A pilot study on 10 patient and 10 control samples was first conducted to choose between hydrophobic interaction & weak cation exchange chromatography magnetic beads. Hydrophobic interaction chromatography C8 beads produced more peaks and achieved better cross validation and recognition capability. Patients and controls were then stratified into the training and validation sets. The training group was used to develop peptide models that could discriminate patients from controls. The validation group was then used to test the predictive power of the model derived from the training set. Sample preparation and mass analysis (peptide profiling)AH samples were thawed on ice and fractionated using magnetic beads based on Hydrophobic Interaction Chromatography (MB-HIC8, Bruker Daltonics Inc.
, Bremen, Germany) based on hydrophobic interation before MS analysis, which was conducted from January to September 2017. These magnetic beads, which exhibit good peptide-capturing performance, were used for protein & peptide enrichment of AH samples according to the manufacturer’s instructions. We added 20 ?L of the AH sample to 40 ?L of binding buffer then 5 ?L of MB-WCX beads in a polymerase chain reaction tube. After careful mixing & incubation for 1 min, tube was placed on a magnetic separator to isolate the unbound solution. The bound peptides were eluted from the magnetic beads after three rounds of bead separation and washing.
Finally, 1 ?L of the peptide eluate was mixed with 1 ?L of HCCA matrix (1.2mg of ?-Cyano-4-hydroxycinnamic acid in 50% acetonitrile with 1% trifluoroacetic acid), which was then spotted onto the sample anchor spots of a polished steel target plate (Bruker Daltonics Inc., Bremen, Germany). The MALDI-TOF MS analyses were performed on an Ultraflex III MALDI-TOF MS device (Bruker Daltonics Inc.) with the following settings: linear positive ion mode, repetition rate of 200 Hz, ion source voltages of 25 kV and 23.65 kV, lens voltage of 6.8 kV, pulsed ion extraction time of 300 ns. All signals with a signal-to-noise ratio of 3 in a mass range of 1,000–20,000 Da were recorded using FlexAnalysis software (version 3.
4; Bruker Daltonics Inc.). The peptidomic patterns and models were processed using ClinPro Tools bioinformatics software (version 3.0; Bruker Daltonics Inc.
)Data processing and statistical analysisThe following workflow was used for the data processing and analysis with ClinPro Tools software: spectra were normalized, baseline subtracted, peaks smoothed and peak areas were calculated for each spectrum. All peak signals were processed for noise reduction using a top-hat baseline in the 800–20,000 Da range. For the peptide peaks, the expressions of the same mass-to-charge ratios (m/z) were compared between the patients & controls. For model construction, spectra of the training set were used. Three algorithms (genetic algorithm GA, supervised neural networks SNN, and quick classifier QC) were used to establish the prediction models. Next, each model was applied to the validation set to test its ability to identify patients and controls.