A new artificial intelligence (AI)-based tool helps clinicians understand and better predict which adverse effects their COVID-19 patients could experience, based on comorbidities. It also helps suggest Food and Drug Administration (FDA)-approved drugs that could improve particular patients’ health outcomes.
Researchers from the Georgia Institute of Technology and Emory University collaborated on this work, which is described in a new study published October 21 in Scientific Reports.
The team’s methodology, or “decision prioritization tool”, is called MOATAI-VIR (Mode of Action proteins & Targeted therapeutic discovery driven by Artificial Intelligence for VIRuses). The researchers say it predicts 24 out of 26 major clinical manifestations of COVID-19 and their underlying disease-protein-pathway relationships.
Those clinical manifestations cover issues including acute respiratory distress, blood clotting issues, cytokine storms, low blood oxygen and white blood cell counts, and bone marrow failure.
“It’s still the question of, what’s causing the side effects?” says Jeffrey Skolnick, professor and Mary and Maisie Gibson Chair in the School of Biological Sciences and a co-author for the study. “So, you lost sense of smell and got brain fog — and another (patient) had respiratory distress, and another can’t remember the day of the week. What we’ve identified are the possible mode of action drivers for these various conditions, which is now setting the stage for who’s getting what side effects.”
Skolnick notes that it makes sense to predict the side effects based on protein interactions.
“Humans are molecular machines, and presumably there are biological and physical rules to dictate our responses,” he says. “We basically built an AI-based approach which was designed given the interactive set of proteins in humans which interact with the [novel] coronavirus,” he adds. “We then asked ourselves, ‘Could we predict, based on biochemical pathways, which interactive proteins are associated with side effects?’”
Skolnick explains that most known diseases are due to the “malfunction and interaction of many proteins,” and says it’s a collective effect — a “many-targeted protein effect.” His team’s new AI methodology is identifying as many targets as possible.
Comorbidities, such as diabetes, obesity, autoimmune disorders, and other conditions that affect the immune system, can play an outsized a role in risk factors related to COVID-19. Such conditions can be plugged into the team’s algorithm.
“Alzheimer’s and hyperthyroidism are strongly correlated — as is diabetes. There are six to eight (COVID-19) comorbidities with a patient that has Alzheimer’s,” Skolnick explains. “It’s not just old age — it’s much more complicated.”
The MOATAI-VIR methodology helps identify the common proteins of the comorbidities in relation to the parent disease. A clinician can then target the diseases with drugs. Researchers report that this specific methodology had 72% success in 123,146 drug-indication pair predictions found by Skolnick’s team.
“For a given disease, we prioritize them by the proteins that are most in common with the comorbid diseases to the given disease, giving rise to the particular complication, such as respiratory failure. This identifies the putative (assumed) driver proteins for the given complication,” he says. “Then we select repurposed drugs in two ways — we screen the most common comorbid proteins for their most frequent binding to repurposed drugs. For the set of comorbid diseases to a given complication, choose the drugs that treat the most complications.”
It’s critical to find the right drugs for those complications and side effects — and using the new “decision tool” can help do that, Skolnick says.
The MOATAI-VIR methodology algorithms can be downloaded at: https://sites.gatech.edu/cssb/moatai-vir/