mosGraphFlow

1. ROSMAP Dataset Processing

• parse the multiomics data from the synapse website of the dataset ROSMAP https://www.synapse.org/#!Synapse:syn23446022

• Genome Variants https://www.synapse.org/#!Synapse:syn26263118

• Methylation https://www.synapse.org/#!Synapse:syn3168763

• RNA sequence https://www.synapse.org/#!Synapse:syn3505720

• Proteomics https://www.synapse.org/#!Synapse:syn21266454

• parse the clinical data https://www.synapse.org/#!Synapse:syn3191087

Check the jupyter nodebook 'ROSMAP_union_raw_data_process_AD.ipynb' or 'ROSMAP_union_raw_data_process_gender_in_AD.ipynb' for details.

1.1 For AD vs. non-AD classification

Use 'ROSMAP_union_raw_data_process_AD.ipynb'

1.2 For female vs. male within AD samples classification

Use 'ROSMAP_union_raw_data_process_gender_in_AD.ipynb'

2. Run the Graph Neural Network Model

2.1 Load the data into NumPy format

python load_data.py --dataset 'ROSMAP'

2.2 Run experiments on ROSMAP dataset

python geo_ROSMAP_tmain_mosgraphflow.py

python geo_ROSMAP_tmain_gcn.py

python geo_ROSMAP_tmain_gat.py

python geo_ROSMAP_tmain_gin.py

python geo_ROSMAP_tmain_gformer.py

2.3 Run analysis on mosGraphFlow results

python geo_ROSMAP_tmain_mosgraphflow_analysis.py

3. Signaling network interaction analysis

The R programing language will be used here, combined with python for data processing, to visualize the result in the file 'Plot_momic.py', with the attention mechanism in model mosGraphFlow.

3.1 Calculate average pathway attention of two sample groups for selected tasks (AD vs. non-AD, female vs. male within AD samples)

python ROSMAP_analysis_path_edge.py

3.2 Data processing and visualization

python Plot_momic.py

Before you run Plot_momic.py, you should configure your own R home directory in part 6 of the script.

Following is an signaling network interaction analysis exmaple. For AD/non-AD Top 70 gene features

• Top 70 important nodes signaling network interaction in AD samples

• Top 70 important nodes signaling network interaction in non-AD samples

• Bar chart displaying the weight of important genes in AD and non-AD samples, ranking by their p-values. (The red dashed line indicates a p-value threshold of 0.05)