Evidence-based methods in studies of biology and data analysis
Embargo until
Date
2018-04-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
When scientists use familiar data analysis methods out of comfort or convenience, disciplines can suffer in their scientific inferences if these methods are not appropriate for their ultimate goals. Older fields experience this when long-standing methods are used simply for their longevity. Newer fields experience this when scientists transfer methods from other areas without evaluating their performance in these new domains. This work represents a collection of methods and results that contribute to evidence-based analytical practice in three different domains: mass spectrometry-based metabolomics, massively parallel reporter assays, and data science training. In the first two domains, we present new methods that improve current practice for comparative (differential) analysis in those fields. Specifically these methods are shown to be statistically calibrated and powerful compared to existing alternatives. In the third domain, we present experimental results regarding the actions and perceptions in data analysis practice. These results have implications for data analysis training and education. Broadly, in these three domains, we provide tools and discuss findings that enable higher quality work in applied research.
Description
Keywords
metabolomics, mass spectrometry, massively parallel reporter assays, evidence-based data analysis, human-data interaction