Representing molecules with minimalism: A solution to the entropy of informatics
Cheminformatics as we know it is possible because so many molecular structures can be represented with datastructures and rules that are at first glance quite trivial. This first impression is highly misleading, since even within supposedly well behaved domains, edge cases arising from issues such as resonance, tautomerization, symmetry and stereochemistry - to name but a few - quickly add up. To supplement these genuine challenges, there is a whole additional class of problems caused by the mismatch between chemists' understanding of molecules and the datatypes that are necessary to capture a structure for informatics purposes. This line is blurred by the convenience of representing structures in a form that is very closely related to the diagram styles that have been in use since the dawn of chemistry. There are currently four major approaches to structure representation: connection tables (e.g. MDL Molfile), sketches (e.g. ChemDraw), canonical strings (e.g. SMILES and InChI) and atomic models (numerous 3D formats). Not only do all of these approaches have valid use cases, but they are deceptively incompatible with each other, even when addressing identical needs. Almost without exception, format conversions are not commutative, and every translation involves losing some amount of data. Given that recording chemical structures in machine readable form has become such a critical part of scientific research, it is essential to define a fundamental representation that captures the key structural definition asserted by the experimental chemist, for a broad and useful range of molecules, and ideally in a way that is closely related to visual drawing mnemonics. The number of data concepts needed to satisfy these conditions is quite small, and is mostly satisfied by the most commonly used subset of the venerable MDL Molfile format. This presentation will discuss how this subset, with a few minor corrections and clarifications, can and should be used as the reference standard for molecules, and how the informatics community can benefit from having well defined standards.
Recommended
More Related Content
Similar to Representing molecules with minimalism: A solution to the entropy of informatics
Similar to Representing molecules with minimalism: A solution to the entropy of informatics (20)
More from Alex Clark
More from Alex Clark (20)
Recently uploaded
Recently uploaded (20)
Representing molecules with minimalism: A solution to the entropy of informatics
- 1. Representing molecules with minimalism: A solution to the entropy of informatics Alex M. Clark ⚡ COLLABORATIVE DRUG DISCOVERY
- 2. Introduction ◈ InChI is great if... ▷ applied within its domain ▷ used for the problems it is designed to solve ▷ supplied with appropriate data 2 J. Cheminf. 7:9 (2015)
- 3. Encoding Types ◈ 1D ◈ 2D ◈ 3D 3 identifier (InChI, SMILES) conformation (QC, MM, crystals) diagram (ChemDraw etc.) representation (Molfile) optimised for presentation optimised for information
- 4. Sketch vs. Cheminformatics ◈ Most cheminformatics data comes in through sketches: ▷ chemists use software designed to match conventions dating back to 19th century ◈ Software is given a choice: 1. capture the stylistic choices (diagram primacy) 2. capture the fundamental meaning (data purity) ◈ We need to do both... 4
- 5. Minimalism ◈ Want to describe the intention of the scientist who has made a discovery ▷ capture the fact without destroying information ▷ display it faithfully and recognisably ▷ compare to other molecules ▷ manipulate easily with many software tools ◈ Start with extremely minimalistic primitives... ◈ ... extend that with semi-optional metadata 5
- 6. Primitives ◈ Connection table style closely aligned with editor tools ◈ Atoms ▷ Symbol ▷ Isotope ▷ Charge (integral) ▷ Virtual hydrogens (explicit vs. computed) ▷ X,Y coordinates ◈ Anything else is a nice-to-have... 6 ◈ Bonds ▷ From, To ▷ Order (0, 1, 2, 3, 4, ...) ▷ Wedge: quasi-3D (stereochemistry in situ)
- 7. Practical Solution ◈ An actual format is necessary... ◈ ... MDL Molfile format is the ad hoc standard ▷ originally poorly specified, though now much better ▷ but too late: literally thousands of scripts/products ▷ doesn't really matter what the spec says (cf. V3000) ◈ Effectively the lowest common denominator feature set of software in current use... ◈ ... need to expand this subset, while deprecating. 7
- 8. OpenMolecule Idea ◈ Determine a level of nonstandard features necessary ▷ level 1.0: features that every MDL Molfile reader & writer can handle with no information loss ▷ level 1.x: increasing likelihood of confusion ◈ Lowest level possible ▷ mark internally ▷ or quarantine ◈ Keep using Molfiles compatibly until replacement... 8
- 9. OpenMolecule Idea ◈ Determine a level of nonstandard features necessary ▷ level 1.0: features that every MDL Molfile reader & writer can handle with no information loss ▷ level 1.x: increasing likelihood of confusion ◈ Lowest level possible ▷ mark internally ▷ or quarantine ◈ Keep using Molfiles compatibly until replacement... 8
- 10. Level 1.0: Minimal ◈ Lowest common denominator of all readers/writers ▷ the drug-like subset of organic chemistry ◈ To guarantee round-trip survival: ▷ Atoms are atomic symbols ▷ Bonds are 1, 2, 3 ▷ Chirality by wedges (not parity) ▷ E/Z always specified by coordinates ▷ Implied hydrogens never in doubt 9
- 11. Level 1.1: Bigger ◈ 1000 atoms / bonds is a Molfile V2000 limit ◈ V3000 fixes this ◈ Limit rarely exceeded for most use cases ◈ V3000 is ugly, adoption slow ◈ Compliance means reading/writing V3000 10
- 12. Level 1.2: H-control, zero bonds ◈ Need to override virtual hydrogens and define a strict rule for implicit hydrogen counting ◈ Need a bond type that doesn't affect valence counting: allow 0-order ◈ Minimal changes - can describe a lot of inorganic chemistry - see later for beautification 11 (Ⅱ) (Ⅳ) J. Chem. Inf. Model. 52, 3149-3157 (2011)
- 13. Inorganics 12
- 14. Inorganics 12
- 15. Inorganics 12
- 16. Inorganics 12
- 17. Molecule Metadata ◈ Metadata = optional information ▷ can beautify the structure, or assist interpretation ▷ but core representation is still meaningful & renderable ◈ Can help interpretation and/or aesthetics, e.g. 13 resonance & symmetry preferred visual style
- 22. Stereochemistry ◈ Best to use 2D coordinates + wedges to derive... ◈ ... label/parity is metadata 15 S R S R S S R R cis trans (R)-2,2′-Bis(bromomethyl)-1,1′-binaphthyl
- 23. Abbreviations ◈ Using abbreviations instead of all heavy atoms must be accompanied by a definition... ◈ ... everyone has their own ◈ Popular but nice-to-have in drug discovery; essential elsewhere ◈ Support it via S-group in Molfile, inline abbreviations in SketchEl ◈ Must be able to expand to get all atom 16
- 24. Enumeration Rules ◈ Formulae for multiple structures are not safe for naïve interpretation (exception to minimalist approach) 17 Mixtures xylenes
- 25. Enumeration Rules ◈ Formulae for multiple structures are not safe for naïve interpretation (exception to minimalist approach) 17 Mixtures xylenes Markush R1 = H, CH3 R2 = Me, Et, Pr, Bu R3 = Ph, tolyl X = N, CH Y = C, N+ Z = Cl, CH2Cl ( ) n
- 26. Enumeration Rules ◈ Formulae for multiple structures are not safe for naïve interpretation (exception to minimalist approach) 17 * htn htn* Cl ran Polymers Mixtures xylenes Markush R1 = H, CH3 R2 = Me, Et, Pr, Bu R3 = Ph, tolyl X = N, CH Y = C, N+ Z = Cl, CH2Cl ( ) n
- 27. Future Work ◈ Slumming it with molfiles: ▷ agree on valid extensions ▷ self-awareness of limitations ◈ Lossless conversions between different formats ◈ Chart a course away from Molfiles... 18