Matlab software version r2012a

All FFPE samples contained at least 30% tumor cells (Supplementary Table S1). RNA isolation, amplification, labelling, hybridization to Agilent full-genome microarrays and data processing of FFPE samples was performed as previous described [10 (link)]. RNA extraction was performed using two sections of 10-μm thickness or four sections of 5-μm thickness. Deparaffinization and total RNA extraction was performed using an RNeasy FFPE kit (Qiagen) according to the manufacturer’s instructions. RNA yield was quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) as described previously [12 (link)]. Extracted RNA was amplified using a TransPLEX C-WTA whole-transcriptome amplification kit (Rubicon Genomics, Ann Arbor, MI, USA). Amplified cDNA was labeled using the Genomic DNA Enzymatic Labeling Kit (Agilent Technologies, Santa Clara, CA, USA) and hybridized onto Agendia’s full genome arrays (custom-designed and produced by Agilent Technologies specifically for Agendia), both according to the manufacturer’s instructions. For FFPE samples, no reference channel was used. Gene expression intensities were normalized using Lowess normalization method implemented in Matlab software version R2012a (MathWorks, Inc., Natick, MA, USA).

The PLSR has been frequently used in spectral quantitative research due to its superiority of dimension reduction and the synthesis. Detailed description of PLSR is available in Wold, Sjöström & Eriksson (2001) (link). In the calculation procedure of the step, PLSR follows a linear multivariate model to associate the independent variables (X, reflectance in this research) and dependent variables (Y, soil salinity in this research) and select latent factors (variables). Thereby, it compresses X variables into a small number of latent variables (LVs) to maximize the covariance between the LV scores and Y variables. To identify the ideal number of LVs, leave-one-out cross validation (LOOCV) was conducted. Parameter optimization and modeling were implemented with the PLS_Toolbox (version 7.9) based on MATLAB^® software version R2012a (MathWorks, Inc., Natick, MA, USA).