Do Structurally Similar Molecules Have Similar Biological Activity?

Abstract

To design diverse combinatorial libraries or to select diverse compounds to augment a screening collection, computational chemists frequently reject compounds that are > or =0.85 similar to one already chosen for the combinatorial library or in the screening set. Using Daylight fingerprints, this report shows that for IC(50) values determined as a follow-up to 115 high-throughput screening assays, there is only a 30% chance that a compound that is > or = 0.85 (Tanimoto) similar to an active is itself active. Although this enrichment is greater than that found with random screening and docking to three-dimensional structures, this low fraction of actives within similar compounds occurs not only because of deficiencies in the Daylight fingerprints and Tanimoto similarity calculations but also because similar compounds do not necessarily interact with the target macromolecule in similar ways. The current study emphasizes the statistical or probabilistic nature of library design and that perfect results cannot be expected.

Similarity-based virtual screening using 2D fingerprints

“…This approach was rapidly taken up and there is now a very large body of evidence that supports the use of fingerprint-based measures for VS [10][11][12][13][14][15][16][17][18]. It must be emphasised that many other types of structural representation have been suggested for the computation of molecular similarity, including physicochemical properties, chemical graphs, topological indices, 3D pharmacophore patterns and molecular fields inter alia [19][20][21][22].…”

Evaluation of machine-learning methods for ligand-based virtual screening

“…Studies in the mid-Nineties explored the extent to which different types of structural descriptor, both 2D and 3D, encoded sufficient structural information to enable the prediction of a range of physicochemical properties [33,34]; more recent work has demonstrated the extent to which one such representation, Daylight 2D fingerprints, enables the prediction of bioactivity in HTS systems [35]. Here in Sheffield, we have had a long-standing interest in the use of similarity and machine-learning methods, principally using 2D fingerprint representations, and in this paper, we discuss briefly some of our recent work in the latter area.…”

Analysis and use of fragment-occurrence data in similarity-based virtual screening

“…The Similar Property Principle [11,18,55] would lead us to expect that the active molecules in an activity class are likely to be more similar to an active reference structure from that class than are the inactive molecules (although there are, of course, many exceptions to this generalisation). Thus, if we plot the frequency distributions for the similarities between the reference structure and the set of active molecules, and the similarities between the reference structure and the set of inactive molecules (which we shall refer to as the Actives distribution and the Inactives distribution, respectively) then we would expect a plot such as that shown in Figure 2a, which is based on the M22 measure.…”

“…A fingerprint is clearly an extremely simple type of structural representation, but still contains sufficient information to enable effective similarity-based virtual screening to be carried out (see, e.g., [11][12][13][14][15][16][17]). Here, the similarity is computed between a reference structure of known biological activity and each of the structures in a database; the most similar structures -the nearest neighbours -are then prime candidates for biological screening.…”