NMF_neurodevelopment

analysis pipeline for non-negative matrix factorization on the Philadelphia Neurodevelopmental Cohort

---

1. install libraries

• install nimfa, scipy, and numpy for python: pip install nimfa

2. place PNC data in the pnc_data directory, data available through NIH dbGaP

• n780_censor.mat
• n780_despike.mat
• ts_36_ev.mat
• age.mat
• cognDataCollection.mat
• coordinatesXYZ.mat
• mvmt.mat
• neuralSystem.xlsx
• sex.mat

3. generate old and young subject groups in subject_indicies directory

• we used the 100 oldest and 100 youngest subjects for our analyses

4. build association matrices using wavelet coherence

• wavelet toolbox used for generated wavelet coherence: http://www.glaciology.net/wavelet-coherence
• this toolbox is included in the matlab_functions directory
• navigate to ts36_wavelets_highpass directory
• run wavelet_coherence.m file in matlab
• it may be useful to split this process into several batch processes
• repeat these stets in the ts36_wavelets_censor directory

5. parameter grid search

• in the parameter_sweep directory run python parameter_sweep.py
• this saves the result in a .mat file in that directory
• the grid search will take a few days to complete running

6. pre-nmf processing and running nmf

• in the ts36_wavelets_highpass directory run python highpass_nmf.py
• this will concatenate the subject matrices and regress out motion, then run nmf
• you can also run python highpass_nmf_supplement.py in the same directory for additional processing pipelines
• in the ts36_wavelets_censor directory run python censor_nmf.py

7. in nmf_master.py running each pipeline will generate figures

8. code to generate figures are located in the figures directory. BrainNetViewer was used to visualize the nodes