aboutsummaryrefslogtreecommitdiff
path: root/src/matlab/quikrTrain.m
diff options
context:
space:
mode:
Diffstat (limited to 'src/matlab/quikrTrain.m')
-rw-r--r--src/matlab/quikrTrain.m41
1 files changed, 41 insertions, 0 deletions
diff --git a/src/matlab/quikrTrain.m b/src/matlab/quikrTrain.m
new file mode 100644
index 0000000..7960cca
--- /dev/null
+++ b/src/matlab/quikrTrain.m
@@ -0,0 +1,41 @@
+function mat=quikrTrain(inputfasta,k)
+%matrix=QuikrTrain(inputfasta,k) Returns the (sparse) k-mer sensing
+%matrix from the FASTA file located at 'inputfasta' for the k-mer size k.
+%This serves to retrain the Quikr method. The ouput can then be fed to
+%quikrCustomTrained().
+%Works for k=1:8 (typically the matrices get too large for k>9)
+%The filename for the inputfasta must be the complete path.
+
+if nargin>2
+ error('too many input arguments');
+end
+
+
+[pathtofile,filename,ext]=fileparts(inputfasta); %get the parts of the file
+
+
+outputfilename=fullfile(pathtofile, [filename sprintf('-sensingmatrixK%d.txt',k)]); %Currently this is coded to write a temporary file. In future versions, this will be all be done in RAM
+%The reason for writing the file to disk first is that Matlab typically
+%crashes when unix() returns as many entries as ./probabilities-by-read
+%does (on the order of ~2*10^10).
+
+kmerfilename=sprintf('%dmers.txt',k); %This contains the list of 6-mers to count. In future versions this will be computed locally instead of being read in.
+
+if (isunix && not(ismac)) %this is for the linux version
+ unix(['./probabilities-by-read-linux ' sprintf('%d',k) ' ' inputfasta ' ' kmerfilename ' > ' outputfilename]); %obtain the k-mer counts of the inputfasta read-by-read
+elseif ismac %mac version
+ unix(['./probabilities-by-read-osx ' sprintf('%d',k) ' ' inputfasta ' ' kmerfilename ' > ' outputfilename]); %obtain the k-mer counts of the inputfasta read-by-read
+elseif ispc %No PC version
+ error('Windows is not yet supported');
+end
+
+fid=fopen(outputfilename); %open the output file
+
+%A=textscan(fid,'%f'); %get all the counts
+%A=A{:};
+A=fscanf(fid,'%f');
+mat=sparse(reshape(A,4^k,length(A)/4^k)); %form into a matrix
+mat=bsxfun(@rdivide,mat,sum(mat,1)); %column-normalize
+fclose(fid); %close file
+delete(outputfilename); %delete the file
+