Evaluation of bioimage normalization strategies to quantify morphogen gradients

Xiaoying Yang, Purdue University

Abstract

Morphogenic molecules play critical roles in directing pattern formation in organisms. Morphogens form spatially non-uniform profiles in tissues and cells are instructed based on thresholds of concentration in the profiles. The numerical profiles usually need to be converted from experimental data, such as images, to common scale for proper comparison. Three normalization methods, Minimum Variance, Extrema Pinning, and Fixed Integral, are commonly used in the conversion processes and their efficacies have never been evaluated. In this study, we compare these methods using computationally simulated data. We constructed a mathematical model to simulate the production, diffusion, and decay behavior of a morphogen to generate 'ground truth' data that contains the true quantitative information of the molecule that cannot be easily acquired experimentally. We then added noise to represent the various errors produced in experiments and processed the noisy data as we would with real experimental data. Statistical analysis, including significance testing and sample size estimation, were performed on data before and after normalization. We found that Minimum Variance is the most efficient method in capturing true data profiles and detecting incremental differences. Extrema Pinning and Fixed Integral, however, exhibit weakness in data distortion and high demand for samples, respectively. Our results indicate that extra care must be taken in experiments in choosing the most efficient method for data comparison.

Degree

M.S.

Advisors

Umulis, Purdue University.

Subject Area

Biomedical engineering

COinS