Quickstart

Virus Assessment using Neural networks for Just-in-time Analysis and Rapid Identification

Vanjari is a tool for the classification of viruses using neural networks. It uses the ICTV taxonomy to hierarchically classify viruses.

Installation

Vanjari can be installed using pip:

pip install git+https://github.com/bloodhound-devs/vanjari.git

Usage

The main Vanjari app can be used as follows:

For more information see:

vanjari --input virus.fa --output-csv virus-predictions.csv

Note

The first time that Vanjari is run, it will download the model weights. These are large files and may take some time to download.

This outputs a CSV file with the predictions for each virus in the input file.

This will build embeddings for all sequences in the input file and then classify them using the neural network.

To save the embeddings to disk as a Numpy memmap array, provide a location for the file and the index listing:

vanjari --input virus.fa --output-csv virus-predictions.csv --memmap-array-path virus-embeddings.npy --memmap-index virus-index.txt

If the memmap and index exist already, then they will be used without recomputing the embeddings.

The input file can also be a directory of FASTA files.

Thresholds

You can set a threshold for the predictions using the --prediction-threshold option. This must be a value between 0 and 1. If the probability of a classification at any rank in the taxonomy is below this threshold, the classification will be set to “NA”.

The threshold can be increased later using the vanjari-tools increase-threshold command:

Faster Inference

To use a Vanjari model without computing the embeddings, use the vanjari-fast command which uses a simplier convolutional neural network:

vanjari-fast --input virus.fa --output-csv virus-predictions.csv

Advanced Usage

More documentation is coming with advanced usage. For now, please see the help:

vanjari --help
vanjari-tools --help
vanjari-fast --help
vanjari-fast-tools --help

Programmatic Usage

Vanjari can also be used programmatically. For example:

from vanjari import Vanjari

vanjari = Vanjari()
results = vanjari.predict("virus.fa")