Display and Analysis of Patterns of Differential Activity of Drugs Against Human Tumor Cell Lines: Development of Mean Graph and COMPARE Algorithm

Abstract

The objective of this study was to develop and investigate an approach to optimally detect, rank, display, and analyze patterns of differential growth inhibition among cultured cell lines. Such patterns of cellular responsiveness are produced by substances tested in vitro against disease-oriented panels of human tumor cell lines in a new anticancer screening model under development by the National Cancer Institute. In the first phase of the study, we developed a key methodological tool, the mean graph, which allowed the transformation of the numerical cell line response data into graphic patterns. These patterns were particularly expressive of differential cell growth inhibition and were conveniently amenable to further analyses by an algorithm we devised and implemented in the COMPARE computer program.

Mining and Visualizing Large Anticancer Drug Discovery Databases

“…The meanzero preprocessing procedure 16 seemed to eliminate the noninformative "inherent" cytotoxicity, thus bringing out the informative differential cell responses. The same conclusion was also drawn from a series of PCA score plots.…”

“…[16][17][18][19][20][21][22][23][24] Similarity in activity patterns very often indicates a similarity in the mechanism of action, mode of drug resistance, and molecular structure of tested compounds. 16,20 Similarly, each column of A can be regarded as the sensitivity profile of a particular cancer cell line to the large number of screened chemicals. Therefore, we can assess the similarity of cancer cell lines in terms of their responses to thousands of chemicals, as will be discussed later.…”

“…Backpropagation neural networks, 20 Kohonen self-organizing maps, 22 principal component analysis (PCA), 21,23 and cluster analysis 18,23 have been used to predict mechanisms of action or to organize compounds into families based on their activity patterns. This "information-intensive" approach to the molecular pharmacology of cancer and anticancer drug discovery 16,[18][19][20]24,50 has proved useful in identifying subgroups of compounds with an apparent relationship to particular biological targets, e.g., the identification of "p53-inverse" agents. 18,23,57,58 We previously investigated the relationship between the A and T matrices.…”

“…16,17,20,21 (b) Compounds similar in mechanism of action tend to cluster together, even when they possess diverse chemical structures. 16,17,20,21 (c) Compounds apparently similar in chemical structure but actually different in mechanism of action are distant from each other in the cluster tree. 23 These observations clearly corroborate the finding that, although activity data for a single cell line are not informative about the mechanism of action, activity patterns generated over the 60 cell lines encode incisive information about the mechanism.…”

“…7,8 More recently, Telik, Inc. (South San Francisco, CA), developed a method for predicting ligand binding to proteins by affinity fingerprinting from a small panel of reference proteins. [9][10][11] After preliminary testing of over 300 proteins from a variety of sources, the authors selected panels of [8][9][10][11][12][13][14][15][16][17][18] proteins that displayed the broadest binding affinities for a set of over 5000 compounds. MLR was used to build "computational surrogates" to predict the binding potencies of additional molecules.…”