Repository navigation
DeepGOPlus
Members:
We developed a novel method for predicting protein functions called DeepGOPlus. DeepGOPlus achieved high performance compared to top performing CAFA3 methods. DeepGOPlus uses only protein sequence in order to predict protein functions. Thus, it can annotate millions of proteins in UniProtKB. Here, we worked on building a searchable, browsable and downloadable dataset of predicted functions for more than 160 million proteins in UniProtKB.
Currently, DeepGOPlus has an online website which can run predictions for submitted protein sequences. However, the website does not provide predictions for download and running predictions take some time. We plan to solve this problem by storing predictions for almost all of available protein sequences.
Firstly, we downloaded UniProtKB/SwissProt and UniProtKB/Trembl protein sequences. Secondly, we use DeepGOPlus and obtain predictions with their scores for every protein sequence. Since UniProt/Trembl contains 167 million proteins, the file with predictions got very large and we decided to work with 550 thousand proteins in UniProt/SwissProt while we are developing the web interface for it. Then, we created relational database tables for storing and indexing the data for efficient search queries and loaded our predictions to the database. Finally, we extended DeepGOPlus website at http://deepgoplus.bio2vec.net and added a page for browsing and downloading the predictions in GPAD format.
In conclusion, we created a dataset of accurate predicted functions and now users can browse and download our data without the need to run the predictor themselves.
In future, we plan to add many other formats (JSON, RDF/XML, GAF) for our download service. In addition, we plan to provide SPARQL Endpoint for our dataset.