Add additional fingerprints

[j-adamczyk]

We need to implement additional fingerprints from multiple sources, as we want scikit-fingerprints to be the only required Python library for computing molecular fingerprints.

RDKit hashed fingerprints:

• Avalon - ref
• E-state - ref
• FPCP - ref
• MHFP - potentially, benchmark with already implemented one, ref
• Layered - ref
• Pattern - ref
• Pharmacophore - ref
• Physiochemical property fingerprints - ref
• RDKit (substructure) - ref
• SECFP - ref

RDKit descriptor fingerprints:

• Autocorr - ref 2D, ref 3D
• BCUT - ref 1, ref 2; blocked by this RDKit issue
• GETAWAY - ref; current version works, but is nondeterministic, see this RDKit issue
• fragments (functional groups) - ref
• MORSE - ref
• MQNs - ref
• RDF - ref
• RDKit2D - ref and references from this paper
• USR, USRCAT - ref
• WHIM - ref
• VSA - ref

Check other libraries and software for fingerprints and descriptors, and add them to lists below:

• PaDEL-Descriptor - ref
• PyBioMed - ref 1, ref 2, ref 3
• MolMap - ref 1, ref 2
• ChemoPy - ref 1, ref 2
• ChemDes - ref 1, ref 2

Other descriptor-based fingerprints:

• Atom triplets - ref 1; this is very similar to atom pairs, just uses triplets of atoms, but we need to implement this from scratch, based on RDKit implementation (including atom invariants)
• Electroshape descriptors - ref 1, ref 2
• CATS descriptors - ref 1, ref 2, ref 3, ref 4; the last one includes an interesting way to use 3D distances
• CheckMol (FP3) - ref 1, ref 2, ref 3, ref 4; rejected, since it's basically covered by Laggner's SMARTS patterns
• Laggner - ref 1, ref 2, SMARTS Patterns for Functional Group Classification by Christian Laggner; also known as CDK Substructure Fingerprint in ref 3
• Ghose-Crippen - ref 1, take just the SMARTS patterns
• Klekota-Roth - ref 1, ref 2
• Lingo - ref
• Mordred - ref 1, ref 2
• PubChem - ref 1, ref 2; note that we need to implement the actual calculation of this fingerprint, not just connect to PUG REST API, since it's extremely unreliable
• SHED - ref 1, ref 2

Other fingerprints:

• 4PT - ref 1
• BCL2D - ref 1; rejected, since it's basically worse ECFP4
• FragFP - ref 1, ref 2; rejected, since it uses custom search patterns engine and it's not really doable to translate them to SMARTS
• Graph signature - ref 1, ref 2; originally proposed in ref 3 and ref 4
• MDFP - ref
• MolPrint2D - ref; rejected, since it's basically worse ECFP4
• NC-MFP - ref, RDKit code is provided, but it requires quite a bit of refactoring
• Spectrophore - ref
• Toxicophore - ref 1, ref 2, SMARTS queries are in supporting materials; note that this is not a unique definition, and e.g. OChem ToxAlerts offers (much larger) alternative