The analysis of
biologically active substances in human test groups is more likely to yield a wide distribution with multiple differences and outlying results, research suggests.
Scientists at the University of Massachusetts Amherst, private health university UMIT and life sciences company BG Medicine examined the differences between human and animal results.
Their focus was on the assessment of
biologically active substances using 'omics' - methods of analysis such as genomics and proteomics.
"Data measured in human studies were characterised by higher biological variation and the presence of outliers," they write.
Comparing different distributions with the classifiers commonly used in such research, they found balanced, Gaussian models most compatible with Prediction Analysis for Microarrays, or PAM.
Asymmetrical distributions and unbalanced class patterns were better assessed using Random Forests as the classifier, they add.
The -omics studies they considered had all made use of multiplexed immunoassay techniques and high-content mass spectrometry as part of highly dimensional data sets.
Product directory
-(1).jpg)