COVID-19 win for AI pharma farmers

The vaccine roll-out is in action all over the world. Nevertheless, given the propensity of SARS-CoV-2 to mutate, we are not yet out of the woods. The reality is, we are likely to be living alongside this virus for some time yet. For a successful post-COVID-19 era, it will be essential to have a variety of useful and effective treatments for the disease. In normal times, drug hunters would expect to invest over a decade of research and around $2 billion in the development of a new drug. Evidently, this sort of timeline is too slow to deal with the current crisis.

To combat the lag in bringing treatments from bench to bedside, researchers have looked to use artificial intelligence (AI) to find useful pharmaceuticals. The circumstances set out by the pandemic have laid a proving ground for AI in drug discovery and this may act as a watershed moment for the future of this technology (see our blog From data to drugs - a brief introduction to AI in drug discovery).

Drugs which already have safety data available present a faster route to approval for new uses. In addition, compounds which have been through clinical trials often have well established pharmacokinetic/pharmacodynamic, safety, and tolerability profiles. All this data is incredibly useful for feeding into AI methods such as deep learning. Indeed, repurposing of approved drugs is a promising route to getting treatments off the ground quickly. Consequently, drug repurposing has been an early focus for COVID-19 therapies. See our blogs What use is computational biology during the coronavirus (COVID-19) pandemic? and Has a cure for coronavirus (COVID-19) already been made? which have previously discussed repurposing efforts.

Sorting the wheat from the chaff

Clearly certain types of drugs represent obvious starting points for treating COVID-19. For example, antiviral and anti-inflammatory treatments are a good starting point for dealing with either the causative agent (SARS-CoV-2) or the pathophysiology of the disease (inflammatory response). However, even within these pharmaceutical subsets, there are a vast number of possible drugs and it is difficult to know where to start.

The Ebola antiviral remdesivir has been shown to be effective in the treatment of COVID-19. However, other drugs with some antiviral activity such as lopinavir-ritonavir (HIV/AIDS) and the malaria drug hydroxychloroquine have both been found to be ineffective in the reduction of mortality from COVID-19, and are no longer recommended for use by the WHO.

To help find the signal in the noise a group from the Cleveland Clinic and Case Western Reserve University, Ohio set up a knowledge graph (CoV-KGE). Using deep learning methodology they identified a number of promising candidates for repurposing. Among the top compounds identified using their knowledge graph were dexamethasone, melatonin and toremifene.


Dexamethasone is an anti-inflammatory drug which is used in the treatment of a range of diseases such as rheumatoid arthritis, allergies, asthma and COPD. The UK based RECOVERY clinical trial has shown that dexamethasone can reduce the mortality rate of patients receiving ventilator treatment by a third.

Indeed, dexamethasone was among the first treatments recommended for use in treating COVID-19 infection by regulatory bodies like the EMA [EU] and MHRA [UK].  


After identification through CoV-KGE, toremifene, a non-steroidal treatment for breast cancer was further validated using computer modelling of the interactions between the compound and SARS-CoV-2 proteins. These studies suggest a possible mechanism through which toremifene could inhibit viral replication.

Another study from the same group delved deeper into the use of melatonin as a potential useful treatment for COVID-19 based on network medicine methodologies. Retrospective analysis of patient data showed a 28% reduction in likelihood of a positive test result for SARS-CoV-2 in patients that used melatonin (rising to 52% for black Americans).

Further use of the knowledge graph built by this team of researchers found that toremifene and melatonin attack different parts of the SARS-CoV-2 disease network. This suggested that this combination of drugs could be complementary in the treatment of COVID-19, while avoiding overlapping toxicity mechanisms. The implication is that there could be a synergistic link between toremifene and melatonin with the possibility of improved treatment outcomes and reduced side-effects. Phase II trials on the efficacy of this combination are set to begin in spring 2021.

This type of analysis is particularly interesting, as the identification of effective drug combinations is notoriously difficult due to the increase in possible permutations. It will be exciting to see if the clinical trials validate this method of predicting combination therapies.


Another team using knowledge graph analysis (at pharmaceutical firm BenevolentAI) identified the arthritis drug baricitinib in early 2020 as a candidate for the treatment of COVID-19. The analysis suggested that baricitinib could provide both anti-inflammatory effects and prevent SARS-CoV-2 entry into lung cells.

Unfortunately, the use of baricitinib alone was not found to result in a statistically significant improvement in treatment outcome versus the standard of care. Nevertheless, more recently the results from the ACTT-2 trial for baricitinib in combination with remdesivir, showed an impressive reduction in recovery time from 18 to 10 days for patients on the verge of requiring invasive ventilation compared with treatment using remdesivir alone.

This combination has received emergency use approval from the FDA.

AI entering the mainstream

There are already hints that AI enhanced drug discovery is becoming more commonplace. Big pharma appears to be starting to reap the rewards of AI partnerships that have been emerging over the last few years.

Exscientia is a pharma company that focuses on the de novo design of drugs informed by AI. In January 2020 they announced that their compound DSP-1181, designed in collaboration with Sumitomo Dainippon Pharma using AI methods, was entering clinical trials for the treatment of obsessive-compulsive disorder. This is the first ever AI designed drug to be tested in humans. Moreover, the development timeline on that project was less than 12 months from initial screening to the end of preclinical testing almost 5 times faster than the industry average of 4.5 years.

In January this year, BenevolentAI and AstraZeneca announced that after almost 2 years their partnership on chronic kidney disease (CKD) has generated a new target that will be entering AstraZeneca’s portfolio. CKD is an incredibly complex disease which is highly prevalent, and few treatment options are available. In the words of Benevolent CSO Anne Phelan, CKD has “defied conventional research efforts” so to have a novel target is both impressive and promising.

COVID-19 success story?

The successful use of AI in the search for COVID-19 treatments has clearly shown the propensity for these methods to improve the drug discovery process. No doubt, the accomplishments have been spurred on by the increase in research openness surrounding SARS-CoV-2. Access to data is especially important for deep learning methods like those using CoV-KGE and the Benevolent knowledge graph. The interesting questions are: will this evidenced success accelerate the uptake of AI methods elsewhere in the pharma pipeline? And what can we expect to see next?

End-to-end integration looks to be key for the future. New companies like Valo are entering the scene with platforms claiming to provide assistance at all stages of the drug discovery process from target identification, to candidate generation and clinical development. Investors seem to be paying attention as Valo closed their Series B funding round with an additional $300 million in investment. Beyond venture capital, investment banks like Credit Suisse are even hosting panel discussions on the intersection of pharma and AI. Boom time for the industry appears to be on the horizon.