Conformational Soft Hash (CSH): A Rotation-Invariant Molecular Fingerprint for Conformation-Sensitive Similarity Search and Blockchain-Based Conformational Registry

Many molecular descriptors are currently available, but each has its limitations. One group, such as the Morgan and ECFP fingerprints, encodes only the connections between atoms. Those that do take the 3D structure into account either reduce the geometry to a few summary values, as USR, USRCAT and 3DAPfp do, or describe only the environments of individual atoms, as E3FP does. The present work uses the Conformational Soft Hash (CSH), a simple method for representing a 3D conformation as a 256-bit binary hash. Each conformer is described by a feature vector containing the distances between pairs of heavy atoms at least three bonds apart, together with the sine and cosine of each torsion angle. Because interatomic distances and trigonometric functions span different ranges of values, they were brought to a common scale. The length of the vector varied with the size of the molecule, so its dimensionality was reduced to 32 components and the result was converted into a 256-bit string by SimHash, yielding the CSH. CSH was compared with USR, USRCAT, 3DAPfp and E3FP on 28 drug molecules and on a separate set of twelve approved drugs of varying size and flexibility, selected after the method had been finalised. For every pair of conformers of a molecule, the similarity of their hashes was correlated with the geometric difference between them, measured as the root-mean-square difference between their interatomic distance matrices (DM-RMSD). At a common budget of 256 bits, CSH reached a mean Spearman correlation of ρ = 0.508 on the first set and 0.637 on the second. It performed better than USR, while the results for USRCAT and 3DAPfp were similar. 3DAPfp has the advantage, however, of reaching the same accuracy without a transformation fitted to each molecule, which allows different molecules to be compared, something CSH does not offer. E3FP, on the other hand, encodes local atom environments rather than global shape and reflected the geometric differences between conformers poorly. A conformational registry was also implemented on the Ethereum blockchain, which timestamps an identifier of a conformer on a public chain without disclosing its coordinates. Eighty conformers of four molecules were recorded. Records belong to the registry of a particular molecule, so all of its hashes can be retrieved at once, browsed and compared with one another. No registry of this kind has been described before for molecular conformations.