This repository provides a lightweight and practical solution for estimating Coupled Cross-sectional and Longitudinal Non-negative Matrix Factorization (CCL-NMF) loading coefficients in new datasets, without the need to re-run the full CCL-NMF pipeline. We provide pre-trained regression models that approximate the subject-level cross-sectional CCL-NMF loading coefficients using regional brain volumes derived from T1-weighted MRI and basic demographic information (age, sex, intracranial volume) to make CCL-NMF loadings more accessible for downstream research. The regional volumes used in these models are not harmonized. This design choice eliminates the need for users to perform data harmonization, which is often impractical in small or single-site datasets.
It is strongly recommended to install the application in a new virtual environment, such as via Miniconda or venv. Please see the respective pages to learn how to set up a virtual environment for python 3.10 on your system and activate it.
When your desired environment is active:
git clone https://github.com/CBICA/CCL_NMF_Prediction.git
cd CCL_NMF_Prediction
pip install -e .
CCL_NMF_Prediction requires 3 arguments:
- -i : The input DLMUSE ROI CSV. This is a comma-separated tabular data containing volumes for ROIs with columns matching the DLMUSE ROI indices. This should be the output of the NiChart_DLMUSE tool.
- -d : The demographics CSV containing at least MRID (string, the subject identifier), Age (integer), Sex ("M"/"F").
- -o : The path to the output CSV file containing CCL-NMF component predictions. Must be writable.
An example command is:
ccl_nmf_prediction -i /path/to/input.csv -d /path/to/demographics.csv -o /path/to/output.csv
Please see test for testing and replication