Shape‐based similarity searching in chemical databases

Abstract

Shape similarity is a key concept and requirement for molecular recognition. As a result, much research has been undertaken to develop methods to represent molecular shape and to quantify the shape similarity between molecules. A great variety of shape descriptions and similarity comparison approaches have been developed, ranging from explicit representations using intersecting atom-centered spheres and molecular superposition to abstract statistical representations that allow alignment-free comparisons. Several of these methods have sufficient computational performance to allow shape similarity searches over extremely large compound databases as a ligand-based virtual screening (VS) technique. As with other approaches to VS, the relative performance of shape similarity methods is dataset and problem specific, with each approach having its merits and limitations and no one approach showing a clear and consistent advantage over the others. Again, as with other VS approaches, most reports of performance are in the context of retrospective validation studies, which show competitive performance against target-based and two-dimensional methods. Prospective studies are rarer in the literature, but a number of successes have been reported. Intensive research continues in the search for improved representations of shape, partial shape matching, and approaches to address the challenges imposed by ligand and target flexibility. C 2012 John Wiley & Sons, Ltd.

ZZS similarity tool: The online tool for similarity screening to identify chemicals of potential concern

“…Future improvements of the models could focus on the evaluation of more sophisticated fingerprints that possibly better define the structural aspects of chemicals, like count‐based fingerprints or 3D‐based fingerprints. Particularly, 3D‐based fingerprints could be relevant as they present a group of important descriptors for determining binding affinity as well as several other properties 28,29 . However, their use also contains challenges and uncertainties related to chemical conformations and alignments, 8,30 and more advanced fingerprints do not necessarily outperform the predictive performance of 2D‐binary fingerprints 31 .…”

“…Particularly, 3D-based fingerprints could be relevant as they present a group of important descriptors for determining binding affinity as well as several other properties. 28,29 However, their use also contains challenges and uncertainties related to chemical conformations and alignments, 8,30 and more advanced fingerprints do not necessarily outperform the predictive performance of 2D-binary fingerprints. 31 Accordingly, the currently evaluated methodology, which shows very reasonable performance, is considered adequate, especially for the proposed screening activities.…”

The Calculation of Molecular Structural Similarity: Principles and Practice

“…Third, while 2D fingerprints rapidly established themselves as the most appropriate 2D representation no such consensus has emerged thus far as to which is the most appropriate of the many different types of 3D representation that have been described. Probably the most common are measures of shape similarity,41 as exemplified by the ROCS system42 (which describes molecular shapes by Gaussian functions that permit the rapid alignment of two molecules so as to maximise their volume overlap) and the alignment‐free USR system43 (which describes molecular shapes by the moments of a set of distributions that are created by measuring the distance between bonded atoms and four reference points). Reviews of current 3D similarity measures are provided by Willett2 and by MacCuish and MacCuish 44…”

Applying Machine Learning to Ultrafast Shape Recognition in Ligand-Based Virtual Screening

“…Alignment-free methods do not require an alignment for the descriptor comparison and are generally more efficient. For a review of shape-based similarity methods please refer to Finn and Morris (2013).…”