AI for COVID-19 Research

As the world grapples with the effects of the COVID-19 pandemic, researchers are working tirelessly to understand how the virus is transmitted, how to improve testing, what existing therapies can mitigate the effects of the disease, and the most likely path to containing the disease.

With a vaccine unlikely to be available until 2021, we’re stuck with trying to prevent the spread and safely treat those with the most dangerous symptoms.

AI can help.

With tens of thousands of medical research papers available worldwide, focused on new and previous research over the past 15 years on various types of coronaviruses and drugs, natural language processing algorithms can save precious time by processing massive volumes of text to uncover key findings and narrow human research to the most relevant content.

On March 27, 2020, we shared our initial AI platform review of the COVID-19 dataset. This initial review focused on the commercial use subset of research papers from our March 24 download from the Allen Institute’s SemanticScholar.org website.

Commercial use subset (includes PMC content) 347Mb:

• PDF - 9365 full text (new: 70, removed: 20)
• PMC - 8995 full text (new: 8995, removed: 0)

Today, we’re sharing our findings from running the additional 30,000+ research papers through the Rehinged data processing engine.

Non-commercial use subset (includes PMC content) 71MB:

• PDF - 2377 full text (new: 54, removed: 27)
• PMC - 2093 full text (new: 2093, removed: 0)

Custom license subset (includes PMC, Elsevier content) 513MB:

• PDF - 23152 full text (new: 2541, removed: 46)
• PMC - 4773 full text (new: 4773, removed: 0)

bioRxiv/medRxiv subset (pre-prints that are not peer reviewed) 19Mb:

• PDF - 1342 full text (new: 348, removed: 59)

Results

Rehinged applied the same NLP algorithms and perception metrics to this corpus as used on the previous corpus, analyzing the 30,000+ research papers for how the authors discussed a variety of drugs in studying COVID-19.

Our AI platform has the ability to interpret and compare any objects on any perception metrics. For this project, we measured the above articles for the words “improvement”, “good”, “effective” and “unique.”

Learning #1:‍

Hydroxychloroquine scored high for “improvement” in this group of papers. This means that the language in these papers referenced improvement.

Article: Gupta R. et al., Diabetes & Metabolic Syndrome: Clinical Research & Reviews Vol. 14, Issue 3, May-June 2020, Pages 251-254 [25 March 2020]

Title: Contentious issues and evolving concepts in the clinical presentation and management of patients with COVID-19 infection with reference to use of therapeutic and other drugs used in Co-morbid diseases (Hypertension, diabetes etc.)

DOI: https://doi.org/10.1016/j.dsx.2020.03.012

“In this study of 36 patients, 20 patients were treated with hydroxychloroquine (HCQ), out of which 6 also received azithromycin, and 16 patients served as controls. At day 6 post treatment the proportion of patients who were negative for SARS CoV-2 was 100%, 57% and 12.5% for those treated with HCQ and azithromycin combination, HCQ only and controls, respectively. Though this is a small study, the results are very encouraging.”

Article: Gautret Ph. et al., International Journal of Antimicrobial Agents (in press) Available online 20 March 2020, 105949

Title: Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial

DOI: https://doi.org/10.1016/j.ijantimicag.2020.105949

“Hydroxychloroquine (an analogue of chloroquine) has been demonstrated to have an anti-SARS-CoV activity in vitro. Hydroxychloroquine clinical safety profile is better than that of chloroquine (during long-term use) and allows higher daily dose and has fewer concerns about drug-drug interactions.” (Researchers - please view our comments on this at the end of the learnings.)*

Learning #2

‍ARB scored high for “improvement” in this group of papers. This means that the language in these papers referenced improvement.

Article: Bansal M., Diabetes & Metabolic Syndrome: Clinical Research & Reviews Vol. 14, Issue 3, May-June 2020, Pages 247-250 [25 March 2020]

Title: Cardiovascular disease and COVID-19

DOI: https://doi.org/10.1016/j.dsx.2020.03.013

“Since SARS-CoV-2 binds to ACE2 to gain entry into human cells, there is a potentially increased risk of developing COVID-19 or developing more severe disease in patients who are already on background treatment with ACEi/ARB. However, to date, no experimental or clinical data have emerged to support these concerns. At the same time, the risks of discontinuing these therapies are well known. Therefore, several leading professional societies have strongly urged to not discontinue clinically-indicated ACEi/ARB therapy in the event the patient develops COVID-19.”

Learning #3:

‍Chloroquine scored low for “improvement” in this group of papers. This means that the language in these papers didn’t reference improvement.

Article: Xiong R. et al., bioRxiv (not peer-reviewed) Posted online 12 March 2020*

Title: Novel and potent inhibitors targeting DHODH, a rate-limiting enzyme in do novo pyrimidine biosynthesis, are broad-spectrum antiviral against RNA viruses including newly emerged coronavirus SARS-CoV-2

DOI: https://doi.org/10.1101/2020.03.11.983056

"Compared with our previous publication of Remdesivir (EC₅₀=0.77μM, SI>129.87) and Chloroquine (EC₅₀=1.13μM, SI>88.5), which are currently used in clinical trials against SARS-CoV-2, S416 had much greater EC₅₀ and SI values (66.5-fold stronger than Chloroquine in EC₅₀) against SARS-CoV-2."

Learning #4:‍

Corticosteriod scored low for “improvement” in this group of papers. This means that the language in these papers didn’t reference improvement.

Article: Qiu C. et al. medRxiv (not peer-reviewed) Posted online 06 March 2020*

Title: Transmission and clinical characteristics of coronavirus disease 2019 in 104 outside-Wuhan patients, China

DOI: https://doi.org/10.1101/2020.03.04.20026005

"Lopinavar/ ritonavir was proved to be substantial clinical benefit against SARS and MERS. Particularly, 80 (76.92%) patients received traditional Chinese medicine therapy, the efficacy in COVID-19 treatment needs more clinical evidence to be confirmed. Controversy about corticosteroids in NCP treatment has not reached a consensus. Evidence suggests corticosteroids did not decrease the mortality of patients with SARS and MERS, but rather delayed the clearance of viral. Chinese guideline recommends a short treatment of corticosteroids in server NCP. Just 8.65% patients of this study received glucocorticoids treatment, most of them are severe patients."

Learning #5:

Ritonavir scored neutral for “improvement” in this group of papers. This means that the language in these papers didn’t reference improvement or degrading with patients.

*Please note: if you're a researcher, we acknowledge that it's important to separate in vitro and in vivo papers, and to also scientifically evaluate non peer-reviewed papers. For this exercise, we simply ran the entire batch of 30,000+ papers through our AI engine. Contact us if you'd like us to perform specialized analyses.

Takeaways

Again, the AI in this use case isn’t likely to uncover a cure by looking at a single analysis. But when researchers evaluate a combination of analyses, many times a story appears.

When we compare the results of our AI analysis on the first batch (posted March 27) with this analysis of the second batch, we find the following:

1. Hydroxychoroquine scored high for improvement in both batches, significantly higher than chloroquine
2. Chloroquine scored negative for improvement in both batches
3. Ritonavir scored positive for improvement in one batch, and neutral in another batch

Next Steps

We’ll continue to review additional datasets and explore some visual methods for displaying our results. Contact us if you'd like to learn more.