A while ago I worked together with Anthony Bradley, Maciej Majewski and Xavier Barril on porting our dynamic undocking approach (Nat Chem 2016) from a MOE - Amber based version to a full open-source version. In that work we used MOE to essentially prepare and parametrize chunks of protein binding sites with their ligands in them. Amber was used to run steered molecular dynamics. The whole source code for running these simulations is available on the Barril lab website.
I love MOE, it’s a great piece of software, but unfortunately it’s not free and now that I’m nor in academia, nor in a pharma company I find myself unable to further develop an approach I implemented in the past which is kind of frustrating to put it politely. The same goes for Amber.
So I started to look for alternatives, also under a gentle demand from Anthony (poking me regularly) who wanted to apply the whole approach in his fragment screening pipeline. That’s in the scope of the crazy impressive XChem project. If you haven’t heard / read about that check it out - it gives you a bit of an insight on what will be possible very near future in xray crystallography!
Back at the time the openforcefield project just started and an alpha version for an rdkit integration was sort of available. After a lot of ugly hacking we were able to prepare and run our dynamic undocking runs using openforcefield and openmm. The code is still available here . The problem was though that this rdkit integration and the forcefield itself didn’t allow to cover the large variety of chemotypes in the fragment library at hand (only 60% were covered back at the time) and we had a lot of issues with parametrizing nitrogens in aryles. Unfortunately that’s rather common in druglike molecules ;)
Recently a few updates of the openforcefield toolkit came out … a game changer, as you’ll see.
The openforcefield initiative
If you have to follow and support something during the years to come, then this initiative. Why? It is an effort to bring an open source small molecule force field to the community. Where before you had to handle MMFF in MOE, OPLS in the Schrödinger tools etc here the idea is to enable molecular mechanics on small and macromolecules jointly using open and freely available software. That’s really awesome in my opinion (given the reasons I mentioned before + the tons of public money that already went into the development of some of the previous forcefields).
It’s lead by a consortium of academic and industrial partners and advisors
It’s all about open science / data
It’ll allow people to develop methods and apply them where before you needed expert molecular modelling tools that integrated and maintained for most of them their own small molecule force fields
It’s openmm native and amber compatible
They apply an orthogonal approach to parametrize the ligands using smirks patterns
Recently version 0.2 was released with this time around an official support for rdkit. Things are moving fast, today 0.4 was released. This post is describing the first tests on preparing ligand topologies with the version 0.4. More to come on more detailed tests in upcoming posts.
My test run here
In my first test run, source code available here, I wanted to see whether the 768 fragments from the XChem fragment library can be parametrized with the new version and how they behave after a simple minimization.
Again a bit of code, but just the relevant parts. The fully functional source code is on github.
My first naive approach intended to read in smiles and use the handy Molecule.from_smiles function to create a new molecule instance with the openforcefield toolkit.