Comparison of SVM and Spectral Embedding in Promoter Biobricks' Categorizing and Clustering

TL;DR

This study compares spectral embedding and SVM in classifying prokaryotic promoters, finding SVM outperforms spectral methods with over 93% accuracy in predicting promoter functions.

Contribution

The paper evaluates and contrasts spectral embedding and SVM on promoter classification, demonstrating the superior accuracy of SVM for this task.

Findings

SVM achieved 93.07% accuracy in promoter classification.

Spectral embedding based on edit distance was less effective.

SVM outperforms spectral methods in this dataset.

Abstract

Background: In organisms' genomes, promoters are short DNA sequences on the upstream of structural genes, with the function of controlling genes' transcription. Promoters can be roughly divided into two classes: constitutive promoters and inducible promoters. Promoters with clear functional annotations are practical synthetic biology biobricks. Many statistical and machine learning methods have been introduced to predict the functions of candidate promoters. Spectral Eigenmap has been proved to be an effective clustering method to classify biobricks, while support vector machine (SVM) is a powerful machine learning algorithm, especially when dataset is small. Methods: The two algorithms: spectral embedding and SVM are applied to the same dataset with 375 prokaryotic promoters. For spectral embedding, a Laplacian matrix is built with edit distance, followed by K-Means Clustering. The sequences are represented by numeric vector to serve as dataset for SVM trainning. Results: SVM achieved a high predicting accuracy of 93.07% in 10-fold cross validation for classification of promoters' transcriptional functions. Laplacian eigenmap (spectral embedding) based on editing distance may not be capable for extracting discriminative features for this task.