One of the really heartening things to come out of the current pandemic is the willingness of many scientists to put aside their own research and throw themselves into the efforts to find a treatment. However, lack of domain expertise is always a problem when scientists enter a new field, so I thought I'd put together a few things to consider.

In silico screening, for docking experiments you need to put considerable effort into ensuring the protein structure used is appropriate, you can't simply download a PDB file from the Protein Data Bank and use it. It will undoubtedly contain errors, you will need check protonation, hydrogen bonds etc. Then there is the issue of deciding which solvent molecules are important. Binding energies, docking scores are not as accurate as many seem to assume and no substitute for an experienced medicinal chemists looking at the bound poses, I've tried to summarise the types of molecular interactions here. Remember to also think about the impact of solvation. For other virtual screening approaches you need to be very careful about the quality of the input data. In many cases it will be heavily biased towards actives.

In silico predictions are no substitute for biological data, if you are using repurposed drugs or available chemicals there is really no excuse for not generating the appropriate in vitro biological data, there are many labs who would be happy to collaborate. If the molecules are novel many custom synthesis companies have offered to help. Remember that the IC50 is probably not that useful, it is likely that you will want to block the target 100% so you need to be above the IC95. In vitro biochemical assays using isolated enzymes will often give a false sense of potency, you should also determine activity in a cell-based assay in the presence of plasma.

If you are proposing a repurposed drug there will be a lot of information about the drug in the public domain, you may well need to search for compound codes, and various drug name synonyms. UniChem is a very useful web service for cross-referencing between chemical structure identifiers.

There are now many free, web-accessible databases some useful starting points are shown in the table below.

Name	Link	Description
ChEMBL	https://www.ebi.ac.uk/chembl/	A database of bioactive drug-like small molecules, it contains 2-D structures, calculated properties (e.g. logP, Molecular Weight, Lipinski Parameters, etc.) and abstracted bioactivities (e.g. binding constants, pharmacology and ADMET data).
PubChem	https://pubchem.ncbi.nlm.nih.gov	Three linked databases within the NCBI's Entrez information retrieval system. These are PubChem Substance, PubChem Compound, and PubChem BioAssay. Many compounds have links to primary literature and patents
Guide to Pharmacology	https://www.guidetopharmacology.org/GRAC/searchPage.jsp	An expert-driven guide to pharmacological targets and the substances that act on them.
DrugBank	https://www.drugbank.ca	The DrugBank database is a unique bioinformatics and cheminformatics resource that combines detailed drug data with comprehensive drug target information
NCI Thesaurus	https://ncithesaurus.nci.nih.gov/ncitbrowser/	NCI Thesaurus (NCIt) provides reference terminology for many NCI and other systems. It covers vocabulary for clinical care, translational and basic research, and public information and administrative activities
Clinical Trials	https://clinicaltrials.gov	A database of privately and publicly funded clinical studies conducted around the world
FDA	https://www.fda.gov	Food and Drug Administration responsible for safety and efficacy of drugs
WIPO	https://www.wipo.int/portal/en/index.html	World IP services

Find out the original target and mode of action. I've seen a couple of proposed compounds that are known prodrugs, the parent compound is designed to either breakdown or be modified in vivo to yield the active compound. The prodrug may have negligible systemic exposure. Covalent modifiers may look attractive but selectivity is always a concern and they may have narrow therapeutic windows.

Look at the original indication, many anticancer drugs are extremely toxic and could not be given other patients. Similarly, drugs that reduce blood pressure or other physiological changes may be problematic. You may well be able to find counter-screening data, this could highlight problematic off-target activities.

Look at the approved dosing regime, if a drug is only approved for doses of 2 ug/kg there might well be good reasons, and if your proposed drug only has uM activity in the in vitro assays you won't be able to generate sufficient plasma concentrations. Check what safety studies have been undertaken, are they sufficient to support multi-day dosing?

Look at the pharmacokinetics, you should be able to model the dosing regime needed to maintain plasma concentrations above IC95, this will may need to be maintained 24 hours a day. Check protein binding and distribution and use in the predictive modelling.

Look for the routes of administration, for in intensive care I suspect many will need the drug to be administered i.v. if there is no intravenous formulation is the drug soluble enough for one to be developed, ber in mind the limitations of intravenous formulations

Many of the patients will be on multiple drugs, both to treat the viral infection but also adventitious bacterial infections and since many are elderly and have pre-existing medical conditions they may have a cocktail of drugs prescribed. Drug-Drug interactions thus become a major concern, any proposed drug to treat the virus that has major interactions with CYP450 enzymes (induction, inhibition or metabolism) is likely to hugely complicate the overall dosing regime.

Check for any toxicity information, particularly black box warnings. HERG inhibition and QT prolongation is an issue that most drug discovery projects have to address at some point. This is particularly worrying if coupled with potential drug-drug interaction described above. You should also be able to find the data from safety studies, these may describe the dose limiting toxicities.

All of this information should be in the public domain, and if you are proposing a compound as a "Drug Candidate" you should not be expecting someone else to pull it all together to decide whether it is worth pursuing clinically.

Updated 26 April 2020