Multimodal Knowledge Graph for Link Prediction and Node Classification in ADNI and PPMI datasets

About

View Final_Report to get idea of this project.

Code adapted from: https://github.com/ronghanghu/tensorflow_compact_bilinear_pooling Paper reference: End-to-End Entity Classification on Multimodal Knowledge Graphs: https://arxiv.org/abs/2003.12383

Codes and their explanations

1. adni_dataprep.py Preparation of ADNI dataset => we also prepared the dataset further if you only want to use adni dataset
2. pd_dataprep.py Preparation of PPMI dataset => we also prepared the dataset further if you only want to use ppmi dataset
3. combined_dataprep.py: Preparation of ADNI and PPMI dataset
4. config.yaml: config file for adni_pd_link_NC.ipynb
5. adni_pd_link_NC.ipynb: Jupyter notebook to run dataset through link prediction and node classification
6. utils folder, consisting of

• dataprep_utils.py: functions needed for preparing dataset for adni_dataprep.py and pd_dataprep.py
• load.py: loading and preparing dataset for tasks of node classification and link prediction
• data_utils.py: classes to prepare images for 3D SqueezeNet

7. model_utils.py: functions used in model development
8. SFCNnet.py: classes defining the SCFN network
9. squeezenet.py: classes defining the SqueezeNet network
10. rgcn.py: classes defining rgcn network.
11. Combined.ipynb: Jupyter notebook to prepare ADNI and PPMI datasets for ablation studies

Other References

I used several pretrained models in this project

1. SFCN: https://github.com/ha-ha-ha-han/UKBiobank_deep_pretrain
2. 3D-Squeezenet : https://github.com/okankop/Efficient-3DCNNs
3. Bio-Clinical BERT : https://huggingface.co/emilyalsentzer/Bio_ClinicalBERT

Data Source

I got the data from ADNI (https://adni.loni.usc.edu/) and PPMI (https://www.ppmi-info.org/). If you like, please request the data from there directly.

How to run the codes

1. Run adni_dataprep.py
2. Run pd_dataprep.py
3. Run combined_dataprep.py (need both step 1 and 2 to be run first)
4. Change config.yaml to reflect updated parameters like model path, embedding paths etc
5. Run adni_pd_link_NC.ipynb