LONDON – COVID-19 research is generating a wealth of data every day, but it is coming from many and disparate sources, making it difficult to assess its quality, dovetail datasets together and decide how to apply it.
“It’s quite a challenge to understand the data and judge which data we can more trust, and which to use. Garbage in, garbage out is a real problem,” said Namshik Han, head of computational biology at Cambridge University’s Milner Therapeutics Institute.
Back in February, when Han first began to look at applying emerging understanding of COVID-19 to an in silico project to hunt for marketed drugs that could be repurposed to treat the infection, the two most relevant datasets available were published in preprints.
The rush to get research findings into the public domain means it remains the case that much of the data are still being published without peer review.
“We had to set up quality control measures and thoroughly test datasets to select what to use,” Han told attendees of the virtual BIA Bioscience Forum.
Even so, there are inconsistencies – in methodologies, cell lines used and recording of patient data – that have to be reconciled. “We needed to design machine learning to suit the datasets; that’s the only thing we can do in practice,” Han said.
Han’s project drew on two categories of data, cross-referencing outputs of scientists who used mass spectrometry to uncover the 332 proteins in the human body that directly bind to SARS-CoV-2, against research identifying changes in host protein expression following infection with the virus.
“We wanted to identify pathways from directly interacting proteins to differently expressed proteins,” Han said. Based on the in silico network they generated, the Milner researchers found that 200 drugs hit targets on those pathways and were likely candidates for repurposing.
The list was endorsed by other research. “It turned out 40 were already in [COVID-19] clinical trials and 30 were reported as being of possible utility,” said Han. “This testified the validity of the approach.”
Further work using neural networks to assess what drugs targeted which pathways pointed to different mechanisms of action. Five drugs were shown to be inhibitors of viral replication. Two of those have been shown in vitro to be particularly effective, Han said. “It shows this is a really very promising strategy.”
Working to Define COVID-19
Given its role of modulating autophagy vs. apoptosis, the molecular chaperone heat-shock protein 90 (Hsp90) has been of interest as a cancer target for more than two decades. But despite a 2012 review of Hsp90’s role in viral processes, showing it has an effect on almost every viral infection ever studied, Hsp90 inhibitors “have not to my knowledge been tested” as antivirals, said Fiona McLaughlin, director of EVA Pharma Consultants.
That position is shifting, after data published early in the pandemic pointed to Hsp90 being central to infection by SARS-CoV-2. McLaughlin said the potential of Hsp90 inhibitors was subsequently confirmed in SARS-CoV-2-infected human cell lines, where they reduced viral replication and the expression of inflammatory cytokines.
Those findings have revived the fortunes of luminespib (AUY-922), a small-molecule Hsp90 inhibitor, discovered by scientists at the Institute of Cancer Research in the UK and the biotech company Vernalis plc. Luminespib was partnered with Novartis AG in 2004 and subsequently reached clinical proof of concept in solid tumors, before being shelved in 2014.
McLaughlin said the drug’s possible effect in treating COVID-19 is now being assessed in animal models. The current owner, Ligand Pharmaceuticals Inc., of San Diego, is looking for partners to take luminespib forward in that indication.
At the start of the pandemic, U.K. research charity Lifearc awarded a £2 million (US$2.6 million) grant to Edinburgh University’s Center for Inflammatory Research, enabling its 150 scientists to be redeployed to work on Define (formerly Stopcovid), a project testing existing drugs to find a treatment to prevent lung inflammation progressing to the point patients need mechanical ventilation.
At the specialist center, the Edinburgh researchers have access to specialist tools and resources that are generating novel datasets. One example is optical sensors that can see inside patients’ lungs in real time, using endomicroscopy techniques. That makes it possible to assess how inflammatory pathways are being affected by the disease process and by possible treatments.
Another is access to postmortem data on the distribution of different cell types. Catriona Crombie, philanthropic fund manager for Lifearc, said from that perspective, different patients show a very different immune response at a cellular level. “We need to get to the bottom of this, to understand how better to treat COVID-19,” she said.
In addition, there is much to work to be done in working out how to manage the interplay of such different types of data as postmortem analyses and genetic data. “We need to better understand how to bring big datasets together,” Crombie said.