Semantic Link Association Prediction for Drug Target Prediction

GO TO SLAP Web Service


The rapidly increasing amount of public data in chemistry and biology allows investigation of complex mechanisms of action of drugs by the systematic integration and analysis of the heterogeneous data into a semantic linked network. Some hidden relations (e.g., drug target relation) in the rich and well organized network could be inferred by statistical models. Our work particularly investigated drug target relation using the data annotated and integrated in our previous work (Chem2Bio2RDF, Chem2Bio2OWL). In this system (SLAP), you are able to do 1) one drug target pair association prediction and network exploration 2) drug target prediction against hundreds of proteins 3) target ligands identification and prediction 4) batch drug target pairs prediction 5) similar drugs identification based on biological functions 6) similar target identification based on ligand information 7) drug target prediction using other predictive models. This system is developed by Bin Chen under the supervision of David Wild and Ying Ding. Qian Zhu joined the development of web site. You are most welcome to send us your feedback


Drug Target Association:

1) For compound/drug input, you can use name, smiles, or PubChem ID; for target input, you can use gene symbol, protein name, or Uniprot ID. You can also input sequence to retrieve its most similar target.
2) P value is used to measure their association. Smaller p value, the stronger association. Usually p value <0.05 is good.
3) If the association already exists in our database, you can further explore it. The link in the SLAP result section will direct to the original RDF set.

Network Exploration:

1) The edge can be filtered by its significance. Smaller score, more important the edge is to contribute the association.
2) We offer several network layout and export formats.
3) The nodes and edges are colored by their semantics.
4) Click the node, you can get its basic information; further click more information link, you can be directed to the original RDF set.
5) Click the edge, and further click the evidence link, you can be directed to their original RDF set which curates their relation.
6) If you like the result for whatever reasons, just click the like button. If you think the result does not make any sense, feel free to click dislike button. This allows us to look at the data and improve our algorithms.
7) The number in the brackets (if there has in the node) shows the number of same nodes with same semantics (e.g., nodes are from the same class, nodes are sharing same neighbors) which are clustered into one node. E.g, 5199 (2) means there are other two compounds sharing same neighbors with compound 5199. They are listed in the node information panel.

Drug Target Prediction:

1) Input compound alone will direct to drug target prediction.
2) SLAP run multiple cores to predict the compound against over 500 proteins.
3) It usually takes ~2min to finish the job. Be patient!
4) The similarity between drugs is assessed by their biological fingerprints composed by their association scores against hundreds of proteins.
5) Drug indications are extracted from paper (drug target network, nature, 2007)
6) example: Ibuprofen, used for Osteoarthritis, has similar drugs Apomorphine, used for Parkinson disease.
7) You can download the biological fingerprints of the input drug as well as the similar drugs.

Target Ligands Prediction:

1) Target ligands are fetched from our chemogenomics hub which integrated over 12 public chemogenomics datasets.
2) Similarity between targets is assessed by their ligand structure similarity using Similarity Ensemble Approach.
3) The compounds of interest to the given target are selected to build the library for virtual screening.
4) Virtual screening usually takes ~2min.

Batch Drug Target Pairs Prediction:

1) You have to follow the format to make your pairs.
2) Only 500 pairs are allowed.
3) For 500 pairs, it takes ~2 min.

Other Predictive Models:

We offer other predictive models, such as similarity ensemble approach, naive bayesian model, occurrence in the literature (will list the articles in which the abstracts include both compound and protein). We have not evaluated these models yet.