Introduction

SARS-CoV-2 and the resulting COVID-19 pandemic is one of the biggest challenges across the globe. Fighting this virus requires the bravery of the social organization, healthcare workers, and technological solutions.

Can Deep Learning Help?

Deep Learning for “Natural Language Processing” (NLP) has been extremely successful. Applications for the COVID-19 pandemic include Misinformation Detection, Literature Mining, and Public Sentiment Analysis. Also, searching through the biomedical literature has been extremely important for drug repurposing. An example of this is the repurposing of baricitinib –  an anti-inflammatory drug for rheumatoid arthritis. The potential efficacy of this drug was found by querying biomedical knowledge graphs. For automated construction, modern knowledge graphs utilize Deep Learning. Other biomedical literature search systems use Deep Learning for information extraction from natural language queries. These Literature Mining systems have been applied towards question answering and summarization models that may transform search altogether.

Computer Vision is another application of the Deep Learning domain. The transformer revolution in Natural Language Processing (NLP) has been successful due to Computer Vision’s pioneering efforts towards large datasets, massive models, and the utilization of a hardware that fast-tracks parallel computation, namely GPUs. Computer Vision applications with respect to COVID-19 include a Medical Image Analysis, Ambient Intelligence, and Vision-based Robotics. Medical Image Analysis has been used to support RT-PCR testing for analysis by classifying COVID-induced pneumonia from CT scans and chest x-rays.

Haque et al recently published a survey on Computer Vision applications for physical space monitoring daily living spaces and hospitals. They termed these applications as “Ambient Intelligence”. This is an interesting phrase to include a massive set of more subtle applications, such as hand washing detection, automated physical therapy assistance, surgery training, and performance evaluation.

Deep Learning can improve virus spread models used in “Epidemiology”. These models use a history of infections, and information, such as lockdown durations, to predict future deaths or cases. The most well-known example of this are Susceptible, Infected, and Recovered (SIR) models. The illustrative SIR model describes how a population transitions from healthy or “Vulnerable”, to “Infected”, and “Recovered” through a set of three differential equations. These equations solve for the infection and recovery rates from the data of initial and recovered populations.

The application of Deep Learning for “Life Sciences” is quite exciting; however, it is still in the early stages. RT-PCR has become the gold standard for COVID-19 testing. This viral nucleic acid test utilizes transcription and primers enzymes to amplify a chunk of DNA, such that fluorescent probes can highlight the presence of the viral RNA.

Another exciting application area is the intersection of molecular engineering and Deep Learning. Deep Learning has gained a lot of prominence for the development of AlphaFold. Given the 1-dimensional string of amino acids, AlphaFold predicts the resulting 3-D structure. These models have been used to predict the 3-D structure of the spike proteins on the outer shell of the coronavirus, as well as its other proteins. Viewing a model of this structure enables biochemists to see possible binding targets for drug development.

These applications of Deep Learning to fight COVID-19 are promising, but it is equally essential to be aware of the possible drawbacks of Deep Learning.