This repository contains the implementation of an artificial neural net in PyBrain for classifying metagenomic taxonomic classifications

Contains all scripts, bins and databases relevant for downloading and processing traingsets from MG-RAST. This implementation uses the Oracle Grid Engine.

Download and processing is wrapped by start.sh (training sets) and start_target.sh (test sets). Processing of training sets needs the reference file 'table.tsv' which references all relevant training sets with an inhouse ID (first column), which is passed to $SGE_TASK_ID for parallel computation. (qsub start.sh -t 1-X)

Processing of test sets is wrapped by start_tartget.sh. This script uses the reference file 'target_sampels/metadata.csv' which references all test sets in a similar manner. Raw sequence data for selected test sets is not included but can be downloaded via ENA accessions. These need to be in fasta format to be processed.

.results/ contains all processed training sets

.target_samples/results contains all processed test sets

Contains all result files relevant for cross validation and classification of selected test sets.

.crossvalidation/ contains bash files to compute, accuracy measurements and ploting scripts to produce surface plots

.30hidden200epochs/ contains heatmap, output and network.obj, a gzipped pickle object of the saved network (needs to be unpacked to use)

Contains the original table from MG-RAST, a script to cure taxonomically classified samples and the resulting processed table which contain all relevant samples.

Processed test sets and training sets respectively.

Crossvalidation can be run with crossvalidation.py Classification of novel samples can be run with classify.py