The SHE assay (pH 6.7) is being considered as a ‘3Rs’ alternative
in animal laboratory studies (1). We have previously developed
a protocol to conduct Fourier-transform infrared spectroscopy
in the Syrian hamster embryo (FTIRS-SHE) experiments, and
corresponding software to build up a FTIRS-SHE database.
Subsequently, we applied machine learning and statistical methods
to analyse our datasets towards chemical-treatment classification,
morphological transformation classification, and extraction
of biomarkers (i.e. spectral wavenumbers) related to chemical
treatment (2). In the present study, we set out to validate and
develop further our biomarker extraction techniques. Biomarker
validation is of extreme importance, for it was found that depending
on different biomarker extraction methods (i.e. computational
algorithms), there was marked variability in the subsequently
identified discriminating biomolecular entities and this would
inevitably give rise to different mechanistic interpretations.
Furthermore, currently a number of techniques used for such
biomarker extraction purposes employed in a variety of fields
were never initially conceived with this intention. In this work,
we compare different techniques used to extract biomarkers and
present rationales for their possible disagreement. We recommend
an analysis framework that can derive robust biomarkers for the
FTIRS-SHE assay based on pattern classification. The application
of our framework can be extended to other studies that use FTIR or
Raman spectroscopy. This work was funded by Unilever and the
SHE assays were conducted at BioReliance, USA.