Dear All
In my case, I have FTIR datasets of cancer cells and cancer cells after treatment. Measuring of them was done in reflectance mode using Nicolet 8700 FTIR Spectrometer with Nicolet Continuum IR Microscope (Thermo Fisher Scientific Inc., USA). The system was equipped with a liquid nitrogen cooled mercury-cadmium-telluride (MCT/A) detector and infinity corrected 15x Reflachromate Objective. Apodization was performed using a Black-Harris apodization function. Each sample of the cells was triplicated and transferred onto the three Low-e glasses. Spectra were collected from the three different most homogeneous sample zones of each Low-e glass. A total of 128 scans with the spectral resolution of 4 cm-1 was set up for both background and sample collection. In the case of cells, the beam was focused to the center of the cell. Approximately 100-150 individual cells were measured of each sample at the spectral range of 4000-650 cm-1. All the data was collected and partially prepared by Omnic 8 (Thermo Fisher Scientific Inc.) software. After, these datasets were exported to Unscrambler software.
I prepared three section for analysis: 3020-2820, 1770-1480 and 1180-950. So, these datasets were preprocessed in Unscrambler such way for each section: 1. SNV; 2. Baseline correction (Baseline offset, Linear baseline correction), 3. 1th Derivative (Savitzky-Golay algorithm using 6 points of smoothing, 2nd polynomial order). After I made a PCA analysis.
As I saw after, I have different spectra in cancer cells dataset. How to identify the good specters in each dataset before preprocessing and calculation of PCA analysis? I need advice, how to do the preprocessing correct.... Thank you.
Hi,
In my own work with Unscramble I would often avoid using Baseline Correction as if the correct points in the actual baseline are not chosen it can really offset the spectra in an undesired way. Off hand (from what I've read) it could be that you might have taken a variable in the baseline correction that is associated with a particular group (the cancer cell dataset).
If you are using SNV and/or any derivative doing a baseline correction is a moot point as both SNV and any derivative should do this.
If you run a PCA you should be able to see the regions of the spectra that created the model. If you get a scattered loadings plot by default you might need to right click on the loadings plot go into the loading tab and click line instead of 2D, now you should be able to flick through the PCs see the regions of the spectra that are used in making that particular PC.
I hope this helps, feel free to message me if you have any more questions or if I completely missed the mark of you question.
All the best,
Pól
Dear Vitalli,
First I don't understand why you split the spectral domain in three and why you apply different preprocessing on them. Second if your experiment is well organised (and it seems to be) there are no good and bad spectra. All spectra have to be used in PCA in order to evaluate deferences between groups but also within groups.
It will be a pleasure for me to help you more in this task. Don't hesitate to contact me by mail if you need help.
Regards,
Ludovic.
- Badges
- Science topic
- Similar topics
- Medicine
- Oncology
- Cancer Research
- Cancer Biology
- Cancer Cells