FASTCAR - Fast Alignment-free Search Tool for Classification and Regression
Release version

Requirements: g++ 4.9.1 or later, requires Homebrew on Mac OS X

Compilation using g++ (homebrew) and GNU Make on Mac OS X
CXX=g++-7 make

see: https://stackoverflow.com/questions/29057437/compile-openmp-programs-with-gcc-compiler-on-os-x-yosemite


Linux/Unix compilation:
make

Usage: bin/fastcar *.fasta --query queryFile1.fasta [--query queryFile2.fasta] [--chunk 10000] [--id 0.90] [--kmer 3] [--chunk 10000] [--output output_first_string] [--sample 300] [--threads 4]
Or:    bin/fastcar *.fasta --all-vs-all [--chunk 10000] [--id 0.90] [--kmer 3] [--chunk 10000] [--output output_first_string] [--sample 300] [--threads 4]

All arguments except for id can be shortened to one dash followed by the first letter of the argument,
for example: --query => -q

--id controls the identity of the sequences, if classification is enabled (by default)

--query is a required parameter which takes in a query file, which searches against the database FASTA files

--all-vs-all can be used instead of supplying a query. This will compute all similarities against itself.

--mut-type will control the mutation type. See --help for details.

--chunk specifies chunk size, or, how many sequences are read in at once and processed. The larger
	the size, the more memory is used. However, this may lead to faster processing as more
	sequences can be processed at once.

--output specifies the output file prefix, to which .searchXXX is appended.

--no-format will not strip the header names after the first space, and it will delimit the output by '!'.
--mode specifies the mode[s], which can be "r" for regression and "c" for classification.
       They can be combined to form the default mode, which regresses identity scores above a cutoff.

--feat specifies which features are used. By default, "fast" (9 features) are used. However, for added
       potential accuracy, "slow" can be used, utilizing 2 more features which are slower.

--kmer decides the size of the kmers. It is by default automatically decided by average sequence length,
       but if provided, MeShClust can speed up a little by not having to find the largest sequence length.
       Increasing kmer size can increase accuracy, but increases memory consumption fourfold.


--threads sets the number of threads to be used. By default OpenMP uses the number of available cores
	  on your machine, but this parameter overwrites that.

--sample selects the total number of sample pairs of sequences used for both training and testing.
	 300 is the default value, generating 1500 positive and 3000 negative pairs.

--all-vs-all Adding the flag --all-vs-all flag will run an all versus all search on the sequences in database file(s) provided.
             The query flag is disregarded here.

--dump weight.txt   Trains the model(s) and dumps the weights
                    to the specified file, and immediately quits.

--recover weight.txt   Recovers the weights from file, and sets parameters
                       from those weights (e.g. k-mer size)

--list     Provide a list of filenames instead of passing as extra parameters
--qlist    Provide a list of filenames instead of passing as query file(s)



If the argument is not listed here, it is interpreted as an input file.


N.B. Running with --dump first and then re-running with the same parameters
     except for replacing --dump with --recover
      may consume less memory, as some data structures are saved in memory throughout.

License

Academic use: The software is provided as-is under the GNU GPLv3.
Any restrictions to use for-profit or non-academics: License needed.