This repository contains the code and resources of the following paper:
ProMeta: A meta-learning framework for robust disease diagnosis and prediction from plasma proteomics
Overview of the ProMeta framework
ProMeta is a few-shot meta-learning framework specifically designed for proteomics data analysis. Addressing the limitation of deep learning in data-scarce scenarios (rare diseases), ProMeta leverages a "learning-to-learn" paradigm to adapt to new disease tasks using as few as 4 patient samples.
- Robust Few-Shot Adaptation: Outperforms transfer learning and traditional ML baselines by ~24.6% in 4-shot scenarios (2 cases, 2 controls).
- Knowledge-Guided Encoding: Uses ConsensusPathDB (CPDB) to map proteins to biological pathways, creating robust functional tokens.
- Dual-View Representation: Handles the entire proteome by processing pathway-mapped proteins and unmapped proteins (via auxiliary tokens) in parallel streams.
- Meta-SGD Optimization: Learns both the model initialization and task-specific learning rates, enabling rapid convergence on novel diseases.
├── data_preprocess/
│ ├── preprocess_data_prevalent.ipynb # Code for prevalent disease data preproocess
│ ├── preprocess_data_incident.ipynb # Code for incident disease data preproocess
├── resource/
│ └── CPDB_pathways_genes.tab # Pathway knowledge database
└── ProMeta/
├── main.py # Entry point for training and evaluation
├── config.py # Configuration and argument parsing
├── dataset.py # Data loaders and Pathway Mask generation
├── model.py # ProMeta model architecture and Loss functions
├── utils.py # Metrics, logging, and helper functions
└── run_ProMeta.sh # Shell script to run experiments
Setup the required environment using environment.yml with Anaconda. While in the project directory run:
conda env create
Activate the environment
conda activate ProMeta
To reproduce the experiments described in the pap er (e.g., 4-shot or 32-shot adaptation), navigate to the source directory and execute the run script:
cd ProMeta
bash run_ProMeta.sh
You can also run main.py directly with custom arguments:
python main.py \
--data_dir "../data/out/" \
--proteomics_csv "../data/proteomics.csv" \
--cpdb_path "../resource/CPDB_pathways_genes.tab" \
--support_size 4 \
--batch_size 8 \
--outer_lr 1e-4 \
--inner_lr 0.005
If you find this code or our paper useful for your research, please cite:
@article{Li2026ProMeta,
title={ProMeta: A meta-learning framework for robust disease diagnosis and prediction from plasma proteomics},
author={Li, Han and Gu, Haoteng and Hu, Lei and Zhang, Zimo and Lv, Yongji and Gao, Peng and Cooper-Knock, Johnathan and Min, Yaosen and Zeng, Jianyang and Zhang, Sai},
year={2026},
}