A C and Python implementation of Brian Cleary's LSA, for eigengenome partitioning.

Run these commands.

git clone https://github.com/mooreryan/eigengenomes.git
cd eigengenomes
make

This will make the executable files needed for the pipeline.

In the source directory, run make clean. Also, if you moved any of the program binaries, you will need to manually remove them.

Run hash_and_count for each fastq file. It assumes reads are paired.

Usage: ./hash_and_count <1: seed> <2: kmer size> <3: number of hyperplanes> <4: seqs.fastq> > seqs.hash_counts

Run weight_counts once on all the output files from the hash_and_count program.

Usage: ./weight_counts <1: num hyperplanes used in previous step> *.hash_bucket_counts > cool_sample_group.hash_counts

There aren't wrapper scripts yet, but here is an example of running the steps implemented thus far.

test_files contains s1.fastq and s2.fastq.

This command will hash the kmers and count them. It will create these files: test_files/s{1,2}.hash_counts.

parallel "./hash_and_count 0 3 4 {} > {.}.hash_counts" ::: test_files/*fastq

This command will weight the counts using something similar to tf-idf weighting. It will create the file test_files/all.hash_counts.

./weight_counts 4 test_files/*.hash_counts > test_files/all.hash_counts

Aaaand, the remaining steps aren't finished yet ;)

• 0: Success
• 1: Argument error
• 2: Couldn't open a file
• 3: Not an even number of forward and reverse reads
• 4: A sequnece is shorter than the kmer length
• 5: A hashed kmer bucket was missing from a counting hash

• The counting arrays are way to memory intensive, consider hash tables for counting.
• Nucleotides are being mapped to 5 different integers instead of complex numbers.

• Hyperplanes are drawn at random through the origin rather than based on existing kmers.
• The idf score is calculated a bit differently.
• It is not designed to be run across many nodes of a compute cluster. This may change.