HPODNets

HPODNets: deep graph convolutional networks for predicting human protein-phenotype associations

Please run the programs in order.

Dependencies

Our model is implemented by Python 3.6 with Pytorch 1.4.0 and Pytorch-geometric 1.5.0, and run on Nvidia GPU with CUDA 10.0.

Preprocessing

extract_gene_id.py: First, please download gene annotations file from http://compbio.charite.de/jenkins/job/hpo.annotations.monthly/ with all sources and all frequencies: ALL_SOURCES_ALL_FREQUENCIES_phenotype_to_genes.txt. Then run the script, you will get a .txt file containing all gene ids. Finally, please upload this file to http://www.uniprot.org/mapping/ to map Entrez Gene ID to UniProt ID.
create_annotation.py: After generating Gene ID mapping file, you can run this script to generate HPO term-protein association file. The output json file contains associated proteins of each HPO term, like
```
 {
   hpo_term1: [ protein_id1, protein_id2, ... ],
   hpo_term2: [ protein_id1, protein_id2, ... ],
   ...
 }
```

Cross-validation

split_dataset_cv.py: First, the script will remove the proteins that are added after a certain previous time. Then, only the HPO terms in PA sub-ontology (HP:0000118) will be left. Finally, the script split the protein set into five folds and generate the dataset pickle file like
```
 store = {
 	"annotation": full dataset
 	"mask": [
 		{
 			"train": training mask of 1st fold
 			"test": test mask of 1st fold
 		},
 		...
 	]
 }
```

Temporal validation

split_dataset_temporal.py: We still focus on HPO term in PA sub-ontology. We use the proteins added before a certain previous time as the training set, and the proteins added after that time as the test set. The generated dataset pickle file is organized as
```
 store = {
     "annotation": full dataset
     "mask": {
         "train": training mask
         "test": test mask
     }
 }
```

Feature Extraction

STRING

string.py: Please firstly open https://string-db.org/cgi/download.pl and choose "organism" as "Homo sapiens", then download "9606.protein.links.v11.0.txt.gz" (version number may change). Meanwhile, download mapping file under "ACCESSORY DATA" category, or open website https://string-db.org/mapping_files/uniprot_mappings/ to download it. After downloading, you can run this code to get a json file containing PPI data organized as
```
 {
 	protein1: {
 		protein1a: score1a, 
 		protein1b: score1b, 
 		...
 	}, 
 	...
 }
```

Here the scores are scaled to [0, 1].

GeneMANIA-Net

genemania.py: First, please download COMBINED.DEFAULT_NETWORKS.BP_COMBINING.txt from http://genemania.org/data/current/Homo_sapiens.COMBINED/, and then download identifier_mappings.txt from http://genemania.org/data/current/Homo_sapiens/. Run the genemania.py and you will obtain the PPI network as a json file.

HumanNet

humannet.py: First, open https://www.inetbio.org/humannet/download.php, and then click HumanNet-XN in the Integrated Networks section. Next, open https://www.uniprot.org/uniprot/?query=*&fil=reviewed%3Ayes+AND+organism%3A%22Homo+sapiens+%28Human%29+%5B9606%5D%22. Click the Columns button in the middle of the page, and click the crosses in the upper right corner of all the dotted boxes in the Columns to be displayed of the new page. Then enter GeneID in the Search: search box in the middle of the Add more columns column, and click the associated word that pops up. At this point, click the Save button on the far right. Jump back to the original page, now only Entry and Cross-reference (GeneID) are left in the form. Click the Download button in the middle of the page, select Format: Tab-separated, and then click the Go button to download the file. Rename the file to entrez2uniprot.txt and place it under data/feature/HumanNet/raw. After the download is complete, in the UniProt page just now, click the Columns button again, in the new page, click Reset to default on the right side (note: do not click the word default, but click Reset), and then click Save. Now, the UniProt interface is restored to its original appearance. Finally, run humannet.py and then get the processed PPI network in json format.

Run the model

main.py: Run the main.py script, and you will obtain the prediction results. Be careful to the cuda device ID!

Evaluation

evaluation_macro.py & evaluation_micro.py: There are two modes of evaluation: 1) macro-averaged; 2) micro-averaged. You can run the corresponding script to get the performance.

Download the prediction results

We upload the prediction results of 15,033 proteins stored in STRING, GeneMANIA-Net and HumanNet databases that have not been annotated using the HPO annotations of 4,424 proteins released in October 2020 as the training set. The data is available at:

https://doi.org/10.6084/m9.figshare.14222732

The file is organized in .json format, where the key is the HPO term, and the values are proteins and their corresponding predictive scores. The file is so large (1.98 GB), and you are free to download it.

liulizhi1996 / HPODNets