Matlab r2021a

The in-house library was constructed by collecting ginsenosides and other steroid compounds in the existing literature and database (i.e., FoodB). In total, 468 compounds were included in our in-house library, and the complete list of these compounds is provided in the supplementary materials (Table S2). For each record, the chemical name, formula, CAS number (if available), FoodB ID (if available), HMDB ID (if available), source of information, accurate mass, and the m/z of possible adduct ions and multiply charged ions in both positive and negative modes, are provided.
The m/z in the ion feature list from the XCMS analysis was compared with the m/z in the in-house database to screen the possible ginsenosides, and if the ∆m/z is within the ±5 mDa, the ion feature is flagged and considered as a potential ginsenoside. This process was conducted using MATLAB R2021a (MathWorks Inc., Natick, MA, USA). The extracted ginsenoside ion features were exported into an Excel spreadsheet and verified manually. The remaining ion features from the XCMS analysis were used as non-ginsenoside features. The chemometric analysis was conducted using MATLAB R2021a (MathWorks Inc., Natick, MA, USA) with the PLS toolbox (Eigenvector Research, Inc., Manson, WA, USA).

Offline, EEG data were imported to EEGLAB toolbox 2021.0 (Delorme and Makeig 2004 (link)) running under Matlab R2021a (The MathWorks, Natick, MA), with the additional ERPLAB 8.20 plug-in (Delorme and Makeig 2004 (link)). EEG signal was digitally filtered using low-pass and high-pass filters set to 30 and 0.1 Hz, respectively. To reduce artifacts from the continuous EEG data (i.e. eye movements, eye blink, muscle contractions and movement) EEGLAB plugin clean_rawdata() including artifact subspace reconstruction (ASR) was applied to the data (Gabard-Durnam et al. 2018 (link); Blum et al. 2019 (link); Chang et al. 2019 (link)). The parameters used were flat line removal, 5 s; electrode correlation, 0.8; ASR, 20; window rejection, 0.5. As a result, the mean channel rejection rate was 6.5% (SD 2.6, range 2.5–13.7), while the mean data rejection rate was 2% (SD 2.9, range 0–14). The rejected channels were interpolated using EEGLAB's spline interpolation function and the signal was then re-referenced to the average electrode. Continuous data were segmented into 600-ms epochs from 100 ms pre-stimulus to 500 ms post stimulus onset and, successively, baseline corrected from −100 ms to 0 ms before the onset of the stimulus. Datasets with less than 11 epochs per condition were excluded from the analyses.

Data were analyzed in a custom-made MatLab program (MatLab R2021a (9.10.0.1602886), Mathworks, Nattick, MA, USA). A Hampel filter was used to detect outliers in the NIRS signal (e.g., from unwanted body movements). These outliers were replaced with the median value if exceeding three standard deviations from the value itself and three neighboring points to each side [30 (link)]. Signals were then filtered using a 5th order Butterworth low-pass filter with a 1 Hz cut-off frequency. Zero-lag filtering was used to prevent phase shifting. The end of the cyclic pattern in the signal was considered as the start-point of the recovery phase [5 (link),18 (link),19 (link)].