Analysis of Neighborhood Behavior in Lead Optimization and Array Design

Abstract

Neighborhood behavior describes the extent to which small structural changes defined by a molecular descriptor are likely to lead to small property changes. This study evaluates two methods for the quantification of neighborhood behavior: the optimal diagonal method of Patterson et al. and the optimality criterion method of Horvath and Jeandenans. The methods are evaluated using twelve different types of fingerprint (both 2D and 3D) with screening data derived from several lead optimization projects at GlaxoSmithKline. The principal focus of the work is the design of chemical arrays during lead optimization, and the study hence considers not only biological activity but also important drug properties such as metabolic stability, permeability, and lipophilicity. Evidence is provided to suggest that the optimality criterion method may provide a better quantitative description of neighborhood behavior than the optimal diagonal method.

Local neighborhood behavior in a combinatorial library context

“…It has recently [1] been pointed out that these indices may yield results that are more intuitive to a chemist's expectation of virtual screening quality than other, more artificial NB scoring schemes, and that a related index (the Cohen j statistics [33], unknown to us when designing X) had already been used in cognitive sciences. • It allows for an unambiguous definition of the optimal dissimilarity radius, reflecting the relative importance of FS versus PFD.…”

“…The global optimality criterion [1,3] was estimated on the basis of all the compound pairs m,M in the data set (excluding the ones in which both m and M were inactives), and plotted with respect to the dissimilarity cutoff d in order to determine the optimal dissimilarity radius d* minimizing X(d*). Alternatively, the ''local'' NB monitoring scheme advocated here consists in estimating, for each active molecule M, its local optimality score X L (d*) specifically over the N-1 pairs excluding the monitored active M-later on referred to as the virtual screening ''query''.…”

“…Neighborhood Behavior (NB) [1][2][3][4], the relationship between structural similarity of molecules and their experimental properties, is a quantitative expression of the classical similarity principle [5,6] The aim of this article is to revisit the NB issue from a ''local'' point of view in the context of a combinatorial library that was tested on five different proteases. The size and quality (experimental homogeneity) of the data qualify it as an opportunity to investigate the behavior of similarity-based virtual screening tools in the context of combinatorial library design.…”

ISIDA Property‐Labelled Fragment Descriptors

“…[52] Similar "hits" are selected at the dissimilarity cut-off maximising the Local Ascertained Optimality Score (LAOS), a "noise-free" variant, [42] focusing on individual active queries, of the neighbourhood behaviour optimality score. [9,11,50] It is characteristic of the considered chemical space (defined as a combination of descriptors and dissimilarity calculation rule, or metric), within the specific context (given query compound, considered target) of the virtual screening experiments. For each reported active of every considered target, similarity-based retrieval of active analogues is performed within every considered descriptor space (i.e.…”

“…[1][2][3][4] Standard methods include quantitative structure-activity relationships (QSAR) [5][6][7][8] and similarity-based virtual screening. [9][10][11][12] However, although an enormous amount of descriptors [13][14][15][16] are available and data mining [17][18][19] methods are implemented for chemoinformatics tasks, significant differences of activity still arise among molecules perceived as similar by the computational tool (activity cliffs). [20,21] Activity cliffs, apparently violating the similarity principle "similar molecules are likely to have similar properties", are a complex problem, and it is not straightforward to distinguish the "genuine" situations from the cases related to an inappropriate chemical space and activity landscape definition (inappropriate molecular representation, i.e.…”