Under active development...
THOR is a tool that will generate PNG images from genomic data, for use in machine learning applications.
So far, it will:
- histosketch a bunch of FASTA files, creating a set of HULK histosketches
- colour these histosketches to RGB values, so that each sketch of length x will be encoded into x RGB values
- build a PNG image from an OTU table where each row of pixels corresponds to a coloured histosketch
It's a work in progress, but we've had some success in using these images in Neural Nets to classify the Human Microbiome Project 16S samples by body site.
When an OTU is not present in the refseq collection, it is ignored by THOR. This needs to be handled properly.