DeepLocPro 1.0 is a multiclass subcellular localization prediction tool for prokaryotic proteins, trained on experimentally verified data curated from Uniprot and PSORTdb. DeepLocPro has been trained to work with prokaryotic proteins from a wide range of organisms covering Archaea, Gram-positive bacteria, and Gram-negative bacteria. It can differentiate between six different localizations: Cell wall & surface, Extracellular, Cytoplasmic, Cytoplasmic membrane, Outer membrane and Periplasmic.
The DeepLocPro 1.0 server requires protein sequence(s) in fasta format, and can not handle nucleic acid sequences.
The paper can be accessed here: https://academic.oup.com/bioinformatics/article/40/12/btae677/7900293
More information about the method can be found at:
https://services.healthtech.dtu.dk/services/DeepLocPro-1.0/
DeepLocPro 1.0 will run and has been tested under Linux and OS X. The only prerequisite is to have Python 3.6 or above installed.
The installation procedure is:
- Install the DeepLocPro 1.0 package:
# Within the deeplocpro directory
pip install .
- Test DeepLocPro 1.0 by running:
deeplocpro -f test.fasta
This will create a directory outputs containing the predictions.
DeepLoc will be installed under the name 'deeplocpro'. It has 4 possible arguments:
-f,--fasta. Input protein sequences in the fasta format.-o,--output. Output folder name.-p,--plot. Plot and save attention values for each individual protein.-d,--device. One of cpu, cuda or mps. Default: cpu.-g,--group. Prevent outer membrane & periplasm prediction when Archaea/positive. One of any, archaea, positive or negative. Default: any
The output is a comma separated file with the following format:
- 1st column: Protein ID.
- 2nd column: Predicted localization.
- 3rd-8column: Probability for each of the individual localizations.
If --plot is defined, a plot and a text file with the feature importance of the position for the prediction will be generated for each query protein.
In case of technical problems (bugs etc.) please contact health-master@dtu.dk.
Questions on the scientific aspects of the DeepLocPro 1.0 method should go to Henrik Nielsen, henni@dtu.dk.
