has spearheaded is contact tracing apps, which use Bluetooth in smartphones to detect when two individuals are nearby, and upon detection, each device stores the others' unique ID.

MIT's solution is called PACT, private automated contact tracing. Should an individual fall ill to COVID-19, he or she can release their history of "chirps" or proximity to other anonymously-identified smartphones to healthcare authorities, who can then send instructions to those affected.


The PACT process is focused on privacy

The PACT process is focused on privacy. Screenshots (modified) used courtesy of MIT

From there, those who have made contact can then self-isolate to prevent potential spread (remember that many of those with COVID-19 can be asymptomatic for weeks). 



In what ways do you see AI as a tool for consequential healthcare crises such as COVID-19? What are the trade-offs of relying on machine learning? Share your thoughts in the comments below.