DeepSUMO is a computational tool that predicts protein SUMOylation sites and SUMO-interaction motifs (SIMs). By applying the deep learning algorithm together with effective feature extraction methods, we trained prediction models based on the published modification data for sumoylation and SIMs prediction. This new method can discover high-level features and achieves robust prediction performance. For convenience, the online service has also been developed and is freely available.

1. Firstly, to use DeepSUMO you will need to upload a protein data file or paste a formatted test in FASTA format.
An example for FASTA format:

 >sp|P10242|MYB_HUMAN Transcriptional activator Myb OS=Homo sapiens GN=MYB PE=1 SV=2
MARRPRHSIYSSDEDDEDFEMCDHDYDGLLPKSGKRHLGKTRWTREEDEKLKKLVEQNGTDDWKVIANYLPNRT
DVQCQHRWQKVLNPELIKGPWTKEEDQRVIELVQKYGPKRWSVIAKHLKGRIGKQCRERWHNHLNPEVKKTSWT
EEEDRIIYQAHKRLGNRWAEIAKLLPGRTDNAIKNHWNSTMRRKVEQEGYLQESSKASQPAVATSFQKNSHLMG
FAQAPPTAQLPATGQPTVNNDYSYYHISEAQNVSSHVPYPVALHVNIVNVPQPAAAAIQRHYNDEDPEKEKRIK
ELELLLMSTENELKGQQVLPTQNHTCSYPGWHSTTIADHTRPHGDSAPVSCLGEHHSTPSLPADPGSLPEESAS
PARCMIVHQGTILDNVKNLLEFAETLQFIDSFLNTSSNHENSDLEMPSLTSTPLIGHKLTVTTPFHRDQTVKTQ
KENTVFRTPAIKRSILESSPRTPTPFKHALAAQEIKYGPLKMLPQTPSHLVEDLQDVIKQESDESGIVAEFQEN
GPPLLKKIKQEVESPTDKSGNFFCSHHWEGDSLNTQLFTQTSPVADAPNILTSSVLMAPASEDEDNVLKAFTVP
KNRSLASPLQPCSSTWEPASCGKMEEQMTSSSQARKYVNAFSARTLVM

2. Next, select the PTM types and thresholds. The PTM Type panel lists 2 options, Sumoylation and SIM. For a better prediction performance, three thresholds of high, medium and low stringency were chosen for each PTM type. In order to balance the prediction performance, the medium stringency was selected as the default threshold. In addition, we offer several examples to try out the web server.

3. Finally, select a prediction threshold for each PTM type to start the calculation. The current status can be view in the Result panel.

4. All of the prediction results are presented in a tabular form containing the information of FASTA title ID, modified position, modified peptide, predicted score, prediction cutoff and modified type. In the position column, the precise modification sites are shown. Also, the predicted peptide for modification is displayed in the peptide column with the modified site shown in red. The cluster score and its corresponding cutoff are presented in the score and cutoff column, respectively. In the last column, the explicit PTM type is indicated.

The prediction models of DeepSUMO were trained using Deep Learning algorithm. After training the prediction models, we carried out 10-fold cross-validation to validate the prediction performance. To balance the prediction accuracy, we selected three thresholds with high, medium and low stringencies based on the evaluation results. The detailed performance under these three thresholds was presented as follow:

DeepSUMO Database collected the published modification data of Sumoylation and SIMs. At present, DeepSUMO database contains 3492 Sumoylated proteins and 221 SIM-associated proteins. Besides the modification data, users also can query or browse contains other protein details from DeepNitro database.

Search
You can input one or multiple keywords (separated by space character) to search the DeepSUMO database. The search fields including DeepSUMODB ID, UniProt Accession, Protein Name, Protein Alias, Gene Name, Gene Alias, Species, Function, Modification Type and Related Disease.

Browse
You can browse the DeepSUMO database by species. With the default setting, you can directly click on the "Submit" button to browse all the proteins that are modified by SUMO in the database.

BLAST search
It could be used to find the specific protein and/or related homologues by sequence alignment. This search-option will help you to find the querying protein accurately and fast. Only one protein sequence in FASTA format is allowed per time. The E-value threshold could be user-defined, while the default is 0.01.

If you are having trouble with DeepSUMO please contact the two major authors: Prof. Jian Ren and Prof. Zhixiang Zuo. We will try to resolve it.