EC is a free, open-source command-line tool for analysis of GWAS (SNP) and other types of biological data. Several modes are available for various types of analysis, including:
Evaporative Cooling (EC) is C++ library that provides a flexible feature selection algorithm for SNPs and quantitative data, using ReliefF and Random Jungle for interactions and main effects, respectively. EC is also available as a standalone tool.
EC is being developed by the In Silico Research Group at the Tandy School of Computer Science of the University of Tulsa. Our research is sponsored by the NIH and William K. Warren foundation. For more details, visit our research website.
-
EC library, available as a source release on the EC project page, and its dependencies:
-
gfortran, sometimes installed alongside compiler tools
-
GNU Scientific library (libgsl)
-
libxml2
-
Boost system, filesystem, and program-options libraries
-
The libz/zlib compression library is required, but this is installed by default on most Unix systems. In MinGW libz is installed via mingw-get.
-
OpenMP is required to take advantage of the parallelized tree growth in Random Jungle and distance matrix calculations for ReliefF. This is another library typically installed alongside the compiler toolchain.
To compile this code, a GNU toolchain and suitable environment are required. GNU g++ has been used to successfully compile the code.
We have successfully built and run EC on:
- Linux (64-bit Ubuntu) (gcc-4.6)
- Mac (10.6 - 10.7) (gcc-4.2.1)
- Windows 7 (32-bit) using the MinGW compiler system (gcc-4.6)
To build EC, first run the bootstrap script
./bootstrap.sh
Ignore any extraneous warnings. This calls autoreconf and generates the configure script. From this point, a standard
./configure && make && sudo make install
will generate the Makefile, compile and link the code, and copy the objects to
the installation directory (default of /usr/local). As is convention, headers
are installed in $PREFIX/include, binary in $PREFIX/bin, and the library in
$PREFIX/lib.
The resulting binary src/ec_static.exe will run as a command-line tool.
Allowed options:
--help produce help message
--verbose verbose output
--convert convert data set to data set - no ec
-T [ --optimize-temp ] optimize coupling constant T
-c [ --config-file ] arg read configuration options from file -
command line overrides these
-s [ --snp-data ] arg read SNP attributes from genotype
filename: txt, ARFF, plink (map/ped,
binary, raw)
--snp-file-type arg Ignore file extension and use type:
textwhitesp, wekaarff, plinkped,
plinkbed, plinkraw, mayogeo, birdseed
-n [ --numeric-data ] arg read continuous attributes from
PLINK-style covar file
-X [ --numeric-transform ] arg perform numeric transformation:
normalize, standardize, zscore, log,
sqrt
-a [ --alternate-pheno-file ] arg specifies an alternative
phenotype/class label file; one value
per line
-g [ --ec-algorithm-steps ] arg (=all)
EC steps to run (all|rj|rf)
-t [ --ec-num-target ] arg (=0) EC N_target - target number of
attributes to keep
-r [ --ec-iter-remove-n ] arg (=0) Evaporative Cooling number of
attributes to remove per iteration
-p [ --ec-iter-remove-percent ] arg Evaporative Cooling precentage of
attributes to remove per iteration
-O [ --out-dataset-filename ] arg write a new tab-delimited data set with
EC filtered attributes
-o [ --out-files-prefix ] arg (=ec_run)
use prefix for all output files
--snp-metric arg (=gm) metric for determining the difference
between subjects (gm|am|nca|nca6)
-B [ --snp-metric-nn ] arg (=gm) metric for determining the difference
between subjects (gm|am|nca|nca6|km)
-W [ --snp-metric-weights ] arg (=gm) metric for determining the difference
between SNPs (gm|am|nca|nca6)
-N [ --numeric-metric ] arg (=manhattan)
metric for determining the difference
between numeric attributes
(manhattan=|euclidean)
-R [ --rj-run-mode ] arg (=1) Random Jungle run mode: 1
(default=library call) / 2 (system
call)
-j [ --rj-num-trees ] arg (=1000) Random Jungle number of trees to grow
--rj-mtry arg (=0) Random Jungle size of randomly chosen
variable sets, DEFAULT: sqrt(ncol)
--rj-nimpvar arg (=1) Random Jungle only necessary if
backsel>0. SIZE=[1-...] how many
variable should remain
--rj-impmeasure arg (=1) Random Jungle importance method (see RJ
docs)
--rj-backsel arg (=0) Random Jungle backward elimination (see
RJ docs)
-Y [ --rj-tree-type ] arg (=1) Random Jungle tree type: 1 (default)-5
(see RJ docs)
-M [ --rj-memory-mode ] arg (=0) Random Jungle memory mode: 0
(default=double) / 1 (float) / 2 (char)
-x [ --snp-exclusion-file ] arg file of SNP names to be excluded
-k [ --k-nearest-neighbors ] arg (=10)
set k nearest neighbors
-m [ --number-random-samples ] arg (=0)
number of random samples (0=all|1 <= n
<= number of samples)
-b [ --weight-by-distance-method ] arg (=equal)
weight-by-distance method
(equal|one_over_k|exponential)
--weight-by-distance-sigma arg (=2) weight by distance sigma
-d [ --diagnostic-tests ] arg performs diagnostic tests and sends
output to filename without running EC
-D [ --diagnostic-levels-file ] arg write diagnostic attribute level counts
to filename
--dge-counts-data arg read digital gene expression counts
from text file
--dge-norm-factors arg read digital gene expression
normalization factors from text file
--birdseed-snps-data arg read SNP data from a birdseed formatted
file
--birdseed-phenos-data arg read birdseed subjects phenotypes from
a text file
--birdseed-subjects-labels arg read subject labels from filename to
override names from data file
--birdseed-include-snps arg include the SNP IDs listed in the text
file
--birdseed-exclude-snps arg exclude the SNP IDs listed the text
file
--distance-matrix arg create a distance matrix for the loaded
samples and exit
--gain-matrix arg create a GAIN matrix for the loaded
samples and exit
--dump-titv-file arg file for dumping SNP
transition/transversion information
All commands will include an input file (-s/--snp-data), and, optionally,
an output file prefix (-o/--output-files-prefix).
To perform a standard, all-default-parameters analysis,
./ec_static -s snpdata.ped -o result
This will use genotype/phenotype information from snpdata.ped, a PLINK
plaintext GWAS file, in the feature selection. All of the output files
produced will be prepended with 'result'.
This produces a file called result.ec, in which the SNPs are ranked
in descending order.
For additional examples, see the EC page on our research website.
See AUTHORS file.
B.A. McKinney, J.E. Crowe, Jr., J. Guo, and D. Tian, ÒCapturing the spectrum of interaction effects in genetic association studies by simulated evaporative cooling network analysis,Ó PLoS Genetics. 5(3): e1000432. doi:10.1371/journal.pgen.1000432; 2009.
McKinney, B.A., Reif, D.M., White, B.C., Crowe, J.E., Moore, J.H. Evaporative cooling feature selection for genotypic data involving interactions. Bioinformatics 23, 2113-2120 (2007). [PubMed]