Test, Test, and Test

A key strength of the UCL community is its ability to work across disciplines and with colleagues, partners and industry to help address the biggest challenges facing the world today. Across UCL, researchers are working to find a vaccine, improving diagnosis and, are advising Government here in the UK. In addition, they are helping to inform public knowledge by providing expert comment on issues as varied as predictions on virus spread, panic buying and stockpiling, broadband provision, and the economic and political impact of the pandemic.

UCLQ researchers have joined this effort, adapting the skills they use to study quantum science and technology to help the global response to COVID-19. For example, Professor Peter Coveney, who leads the EU H2020 Computational Biomedicine Centre of Excellence, and his colleagues at the UCL Centre for Computational Science, are part of a consortium of more than a hundred researchers from across the US and Europe, who are using an exceptional array of supercomputers – including the biggest one in Europe and the most powerful on the planet – to study several aspects of the virus and disease in detail.

They are using two of the world’s most powerful supercomputers – Summit at Oak Ridge National Lab, USA (1st) and SuperMUC-NG at GCS@LRZ, Germany (9th) – to screen libraries of drug compounds to identify those capable of binding to the spikes on the surface of the novel coronavirus, which the virus uses to invade cells, so as to prevent it from infecting human cells. The libraries include known and licensed drugs for quick repurposing opportunities (e.g. DrugBank), 100 million known small molecules that are drug like (e.g. PubChem) and large-scale libraries (e.g. Enamine, ZINC) with billions of compounds that could be manufactured quickly for testing.

Professor Coveney (UCL Chemistry), said: “We are using the immense power of supercomputers to rapidly search vast numbers of potential compounds that could inhibit the novel coronavirus, and using the same computers again, but with different algorithms, to refine that list to the compounds with the best binding affinity. That way, we are identifying the most promising compounds ahead of further investigations in a traditional laboratory to find the most effective treatment or vaccination for COVID-19.

“We are able to scan existing drug libraries, so many of the compounds we are looking at already have approval for use in humans as they are used to treat existing diseases so could be repositioned to target COVID-19. We are also able to computer-generate new compounds that should bind well to the virus, which gives us a fantastic head start on discovering potential new drugs.”

The compounds screened include chemicals, herbal medicines and natural products that have either been studied in humans or are already approved drugs already considered safe for humans. The supercomputers are able to complete the scanning tasks in days, where it would take regular computers months.

“This is a much quicker way of finding suitable treatments than the typical drug development process. It normally takes pharma companies 12 years and $2 billion to take one drug from discovery to market but we are rewriting the rules by using powerful computers to find a needle in a haystack in a fraction of that time and cost,” he added.

“Supercomputers are a remarkable resource for the development of COVID-19 treatments as they can identify possible treatments through a variety of ways including machine learning, complex molecular dynamics and artificial intelligence methods. Not only do we need to find molecules that bind to the spikes on the coronavirus, but we also need to model how well these bind when we know the spikes move around.”

In addition to the spike proteins on the virus surface, Professor Coveney’s group is simulating how pharmaceutical drugs interact with proteins involved in various other stages of the virus lifecycle. For instance, 3CL-protease is a protein that is key to the virus replicating itself, allowing them to grow in number and cause further damage in the body. Drugs that bind well to this protein may slow or halt its replication. Simulating this drug-protein interaction can identify further drug targets.

Professor Coveney, who is leading the European side of the global effort, welcomes colleagues from across UCL to join the collaboration, and in particular, those with expertise in machine learning and generative methods for compound discovery, and physics-based methods for calculating binding free energies. Anyone with interests across the wider scope of this collaboration should also feel free to get in touch.

Early in the pandemic, SARS-CoV-2 test kits were in critical shortage in many countries, limiting large-scale population testing and hindering the effort to identify and isolate infected individuals. UCLQ research fellow Dr Nikolas Breuckmann, with colleagues from the Max Planck Institute for Mathematics and the Center for Vaccinology at the University Hospitals of Geneva, set out to develop and evaluate multi-stage group testing schemes to make the most of the limited testing resources.

“We showed that the testing capacity for COVID19 can be greatly increased, without having to change the test kits themselves”, said Breuckmann.

Using the existing test kits, the team developed and evaluated ways to group several samples in multiple stages. Through their approach, groups of negative samples can be eliminated with a single test, avoiding the need for individual testing and achieving considerable savings of resources. If, on the other hand, one or more samples in the group are positive then further testing has to be done.

The team designed and parametrised various multi-stage testing schemes and compared their efficiency at different prevalence rates using computer simulations with were optimised with clinical feasibility in mind, i.e. only considering those schemes which are practical.

They found that three-stage testing schemes consisting of pools with a maximum of 16 samples can test up to three and seven times as many individuals with the same number of test kits for prevalence rates of around 5% and 1%, respectively.

Breuckmann said: “We show that, for example, in the case of South Korea five times as many people could have been tested with the same number of tests.”

The team proposes an adaptive approach, where the optimal testing scheme is selected based on the expected prevalence rate, and believe that these group testing schemes could lead to a major reduction in the number of testing kits required, helping to improve large-scale population testing in general and in the context of the current COVID-19 pandemic.

Find out more about COVID-19 research at UCL.

This article was featured in UCLQ’s 2019/20 annual report.