It is noteworthy that a system using AI technology was the first to detect the coronavirus epidemic. That was done when it was still localized in Wuhan. The system was able to show that the outbreak could become a global pandemic. Even though a large body of data regarding diseases has been amassed over decades (due to the findings of on-going biomedical research), there have been only a few months of studies on CoViD-19 (so named because it was first identified in the last days of 2019). Using conventional methods, this may pose a problem. However…AI can expedite what is normally a protracted process.
As it turns out, “smart” technology can use available information to track down other viruses with similar elements and see how they function. Such systems can then work out which drugs could be used to inhibit the virus. For example, DeepMind (the AI arm of Google’s parent company, Alphabet) has already used data on genomes to predict the protein structure of various microbes. The system is now being used to analyze the structure of the novel corona-virus; and thereby determine which drugs could work effectively against COVID-19.
DeepMind created a deep-learning library called “AlphaFold”, which uses neural networks to predict how the proteins that constitute biomes fold. It models such folding based on their genome. This is extremely useful, as proteins determine the shape of receptors in an organism’s cells. Once you know what shape the receptor is, it becomes possible to work out which drugs could bind to them and disrupt vital processes within the cells. (Recall that corona-viruses are named for the shape of the nobs at the end of the spindles protruding around the virus’s spherical body.) The idea is to either completely disrupt how the virus binds to human cells…and/or to drastically slow the rate at which it reproduces.
DeepMind culled the resources yielded by AlphaFold to work on COVID-19’s genome. After running large genomic data-sets on its “machine learning” software, the system was able to ascertain the link between the organism’s genome and how its proteins are shaped. These predictions may contribute to an understanding of how the virus functions. DeepMind has not yet tested AlphaFold’s predictions outside of a computer, but it is releasing the results into the world in case researchers ANYWHERE can use them to develop treatments for COVID-19. Here, the scientific community is working together.
Meanwhile, Canadian start-up known as DarwinAI has developed a neural network that can screen X-rays for signs of COVID-19 infection. Using swabs from patients is currently the default for testing for coronavirus; but analyzing a “smart” system that analyzes chest X-rays can provide an alternative to hospitals with inadequate staff (or testing kits) to process all patients in a timely manner. Massive data-sets of X-rays were used to train the system. Machine-learning has enabled DarwinAI to learn from tens of thousands of images. In addition to expediting the process, this revolutionary approach reduces exposure.
Such technology needs to be shared, as this is a collaborative endeavor on a global scale. That is why DarwinAI released COVID-Net as an open-source system. Once it is out there for all to see, countless researchers will be ready to make helpful suggestions “from the field” for making improvements, and share their own insights.
DarwinAI is working on transitioning COVID-Net from a merely experimental (highly specialized) system to something that can be used by healthcare workers. It is also developing a neural network for “risk-stratifying” CoViD-19 patients–that is: determining levels of risk in order to separate those with the virus who might be able to recover at home vs. those who would be better-suited coming into a hospital. This is yet another reminder that machine-learning is playing an ever-more vital role in solving public health problems.
Open source is about harnessing the wisdom of crowds–an opportunity for ARTIFICIAL intelligence to make use of COLLECTIVE (human) intelligence.