Machine Learning Classifier for CRESS-DNA viral capsids
Built by Ryan O. Schenck Contact: ryanschenck at mail dot usf dot edu
Note: This is not a peer-reviewed published tool. Use at your own risk.
This script is designed for high throughput classification of circular rep-encoding eukaryotic single-stranded DNA (CRESS) viruses.
The dataset was built from this publication: DOI 10.3389/fmicb.2015.00696
This package utilizes an ensemble (1000) of Decision Trees (non-parametric supervised machine learning) for classification of capsid types of CRESS viruses. This learning method will provide a user with a file that provides ‘Sequence Name, Type Predicted, an array showing the probabilities for that type classification.’
Based on the training and testing sets (80/20 split) taken from Rosario et. al. 2015 the accuracy of this machine learning method is between 94.23% and 96.15%.
VL3 raw scores for the first 125 amino acids for each sequence (sequences <125 amino acids will not be possible) in a comma-separated spreadsheet (*.csv or *.txt):
| Identifier | Score.1 | Score.2 | Score.3 | Score.4 | Score.n |
|---|---|---|---|---|---|
| Sequence_Name1 | Score1 | Score2 | Score3 | Score4 | Scoren |
| Sequence_Name2 | Score1 | Score2 | Score3 | Score4 | Scoren |
$python CapClassifier.py input_file.csv
In the extra folder is a webdriver for obtaining VL3 scores when given a fasta file of multiple sequences. There are several dependencies for this script and thus will not be covered in this READ_ME file, but the script is there for those who need it. The Disprot website has strict Rate Limits when using this script and thus only a number of samples can be ran at one time. Paths will need to be changed in this script to suite your environment.