glk-SVM

Introduction

The glk-SVM provides implementation of a novel SVM kernel method using glk-patterns as features for DNA sequences and the related classification tasks.

Installation

Download the zip file and  extract the source codes.

$ cd glk-SVM

$ make

If successful, you could find the following executables in the current directory:

• glk_kernel
• glk_train
• glk_classify
• glk_pattern

make install will simply copy these three executables to the ../bin direcory

Usage

Here we introduce the basic workflow of glk-SVM. Please refer to help messages for more detailed information of each program. You can find the instructions by running the programs without any argument.

(1) Calculating kernel matrix

A pre-computed kernel matrix is needed to train SVM classifiers. Here we are going to use test/test_positive.fa as a positive sequence set, and test/test_negative.fa as a negative set. The resulting kernel matrix is stored in test_kernel.out.

Type:

$ ./glk_kernel -g 2 -l 10 -k 8 test/test_positives.fa test/test_negatives.fa test_kernel.out

The kernel matrix is in test_kernel.out. We also add the same file in the test directory test/test_kernel.out and these two files should be identical.

(2) Training SVM

Now we can train the SVM classifier using the kernel matrix from previous step. This training method will take four arguments: kernel file, positive sequence file, negative sequence file, and output file prefix.

Type:

./glk_train test_kernel.out test/test_positives.fa test/test_negatives.fa test

The training process will generate two files, test_svalpha.out and test_svseq.fa. test_svalpha.out contains the weight values for the support vector sequences, test_svseq.fa contains the sequences in fasta format. These two files will be used in the next classification step.

(3) Classification using SVM

Now we can use glk_classify to score/classify any set of DNA sequences. Here, we will use trained classifier to score the input positive training sequences. We should specify the same set of parameters as used in the first step to ensure correct classification.

Type:

$ ./glk_classify -g 2 -l 10 -k 8 test/test_positives.fa test_svseq.fa test_svalpha.out pos_classify.out

(4) Generating the most differential glk-ci-patterns

This program could be used to generate the glk-ci patterns with strongest differential occurrence between the positive and negative sequences. And we could set the proper parameter value of $l$ according to the maximal differences. A possible range of $l$ could be provided using -f from -t to.

Type:

$ ./glk_pattern -g 2 -f 8 -t 10 -k 8 test/test_positives.fa test/test_negatives.fa test_patterns.out

In the test_patterns.out will list the top N glk-ci patterns with strongest differential occurrence between the positive and negative sequences and the corresponding count in the positive and negative set.

Distribution

• README -- This file
• Makefile -- Makefile for compile
• src/ -- The source code files
• test/ -- Example sequence files for testing