Correlations are abundantly used in the life sciences and other areas of science. They are easy to calculate and understand. However, correlations based on comprehensive “omics” measurements - such as RNAseq, proteomics or metabolomics - are not that easy at all. The measurement error structure of such omics data is complex and affects the underlying correlations dramatically. Age Smilde from the Swammerdam Institute for Life Sciences of the UvA (SILS-UvA), together with Edoardo Saccenti (WUR Wageningen) and Margriet Hendriks (DSM Delft) explain their findings in the scientific journal Scientific Reports.
The use of correlation coefficients can be limited to simple experimental analysis, but they are also basic ingredients for sophisticated multivariate (or “big”) data analysis methods. It is therefore important to have reliable estimates for correlation coefficients. In current-day life sciences, comprehensive measurement techniques are used to measure all kinds of complex data, such as metabolites, proteins, gene-expressions and more. All these measurement techniques have errors. Whereas in the old days, with simple measurements, the errors were also simple, that is not the case anymore. Errors are heterogeneous, non-constant and not independent. This hampers the quality of the estimated correlation coefficients seriously and it can even change the sign of the ‘true’ biological correlation. If this is the case, it can severely affect the outcome of an experiment.
In their findings, the researchers discuss the different types of errors as present in modern comprehensive life science data and show with theory, simulations and real-life data how these affect the correlation coefficients. They also opt ways to improve the estimation of such coefficients. Based on their analysis, the researchers warn that in general correlation coefficients are trusted too much on face value and they hope to have triggered some doubts and pointed to precautions regarding the use of correlations for future experiments.