SLInterpreter-Demo

VIS 2024

This is a Demo for SLInterpreter.

Due to the large size (above 3GB) of the model prediction result files from KR4SL, and the double-blind platform's limitation of a maximum single file upload size of 8MB, the prediction results in this demo only include the cases mentioned in the paper. Other data such as the Knowledge Graph, which are relatively smaller in size, are provided in full. After the double-blind period, we will release all source code and datasets through other platforms.

Important:

Data files exceeding 8MB have been uploaded in compressed form. Therefore, before using this demo, ensure that all files listed in the unzipped list are correctly extracted to their respective locations, otherwise the functionality will not work properly.

Files to be unzipped:

• static/KG/facts_full_type.zip
• static/KG/facts_full_type_index.zip
• static/predicted_gene/predicted_gene_all.zip

User Notice:

1. Since this demo is a build version, we recommend using the Live Server plugin of VScode for running it. Directly opening the .html file may not work correctly.
2. This demo only includes the prediction result data from the Cases mentioned in the paper. Therefore, selecting other genes will not display the Interpretative Path View correctly. Please first enter Thyroid Cancer in the Disease search box and select Thyroid Cancer. Then enter CDK1 in the Primary Gene search box and select CDK1 to reproduce the Case in the paper or explore freely. Or you can enter CDK1 in the Primary Gene search box and select CDK1 directly.
3. As this demo is a trial version, the system does not have access to the full data from our test runs. Therefore, unexpectable bugs may occur during operation. You can raise issues on our subsequent open-source homepage. We greatly appreciate your understanding!
4. The Operation List and Model Log in the demo contain some built-in data for demonstration purposes. This data is not real and is only used for display.
5. Since connecting the frontend to the backend predictive model requires some configuration, we have separated the backend design and model training functionality in the demo to keep it concise and clear. You can obtain your own prediction results by training with KR4SL and connect them to the frontend system according to our file format. We will also gradually make the relevant code and used data publicly available in the future.

Dataset Links:

1. SynLethDB is a comprehensive synthetic lethality knowledge base towards discovery of selective and sensitive anticancer drug targets.

2. ProteinKG25 is a large-scale KG dataset with aligned descriptions and protein sequences respectively to GO term and proteins entities. It contains about 612,483 entities, 4,990,097 triples (including 4,879,951 protein-go triplets and 110,146 Go-Go triplets).