geogav/netbid-in-covid-19-plasma-proteomics-

NetBID-in-COVID-19-plasma-proteomics

Network-based Bayesian Inference of Drivers NetBID

Create an appropriate format for the dataset

• 1.MGH_data_Preprocessing.R

Filtering and data construction for SJARACNe algorithm.

• 2.Network_Reconstruction.R

Perform SJARACNe algorithm, extracting non-linear correlations between proteins

• 3.SJARACNe.sh

Perform Bayesian Inference algorithm for feature selection

• 4.Driver_Inference.R

Load dataset into Cytoscape for better visualization of results for annotation and Enrichment analysis.

• 5.Cytoscape_load.R

Advanced analysis for plotting and interpretation

• 6.Advanced_Analysis.R

biological informed Neural Network (b-iNN)

For the biological informed neural network (b-iNN) is followed the PasNet architecture. The changes are committed to the b-iNN are described in the following steps:

1. Pathway Layer At this step it is created the REACTOM_pathway_mask.csv and GO_pathway_mask.csv documents for Reactom and GO pathway libraries, respectively. These documents are used for the two different b-iNN that are created for this study.

2. Explainable AI feature Next, the Train.py file of PasNet modified (the modified python script is in Neural network folder)

3. Receiver Operating Characteristic plot The EvalFunc.py file of PasNet modified (the modified python script is in Neural network folder)

4. Train and Test data For the training and test of the model has used the MGH dataset. The MGH dataset is normalized with mean = 0 and standard deviation = 1. The produced datasets are the following:

   • Train: std_train_0_0.csv
   • Test: std_test_0_0.csv

5. Execution of b-iNN For the execution of the experiment first it is executed the Run_EmpiricalSearch.py. After optimal L2 and LR attributes are detected, it is executed the Run.py.

scRNAseq from MGH

The .h5ad scRNA-seq data for the analysis are accessible in the following link: https://www.covid19cellatlas.org/index.patient.html (“00803_MGH_Broad00803_MGH_Broad”). The scRNA-seq analysis can then be recreated using the “scDIOR Python MGH.ipynB” script in Python to make an adata object and then the “MGHscRNAseq.R” script in R to convert the adata object into Seurat object and then perform downstream analysis.

Drug Repurposing

D-SJARACNe was assembled using the Day 7 SJARACNe network (“Cytoscape_network_D7.cys”) with data from Zhou et al. (https://doi.org/10.1038/s41587-022-01474-0). The drug-drug interaction network was assembled by using the drugs from D-SJARACNe as input in the STITCH database. Visualizations were performed in the Gephi v0.9 platform.