The crystal structures of the non-covalent fragment hits were visually assessed to establish reasonable vectors for growth within the pocket. Suitable fragments were then virtually enumerated using a bespoke KNIME workflow and the enamine compound library. Classical one-step coupling reactions were used e.g. amide couplings and N-alkylations to speed up the future synthesis of any compounds. The individual fragments were assessed prior to enumeration to decide on suitable coupling chemistries for that particular fragment. For example, a methyl amide pointing towards an empty pocket would suggest that enumeration using amide couplings and a library of amines may be tolerated. The aim was to minimise disruption to the existing binding pose and create new interactions by extending into unoccupied pockets. A summary for this process has been created and attached. COVID19 fragment hit enumeration summary.pdf (861.7 KB)
The generated library was then passed through a bespoke PAINS and property filter using KNIME. The workflow for this will hopefully be uploaded shortly, so that others can use the tool for filtering ideas. An excel spreadsheet for the results so far has been created and needs to be uploaded.
To filter the ideas further for synthesis we plan to dock each of the enumerated libraries into their corresponding crystal structures and filter by docking score. Ideally we plan to use maestro nodes in KNIME to automate the process, since the potential library is very large. We are currently waiting for a license to do this remotely. We also plan to use FAF-Drugs4 (https://mobyle.rpbs.univ-paris-diderot.fr/cgi-bin/portal.py?form=FAF-Drugs3#forms::FAF-Drugs4) to generate ADME and tox data, and have set up a KNIME workflow to process this data in a more ‘user friendly’ format. Again we hope to upload this soon.
Hi Holly,
I recently build a very similar KNIME Workflow (I called it the FragXplorer), which is available on https://www.biosolveit.de/KNIME/.
I use a template-based docking for it, which keeps your starting-point fragment in place and docks only your new parts of the enumerated molecules. Maybe you can recycle parts from my workflow for yours.
(BioSolveIT sponsors licences for this project anyway).
Best,
Franca
Hi,
I look forward to seeing the KNIME workflow for this if you’re able to share - would be really informative and educational for those of us who are non-expert Knime dabblers!
Just some suggestions on your process. The PAINS filter is an interesting step but assuming we’re dealing with xray structure and / or anti-viral assays as a primary assay endpoint we need to be sure the PAINS question is relevant - PAINS are certainly relevant in terms of interference in target-based biochemical assays, but arguably less so in biophysical assays like x-ray crystallography, and also less relevant in anti-infective phenotypic assays with an anti-pathogen endpoint.
For the property filtering step I’d just advise to be careful comparing the properties of what are currently early hit-stage compounds with the properties of actual drugs. I agree it’s always important to keep in mind what sort of properties are displayed by drugs, and we can certainly prioritize hits based on their property profile, but we but we shouldn’t filter out compounds completely at the hit stage based on the similarity of properties to actual drugs, it’s a bit like comparing apples to oranges. Medchemists are generally really good at tuning properties in the direction they want during optimization from hit to lead to candidate.
I have created a basic SOP to explain to new users how to download and use the PAINS filter. Hopefully this can be useful for others to use for this project
Hi
the problem we’ve had is that getting the tools up and running in high throughput mode has taken a week or so, but now with the fantastic input of Holly and others as we go deeper into the project, what is now running can be applied systematically from now on and efficiently moving ahead. Definitely agree that at the early fragment stage the profiling is for guidance at the moment. The filters highlight those that maybe need a second look.
Thanks for the message. What kind of forcefield does the biosolevit docking use? My plan going forward was to dock the enumerated fragments using the schrodinger nodes in KNIME to speed up the process. Just like you suggested, I was planning on keeping the fragment pose relatively rigid and only allow the enumerated portion to be flexible. However, it would be great to have a go with your templated docking protocol, as this would speed up the process! Do you have any idea of how I can get a license?
I have attached the PAINS KNIME workflow below, as well as a beginners guide on how to access and use this. Hopefully it is interesting/useful!
Yes, I agree often PAINS filters can remove interesting compounds from the data set. As such this filter only contains around 20 of the most common structures. The rationale behind this is detailed a bit more in the summary document. The structures as well as being PAINS tend to have reactive functional groups, which would also be undesirable moving forward, so hopefully filtering these out at this stage is reasonable.
The property filtering step can be customised in the workflow using a GUI, so at this early stage we can modify this to suit i.e. initially we can be far from drug like property space, but it can be useful to filter out ludicrously large compounds etc that don’t give much scope for optimisation etc.
