Feature SDF

[chryswoods]

I've added in a complete SDF (MOL) file parser. This can read and write the entirety of a SDF file. Note that this does not infer any missing info, e.g. stereochemistries etc. This just preserves the data that comes from the original SDF file.

To add this, I have

• Added support for individual bonds to be given properties, e.g. connectivity.edit().setProperty(bond_id, "key", value).commit()
• Added properties to represent bond types (single, double etc), stereoscopies (up, down) and atom radical type (singlet, doublet etc)
• Added in better implicit conversion of array properties, plus special classes to hold arrays of values with a single property, e.g. setProperty("key", [value0, value1, value2])
• Added in implicit conversion of dictionaries to Properties, so you can do setProperty(key, {"cat": "meow", "dog": "woof"}) etc.
• Automated the calling of wrap from the Python side. This means that objects are automatically wrapped if there isn't an existing Sire object to hold them. This is completely automatic and recursive, so you can pass in dictionaries of lists of dictionaries, mixing Sire and non-Sire objects.
• Added functions on the C++ side to make it easier to hand arbitrary properties without having to know their type. Added "propertyAsVariant" and "propertyAsProperty" member functions that can be used to extract property values from AtomProperties (and others) so that you don't need to know the type to call property<T>.
• Added convenience functions on the Python side that let you index by integer, e.g. system[i] will return the with molecule (meaning you can also write for molecule in system:. Also cleaned up the Properties API so that it fully matches the Python dictionary API and can be used like a dictionary. This includes assignment, e.g. properties["cat"] = "meow" will now work (note the cool implicit call to SireBase::wrap in the assignment :-)
• Added better logic in the ParserFactory search to find a parser to write a molecular file. If it can't find a parser that matches the primary suffix, then it will look for the first parser that has a matching secondary suffix. This lets you write "file.mol" using the SDF parser. Will also let you write "prmtop7" with AmberPrm etc.

[chryswoods]

I am working on some unit tests for this in SireUnitTests. I found that I need to add the new atom properties to the kludge so that they are correctly loaded. This means I am adding a couple more commits (as I need to use these to generate the wrappers). I will let you know when this is ready.

[chryswoods]

Ok - that's done. I had to add in the PackedArray2D wrappers, and then add the new properties to the kludge list in Mol/init.py. I have added a test to SireUnitTests, which I will expand.

[jmichel80]

Awesome, didn’t know you were working on this. @annamherz and @JenkeScheen may find this feature immediately useful for their projects

[lohedges]

Thanks, @chryswoods. This will really help with OpenForceField integration within BioSimSpace, as well as improving the robustness of our MCS mapping routines. (Both matching atoms using RDKit directly, and generating perturbation networks using LOMAP2.) The additional updates will simplify a lot of the BioSimSpace code too, e.g. selecting molecules by index, for which workarounds already exist.

I'll hop onto your branch locally, rebuild, and test at my end. I'll also make sure that BioSimSpace can be built on top of it.

Cheers.

[JenkeScheen]

Awesome work! An additional thing I can think of that we could use the SDF parser to create the interphase between BSS and RDKit as discussed previously in https://github.com/michellab/BioSimSpace/issues/190

[lohedges]

Yes, at some point I'm hoping to implement some convenience functions in the Molecule class, e.g. toRDKit(), toOpenFF(), toOpenMM(), etc. (Just need to figure out exactly what properties are needed to have a reliable outcome in each case, and what intermediates to use.)

[chryswoods]

Yes, that was my thought too. Although I think it would need to be add to the System class rather than Molecule (or, we could additionally add it to Molecule that automatically creates a one-molecule System, and then calls the function on System).

I am now planning on a new branch to create a new website for Sire using Sphinx. I will be using tutorial-driven development to write a number of Sire tutorials, and will be cleaning up some of the higher level API to make those tutorials easier to write. This will all be backwards-compatible. It will make Sire a lot easier to learn and use.

[lohedges]

Everything is working just fine at my end. The only thing that I've noticed is that, although you can now index molecules in a system by index, e.g. mol = system[-1], and loop over them, for mol in system:, Python slicing doesn't work, e.g. mols = system[3:10].

This is something that I have already implemented in BioSimSpace, so isn't really an issue. I just wondered if it was easy to support slicing in Sire too.

[chryswoods]

Yes - I can add in slicing. I will do that in feat_web as it can be part of my API modernisation as I am writing the new tutorials. This will include some further improvements, e.g. having a string index automatically converting to the appropriate XName class, and adopting "Name:Number", plus "Name[2]" to do more controlled matching.

[lohedges]

That sounds great. I'll merge this across now and work on adding the SDF support into BioSImSpace when I get a chance. (Essentially detecting when a molecule was originally loaded from SDF and bypassing some of the clunky intermeditate conversions when that is the case, e.g. PDB --> RDKit --> SDF, which is currently the most robust for OpenFF.)

Cheers.