DeepMind, the artificial intelligence (AI) arm of Google, has joined the global research community studying the novel coronavirus, COVID-19.
DeepMind is best known for its AI that easily defeated the world's best Go and StarCraft II players. The research laboratory is currently using its system to help researchers fight the epidemic.
To study a virus and develop a vaccine, scientists must first understand how it functions, namely the structure of viral proteins. This is a lengthy process that takes months and may not always yield results. Scientists have turned to computer predictions using a deep learning system known as AlphaFold.
Work on the coronavirus is underway in laboratories around the world. DeepMind hopes to help these studies by "releasing structural predictions for several little-known proteins associated with SARS-CoV-2, the virus that causes COVID-19." The system uses a machine learning method without environmental modeling, which can be used to predict protein structures in the absence of similar protein structures.
DeepMind hopes to save scientists the months it usually takes to determine the protein structure of the virus. “Knowing the structure of a protein provides an important resource for understanding how it functions, but experiments to determine the structure can take months or more,” says the company's official blog.
Given the "potential severity and time frame," DeepMind said it is going to skip the experimental validation process or wait for peer review by the academic community prior to publication. This is in line with other academic research on the topic, which appears in both peer-reviewed journals and non-peer-reviewed preprints, as the process can take months.
“We emphasize that these structural predictions have not been experimentally tested, but we hope they can contribute to the scientific community on how the virus works and serve as a platform for generating hypotheses for future experimental work on the development of therapeutic agents.” said in a blog post.
The team notes that the data provided “is not a primary focus of current therapeutic activity,” but may help a general understanding. “It is important to note that our structure prediction system is still in development and we cannot be sure of the accuracy of the structures we provide, although we are confident that the system is more accurate than our previous CASP13 system. We have confirmed that our system provides an accurate prediction for the experimentally determined SARS-CoV-2 structure stored in the Protein Data Bank, giving us confidence that our model predictions for other proteins can be useful,”the researchers said.
An open license will allow any researcher to develop, adapt or share the results of DeepMind research. Google acquired London-based research organization DeepMind for £ 400 million back in 2014. The company has previously used AI for health research, developing models for identifying eye diseases and detecting neck cancer.
Alibaba is also doing coronavirus research. For example, researchers from a Chinese corporation announced the development of a machine learning algorithm that can detect pneumonia caused by the new coronavirus COVID-19 with an accuracy of 96%, distinguishing it from inflammations of a different nature. According to the Nikkei Asian Review, the analysis will require a CT scan of the patient's chest. After analyzing the image for 20 seconds, the system gives an answer - the doctor would need numerous images and at least 15 minutes of time.
The algorithm has been trained on 5,000 images of the lungs of patients with confirmed coronavirus infection and is already being used in at least 100 hospitals across China.