（For subsequent updates, suggestions, or inquiries, please contact the author via email: oceanddl@sina.com）
This EC (endometrial cancer) repository includes data preprocessing, model construction/validation, and 2M-EC platform deployment. The repository also incorporates our in-house developed tools for molecular matching and ROMA data collection.
Stores trained models and data processors. Deployment programs. Includes 4 test samples for platform validation.
Used to optimize analyte-algorithm combinations. The model training methodology referenced literature（Osipov, A., et al. The Molecular Twin artificial-intelligence platform integrates multi-omic data to predict outcomes for pancreatic adenocarcinoma patients. Nat Cancer 5, 299-314 (2024).
Training, saving and validation of mass spectrometry models and fusion models.
New sample data requires spectral preprocessing followed by binning.
Protein molecular weight calculation and matching.
Web-based clinical cohort input and result collection for ROMA.
Prior to new sample prediction, spectrum data must undergo preprocessing and spectral binning, with the binning methodology following published literature protocols（Weis, C., et al. Direct antimicrobial resistance prediction from clinical MALDI-TOF mass spectra using machine learning. Nat Med 28, 164-174 (2022).）
This repository provides all datasets supporting the project, consisting of four essential files : Cervical2.zip, Uterine2.zip, Plasma-PM2.zip, Plasma-PP2.zip. Each archive contains both training datasets for model development and independent validation datasets for performance evaluation, enabling complete pipeline assessment from training to validation. All files are accessible via: Baidu Netdisk (https://pan.baidu.com/s/1Bxc0J5z0vab9ggsvZ2s_RQ?pwd=sf1b, extraction code: sf1b).
Database of 1160 multi-omics raw profiles from uterine secretions, cervical secretions and plasma with MALDI-TOF Mass spectrometry, processed dataset and clinical metadata
For each site, the data consists of MALDI-TOF mass spectra in the form of .txt files and a aggregated meta-data file (Metadata_CI.csv) with clinical information to model and align.
The EC folder structure obtained after download is the following:
EC
├── Cervical
│ ├── raw_192CM
│ ├── M1/2
│ ├── TS1
│ └── vali_C
│
│
├── Uterine
│ ├── raw_246UM
│ ├── M2
│ ├── TS2
│ ├── NSMP
│ ├── p53
│ └── vali_U
│
├── Plasma
│ ├── raw_361PM
│ ├── raw_361PP
│ ├── M1/2 (PM)
│ ├── M1/2 (PP)
│ ├── TS1
│ ├── vali_PM
│ └── vali_PP
│
├── Metadata_CI.csv
│
├──README.md
University of Fudan, China
For details on the dataset extraction and preprocessing, please refer to the Methods section in the article corresponding to the publication https://www.nature.com/articles/s41591-021-01619-9.
Raw MALDI-TOF MS spectra were converted to .txt format using flexAnalysis software (Bruker, Germany). Spectral preprocessing was performed with R packages MALDIquant and MALDIquantForeign, implementing: 1) square root transformation for variance stabilization, 2) Savitzky-Golay filtering (15-point window) for smoothing, and 3) SNIP algorithm (20 iterations) for baseline correction. Following established protocols, we conducted mass-to-charge ratio (m/z) binning with sample-specific resolutions: 9,000 evenly distributed bins for uterine/cervical metabolic profiles (100-1,000 Da), 900 evenly distributed bins for plasma metabolic profiles (100-1,000 Da) and 675 evenly distributed bins for plasma peptidomic profiles (3-30 kDa). The binning approach was adjusted based on the mass spectrometer's resolution. All processed data retained intensity values normalized to total intensity value of the spectrum.
We recommend using Python package for MALDI-TOF MS preprocessing and machine learning analysis, maldi-learn (https://github.com/BorgwardtLab/maldi-learn), to load and analyse EC data.
The github package comes with an elaborate README.md file, which gives details on installation and usage examples.
The code tools for data processing can be found at https://github.com/lmsac/2M-EC.git.
We implemented a stratified cohort design comprising a modeling cohort and an external test cohort. Modeling cohort (n=436) was partitioned into three M sub-cohorts, M1 (cervical and plasma, n=314), M2 (cervical, plasma and uterine, n=436), and M3 (uterine annotated with subtyping, n=77). The raw spectra processed dataset is uploaded in a tabular form as a stratified cohort. The raw data for p53 and nsmp were obtained from Uterine and the processed dataset was uploaded. The raw data files can be aligned according to Samples.
The first column, "Samples," represents the clinical identification numbers of the patients, and "target" indicates the patient grouping, where 1 denotes the endometrial cancer group and 0 denotes the non-endometrial cancer group. Additionally, for the "p53" and "nsmp" groupings, 1 represents the specific molecular subtype group of the cancer, while 0 represents the other three coded molecular subtype groups combined.