Data Mining the NCI60 to Predict Generalized Cytotoxicity

Abstract

Elimination of cytotoxic compounds in the early and later stages of drug discovery can help reduce the costs of research and development. Through the application of principal components analysis (PCA), we were able to data mine and prove that ∼89% of the total log GI 50 variance is due to the non-specific cytotoxic nature of substances. Furthermore, PCA led to the identification of groups of structurally unrelated substances showing very specific toxicity profiles, such as a set of 45 substances toxic only to the Leukemia_SR cancer cell line. In an effort to predict non-specific cytotoxicity based on the mean log GI 50 , we created a decision tree using MACCS keys that can correctly classify over 83% of the substances as cytotoxic/non-cytotoxic in silico, based on the cutoff of mean log GI 50 = −5.0. Finally, we have established a linear model using least squares in which 9 of the 59 available NCI60 cancer cell lines can be used to predict the mean log GI 50 . The model has R 2 =0.99 and root mean square deviation between the observed and calculated mean log GI 50 (RMSE) = 0.09. Our predictive models can be applied to flag generally cytotoxic molecules in virtual and real chemical libraries, thus saving time and effort.

The slow cell death response when screening chemotherapeutic agents

“…Unfortunately, the SRB assay is widely described as a “cytotoxicity assay” [2, 15, 18, 19] which, based on the lack annexin and vital fluorochrome binding seen with the no cell death response, can be erroneous. Beyond the terminological issue, SRB assay curves can be interpreted as defining cytotoxicity when high concentrations of an agent produce a final cell mass which is below the initial cell mass (see Fig.…”

Network-based characterization of drug-regulated genes, drug targets, and toxicity

“…Several computational methods [26,42,43,48] can predict cytotoxicity, but rely on data unavailable for most substances. We propose that topological properties of drug-affected genes (e.g., degree in PPI networks) may be valuable as additional predictive variables.…”

“…NCI60 data consist of GI50 levels (drug concentrations required for 50% growth inhibition) for drugs tested in 59 human cancer cell lines. We defined the cytotoxicity of a drug to be its mean -log(GI50) across cell lines, as in previous studies [9,43].…”

Two-Dimensional (2D) In Silico Models for Absorption, Distribution, Metabolism, Excretion and Toxicity (ADME/T) in Drug Discovery

“…57]. The methods to assign a cytotoxicity score or to classify whether a compound is cytotoxic / noncytotoxic include neural networks, proteomic profiling and QSAR [57].…”

“…57]. The methods to assign a cytotoxicity score or to classify whether a compound is cytotoxic / noncytotoxic include neural networks, proteomic profiling and QSAR [57]. Using principal component analysis (PCA), NCI60 GI 50 (concentrations required to inhibit growth by 50% using a panel of NCI60 cell lines) data was mined and the authors proved that 89 % of the total variance in the GI 50 data was due to the nonspecific cytotoxic nature of the substances.…”