Synthesis of substrate materials
1. Ferric chloride was first dissolved in ethylene glycol solution.
2. Trisodium citrate (weights from 0 to 0.8 g) was then added to tune the surface charge of the products.
3. Sodium acetate was added to the mixture and sonicated at room temperature for 30 min.
4. The reaction mixture was transferred to a Teflon-lined stainless-steel autoclave (capacity 50 mL) and held at 200°C for 10 h for the formation of ferric particles.
MS data acquisition
1. Ferric particles were dispersed in water as the matrix for LDI MS analysis at a concentration of 1 mg mL-1.
2. 500 nL of matrix slurry was mixed with 50-500 nL of analyte solution on the plate and dried for LDI MS analysis.
3. Set the 5800 Proteomics Analyzer with a repetition rate of 200 Hz and an acceleration voltage of 20 kV. The delay time for this experiment was optimized to 250 ns.
4. Collected the raw MS data based on the experimental parameters above.
Preparation of clinical samples
1. All blood samples were drawn by venepuncture and clotted at room temperature within 40 minutes.
2. Serum samples were obtained by centrifuging at 5,100 xg and 4°C for 10 minutes.
3. After centrifugation, the precipitate was discarded and the supernatant serum was stored at -80℃ immediately (within 15 minutes).
4. The elapsed time was within 1 hour between blood draw, centrifugation, and ultimate storage at -80℃.
Machine learning and computer-assisted diagnosis
1. Pre-processing of the raw mass spectra data, including baseline correction, peak detection, extraction, alignment, normalization, and standardization, was carried out by MATLAB (R2016a, The MathWorks, Natick, MA) prior to pattern recognition analysis.
2. The pre-processed MS data were considered as the inputs (serum metabolic patterns) to train and test the classifier. A 5-fold cross-validation approach was performed to estimate the performance of the classifier for both the inner-loop and outer cross-validation (20 rounds for each fold, thus 100 models for outer cross-validation in total).
3. An external double-bind test for differentiating early-stage LA from healthy controls were conducted based on the as-trained classifier. The disease labels of the double-bind test cohort were unknown and predicted by the classifier.