Evaluation of microelectrode array data using Bayesian modeling as an approach to screening and prioritization for neurotoxicity testing

• Published in: Neurotoxicology (Park Forest South) 2013-05, Vol.36, p.34-41
• Main Authors: LeFew, William R., McConnell, Emma R., Crooks, James L., Shafer, Timothy J.
• Format: Article
• Language: English
• Subjects: Action Potentials - physiology; Animals; Bayes Theorem; Biological and medical sciences; Chemical screening; Data analysis; Drug Evaluation, Preclinical; Humans; Medical sciences; Microelectrode array; Microelectrodes; Models, Neurological; Neurons - physiology; Neurotoxicity; Neurotoxicity Syndromes - diagnosis; Prevention and actions; Public health. Hygiene; Public health. Hygiene-occupational medicine; Toxicology
• ISSN: 0161-813X; 1872-9711
• DOI: 10.1016/j.neuro.2013.02.006

► MEA data from cortical cultures treated with 30 chemicals were analyzed using a Bayesian approach. ► The probability of chemical effect on firing rate distributions across electrodes was determined. ► These probabilities can be used to determine chemical “hits” and prioritize chemicals for testing. ► The Bayesian approach was less robust at detecting hits than the weighted mean firing rate. ► The weighted mean firing rate and Bayesian combined correctly classified >95% of chemicals. The need to assess large numbers of chemicals for their potential toxicities has resulted in increased emphasis on medium- and high-throughput in vitro screening approaches. For such approaches to be useful, efficient and reliable data analysis and hit detection methods are also required. Assessment of chemical effects on neuronal network activity using microelectrode arrays (MEAs) has been proposed as a screening tool for neurotoxicity. The current study examined a Bayesian data analysis approach for assessing effects of a 30 chemical training set on activity of primary cortical neurons grown in multi-well MEA plates. Each well of the MEA plate contained 64 microelectrodes and the data set contains the number of electrical spikes registered by each electrode over the course of each experiment. A Bayesian data analysis approach was developed and then applied to several different parsings of the data set to produce probability determinations for hit selection and ranking. This methodology results in an approach that is approximately 74% sensitive in detecting chemicals in the training set known to alter neuronal function (23 expected positives) while being 100% specific in detecting chemicals expected to have no effect (7 expected negatives). Additionally, this manuscript demonstrates that the Bayesian approach may be combined with a previously published weighted mean firing rate approach in order to produce a more robust hit detection method. In particular, when combined with the weighted mean firing rate approach, the joint analysis produces a sensitivity of approximately 96% and a specificity of 100%. These results demonstrate the utility of a novel approach to analysis of MEA data and support the use of neuronal networks grown on MEAs as a for neurotoxicity screening approach.