A common annoyance when using mid-infrared spectrometers is the sensitivity of the technique to water vapour along the beam path. To combat this, manufacturers will use desiccants or design purge systems with inert gases (such as nitrogen or dry air) however, all too often these systems will fail and water vapour from the atmosphere or other sources will find its way into the system.
Consider a spectrometer that has a multi-position setup to automatically measure 96-well silicon plates. This kind of instrument can be used to sample and measure microlitre volumes of biological fluids with an aim to create a simple, cheap and effective way to detect diseases by means of a multivariate data model. A caveat to the setup is that the user must specify if they want to collect a background spectrum before measuring every position (essentially a doubling of measurement time) or if they would like to use a single background spectrum, collected prior to measuring samples, and use it for every sample measured on the plate. Another caveat to the instrument is that the manufacturer of the silicon plates used decided to change the etching protocol resulting in two distinct designs of surface microstructure. According to the instrument manufacturer, these new plates pass design specifications and shouldn’t cause any problems.
Faced with these decisions, a spectroscopist designs a series of test, methodology experiments to study the effects of choosing one over the other in conjunction with other variables such as the number of scans used and differing humidity levels. For the purposes of this example, a 1 microlitre aliquot of the same stock blood plasma solution has been loaded on every position of a plate (except one for the background), for each etch design, and allowed to dry at room temperature. The humidity within the instrument has been controlled to measure the same samples at an instrument hygrometer reading of 0%, 10% and 20% relative humidity (RH). A further caveat to this value is that the instrument hygrometer records at intervals of 10% only hence a reading of 0% could be 0-9% whilst a reading of 10% could be a 10-19% RH.
When a background is selected prior to each measurement the change in humidity is still a major source of variation in the data which would be problematic when constructing a predictive model.