A coverage criterion for spaced seeds and its applications to SVM string kernels and k-mer distances

[for additional scripts, see this webpage]

This page contains experimental datasets, results and associated scripts related to the paper A coverage criterion for spaced seeds and its applications to SVM string-kernels and k-mer distances. The content is divided into two parts that are related respectively to the sections 4.1. Coverage sensitivity and spaced seed string kernels and 4.2 Coverage sensitivity and alignment-free distance for sequence comparison of the paper.

The C++ is integrated into the Iedera 1.06α7 tool available at http://bioinfo.lifl.fr/yass/iedera.php

The Matlab (Octave compatible) code is available upon request to Donald E. K. Martin.

4.1. Coverage sensitivity and spaced seed string kernels

RFAM v11.0 → (splitfasta.pl) → Rfam/RF*.fa → (main.pl) → results

The initial dataset used is a multi-fasta file composed of 383004 sequences belonging to 2208 families from RFAM v11.0 [ftp://ftp.ebi.ac.uk/pub/databases/Rfam/11.0/Rfam.fasta.gz]. This multi-fasta file is processed to generate a set of (name normalized) fasta files using a script [splitfasta.pl].

The "Rfam" directory containing the fasta files is then processed by a second script [main.pl].

For each of set of seed(s) [seeds.txt], this script does most of the work :

  1. it computes and splits (into train and test) the features (spaced k-mer count),
  2. it learns RNA classes on the training set, then classifies the testing set, with help of the SVM classifier from the svm-multiclass package [dir|url],
  3. it outputs, for each seed, the zero one-error rate for the SVM, computes and outputs the theoretical sensitivities (for each model : single-hit, multiple-hit, coverage) with the help of iedera [dir]),
  4. it outputs the correlation values between zero one-error rates for the SVM, and each or the theoretical sensitivities (for all the seeds of a given family size and a given weight) with the help of a script [corr.pl], and plots them [main-w3-n1.gp|main-w3-n2.gp|main-w4-n1.gp|main-w4-n2.gp].

main-w3-n1main-w3-n2 main-w4-n1main-w4-n2

Tools, Data & Results :

4.2 Coverage sensitivity and alignment-free distance for sequence comparison

Given a set of seed(s), [seeds.txt], a main script [main.step.pl] computes, for each seed, for each alignment percentage of identity, and each of the 1000 trials, the Hamming distance, the estimated multi-hit and coverage values. During computation, it updates two correlation values, between the Hamming distances and each of the multi-hit values and coverage values. At the end, it outputs these two correlation values for each seed.

Correlation values [results.w3-n1.dat|results.w3-n2.dat|results.w4-n1.dat|results.w4-n2.dat] are then plotted with a script [results.gp].

[For additional tests, please see this webpage]

results-labels

Tools, Data & Results :