Machine Learning models can be used to understand the protein complexes involved with infection and what treatments/drugs produce a given relationship.

Protein structure prediction

Unlike pathogenic laboratory testing which produces a significant amount of false-negatives while still being a time-consuming method, Machine Learning models  can display radio-logical and image patterns via CT chest scans. 3D CT volume can detect the virus's transmission by using a lung-region 3D CT scan which is segmented and fed into a 3D deep neural network to predict the infection rate.

Large-Scale Analysis of CT Scans: 

Machine Learning models can provide healthcare systems with a forecast of patient outcomes, risk of hospitalization, and probability of complications ranging from respiratory distress to organ failure. The model takes in blood samples and CT scan datasets to develop machine learning models of high risk patients. 

At high speed, these tools (logistic and FbProphet models) also observe trends in the accumulated epidemiological data with which they predict how the future epidemic curve will look like. It's determined that by end of October, approximately 14.1 million people will be infected, in a cumulative aspect. 


Pandemic Predicitions

Pandemic Simulations

Machine Learning models can be used to better understand the proteins and complexes that are incvolved with the novel Coronavirus. This includes producing predictions of pandemic regarding hospitalizations, patient outcomes, and simulations.

University of California, Santa Cruz

Smith College

Combines computational and statistical techniques, mathematics, and engineering to analyze and interpret biological data of patients affected by COVID-19

Bioinformatics for COVID-19

Biomedical engineers have put in extensive time and research in testing methods and diagnostics regarding the novel coronavirus. The test results of the population have provided mathematical models to visualize the pandemic, alternatively the private sector has produced a high-volume of nasopharyngeal, throat, and nasal test kits by bioengineering front-runners including: Abbot, Xpert, Lightmix, and Genmark.

Diagnostics and Testing with Biomedical Engineering

Allam, M., Cai, S., Ganesh, S., Venkatesan, M., Doodhwala, S., Song, Z., . . . Coskun, A. F. (2020). COVID-19 Diagnostics, Tools, and Prevention. Diagnostics, 10(6), 409. doi:10.3390/diagnostics10060409

https://www.mdpi.com/2075-4418/10/6/409

COVID-19 Diagnostics, Tools, and Prevention

- Oxford University Press, Bioinformatics 
- Journal of Translational Medicine
- Frontiers in Immunology

Bioinformatics for COVID-19 References by Journal

Machine Learning Analysis 

Understanding of COVID-19 target-ligand interactions represents a key challenge, particularly in observing the interactions in vitro. To combat this, the 
COVID-19 Docking Server was developed. It is a web server that predicts binding modes between COVID-19 targets and ligands (small molecules, peptides, antibodies, etc). 

In a very short time, many proteins essential to SARS-CoV-2 have been discovered by researchers in China, including the main protease and spike protein in binding with angiotensin-converting enzyme 2 (ACE2) (this is the primary mode of entry that SARS-CoV-2 uses when initiating cell entry).

As of May 2020, 181 structures were released in the Protein Data Bank, which were  related to 11 different kinds of SARS-CoV-2 proteins.

Structures of the other proteins remain unknown, so the modeled structures of SARS-CoV-2 in the dock server are based on the homologous structures of SARS-CoV, due to high amino acid identity between the two coronaviruses (77.2%). 


INTRODUCTION: COVID-19 Docking Server 

- AI-based programs are trained with large amounts of existing data for their individual purposes. The two types used in this study are:
- (1) Fluency, the AI-based binding affinity prediction platform, which was used on existing FDA approved drugs to test their binding to ACE2 or TMPRSS2 which are 2 proteins used by the virus to enter cells. 
- (2) The Disease Cancelling Technology platform, which was used to assess different drugs' ability to induce gene expression signals that counteract disease-associated signals (identified by examining existing data on gene expression in mouse models). 

Methods of Using Bioinformatics to Identify Potential Therapeutics for COVID-19

- The main benefits of using these unbiased AI programs are (1) speed and (2) widespread availability. 
- Because time is of an essence  in a pandemic, these programs can aid in identifying existing drugs for repurposing and select candidates for clinical trials. 

Benefits of Using Bioinformatics 

Virus-CKB or the viral-associated disease-specific chemogenomics knowledgebase is a platform created to quickly predict existing FDA-approved drugs which can be repurposed and progress to clinical trials to be used to treat COVID-19. New and effective anti-COVID-19 drugs are in urgent need, but discovering new drugs can take up to 8 years, can cost up to $2 billion and have low approval ratings, hence why re-purposing drugs could be useful. Essentially, the database was created to make viral infection research easier.


The Virus-CKB Platform

Computational vaccinology involves using tools that facilitate in silico prediction of immune responses to pathogens by:
- Epitope (art of an antigen molecule to which an antibody attaches itself) mapping
- Antigen selection
- Immunogen design

Use of Bioinformatics in Vaccine Development for COVID-19

Key Features of Diagnostic Testing and Sampling

Specificity measures the ability for the test to produce a correct negative result for people who don't have a given condition.
ex) A test with a 80% sensitivity will return a correct positive result 85% of the time, but a a false negative 15% percent of the time.

Specificity in Testing

Sensitivity is a measure of how often a "true positive" result is produced. A test with a sensitivity of 85% will produce a "true positive" 85% and a false negative 15% of the time.

Sensitivity of Testing

Currently Used Methods for Diagnosis

Sample Collection Methods for Diagnosis

- Isothermal Nucleic Acid Amplification
- RPA
- CRISPR
- Cas13a Assay
- Next Generation Sequencing 
- Nanomaterials (nanotrap particles, etc.)
- Wondfo Qualitative Rapid Chromatographic Immunoassay (RCI)

Emerging Methods for Diagnosis of SARS-CoV-2

- A reduction of ADAMTS13 levels can potentially predict mortality risk for patients infected with SARS-CoV-2

Potential Diagnostic Methods for COVID-19 Mortality Prediction  

Weekly surveillance testing along with isolation would more effectively limit COVID-19 outbreaks than the slow and expensive (yet precise) qPCR (quantitative polymerase chain reaction) testing method that is currently employed.

Testing Strategy for Controlling COVID-19 Outbreaks

In August, the CDC officially recommended that those without COVID-19 symptoms didn't need to be tested- even if they knew they'd been exposed. This decision received a lot of backlash from public health officials pushing for more testing.  It was later revealed that that the original guidelines had been re-written by those involved with the Department of Health and Human Services and the US COVID task force.

Learn Before

Related

Learn After