A three-stage algorithm to make toxicologically relevant activity calls from quantitative high throughput screening data - PubMed (original) (raw)

A three-stage algorithm to make toxicologically relevant activity calls from quantitative high throughput screening data

Keith R Shockley. Environ Health Perspect. 2012 Aug.

Abstract

Background: The ability of a substance to induce a toxicological response is better understood by analyzing the response profile over a broad range of concentrations than at a single concentration. In vitro quantitative high throughput screening (qHTS) assays are multiple-concentration experiments with an important role in the National Toxicology Program's (NTP) efforts to advance toxicology from a predominantly observational science at the level of disease-specific models to a more predictive science based on broad inclusion of biological observations.

Objective: We developed a systematic approach to classify substances from large-scale concentration-​response data into statistically supported, toxicologically relevant activity categories.

Methods: The first stage of the approach finds active substances with robust concentration-response profiles within the tested concentration range. The second stage finds substances with activity at the lowest tested concentration not captured in the first stage. The third and final stage separates statistically significant (but not robustly statistically significant) profiles from responses that lack statistically compelling support (i.e., "inactives"). The performance of the proposed algorithm was evaluated with simulated qHTS data sets.

Results: The proposed approach performed well for 14-point-concentration-response curves with typical levels of residual error (σ ≤ 25%) or when maximal response (|RMAX|) was > 25% of the positive control response. The approach also worked well in most cases for smaller sample sizes when |RMAX| ≥ 50%, even with as few as four data points.

Conclusions: The three-stage classification algorithm performed better than one-stage classification approaches based on overall F-tests, t-tests, or linear regression.

PubMed Disclaimer

Conflict of interest statement

The author declares he has no actual or potential competing financial interests.

Figures

Figure 1

Figure 1

Three-stage algorithm used to classify the activity of a substance from normalized qHTS data. The tree is defined by stages (circles), where the result of each stage determines the next stage to apply. The process continues until the path terminates in a call (rectangles). The number in the brackets designates the direction of the assay as described in the text (“+” refers to activation; “–“ refers to inhibition).

Figure 2

Figure 2

Example response profiles from experimental data obtained within Tox21 qHTS studies. _p_-Values shown are from the overall _F_-test using the nonlinear least squares approach (_p_F.NLS), the overall _F_-test using the weighted nonlinear least squares approach (_p_F.WLS), Student’s _t_-test comparing the mean response to 25% response followed by comparison to –25% response in parentheses (_p_t.student), and a weighted _t_-test comparing the mean response to 25% response followed by comparison to –25% response in parentheses (_p_t.weighted). Activity calls resulting from the proposed algorithm are indicated on the figure.

Figure 3

Figure 3

Contour plots to evaluate classification performance of proposed approach to make activity calls from 14-point concentration–response curves. The plots summarize the performance characteristics of the proposed classification algorithm based on AUC of the ROC curve generated from a broad parameter space of |RMAX| and AC50 under different residual error scenarios. Regions of each plot with AUC ≥ 0.75 indicate moderately good performance, and regions with AUC > 0.9 represent excellent performance. The significance level for statistical tests is 0.05.

Figure 4

Figure 4

Case 2 ROC curves for different parameter configurations for σ = 25% error. Sensitivity versus (1 – Specificity) are plotted for 63 different parameter configurations of AC50 (0.001, 0.1, 10 μM), |RMAX| (25%, 50%, 100%), and SLOPE (0.01, 0.1, 0.5, 1, 2, 10, 100) for R0 = 0. The diagonal line indicates random performance. The significance level for statistical tests is 0.05.

References

    1. Collins FS, Gray GM, Bucher JR. Toxicology. Transforming environmental health protection. Science. 2008;319:906–907. - PMC - PubMed
    1. Cui X, Hwang JT, Qiu J, Blades NJ, Churchill GA. Improved statistical tests for differential gene expression by shrinking variance components estimates. Biostatistics. 2005;6:59–75. - PubMed
    1. Fawcett T. An introduction to ROC analysis. Pattern Recognit Lett. 2006;27:861–874.
    1. Hill AV. The possible effects of the aggregation of the molecules of haemoglobin on its dissociation curves. J Physiol. 1910;40:4–7.
    1. Huang R, Southall N, Cho M-H, Xia M, Inglese J, Austin CP. Characterization of diversity in toxicity mechanism using in vitro cytotoxicity assays in quantitative high throughput screening. Chem Res Toxicol. 2008;21:659–667. - PMC - PubMed

Publication types

MeSH terms

Substances

LinkOut - more resources