Feature selection for gene prediction in metagenomic fragments

Abstract

BackgroundComputational approaches, specifically machine-learning techniques, play an important role in many metagenomic analysis algorithms, such as gene prediction. Due to the large feature space, current de novo gene prediction algorithms use different combinations of classification algorithms to distinguish between coding and non-coding sequences.ResultsIn this study, we apply a filter method to select relevant features from a large set of known features instead of combining them using linear classifiers or ignoring their individual coding potential. We use minimum redundancy maximum relevance (mRMR) to select the most relevant features. Support vector machines (SVM) are trained using these features, and the classification score is transformed into the posterior probability of the coding class. A greedy algorithm uses the probability of overlapped candidate genes to select the final genes. Instead of using one model for all sequences, we train an ensemble of SVM models on mutually exclusive datasets based on GC content and use the appropriated model to classify candidate genes based on their read’s GC content.ConclusionOur proposed algorithm achieves an improvement over some existing algorithms. mRMR produces promising results in gene prediction. It improves classification performance and feature interpretation. Our research serves as a basis for future studies on feature selection for gene prediction.

Literature on Applied Machine Learning in Metagenomic Classification: A Scoping Review

“…This can safely be done for redundant features, i.e., features that do not give meaningful information about the class, are correlated, or are derived from other features in the data set. Amani Al-Ajlan and Achraf El Allali proposed a methodology [ 72 ] for feature selection using maximum Relevance Minimum Redundancy (mRMR) to find the most relevant features. The feature extraction algorithm has shown good results in improving classification results from Support Vector Machine (SVM)-based models.…”

Machine learning applications in RNA modification sites prediction

“…In addition, strategies such as principal component analysis (PCA) [37] , n-Grams, minimal-redundance maximum-Relevance (mRMR) [38] are widely used in order to select a subset of the features. Studies show that applying feature selection algorithms produce better performance than using the extracted features directly or applying a multi-layer machine learning approach [39] , [40] . However, feature selection should be performed on a different dataset than the training to avoid biases in the performance analysis during testing.…”

ORFhunteR: an accurate approach for the automatic identification and annotation of open reading frames in human mRNA molecules

“…Other tools, often used for metagenomics analysis, are based on vectorization of sequence features of ORF-candidates, which is an efficient conversion of nucleotide sequence into a vector of sequence features (Al-Ajlan and El Allali, 2018;Al-Ajlan and El Allali, 2019;El Allali and Rose, 2013;Hoff, et al, 2008;Trimble, et al, 2012;Zhang, et al, 2017). These features can be evaluated by a machine learning models in order to select true ORFs.…”

“…A promising way that could wave the main drawbacks of existing algorithms, is to select the informative parameters describing candidate ORF fragments using advanced algorithms of nucleotide sequence vectorization (Bao, et al, 2014;Mao, et al, 2014), and then to apply an optimal prediction algorithm and identify the most probable or true ORF sequence among candidates. This approach needs only a -4 -limited set of input features (here we used 104), which is significantly less than 4000-5000 considered earlier (Al-Ajlan and El Allali, 2018). The use of a random forest classifier (Breiman, 2001) has several advantages over more complex deep learning techniques (Al-Ajlan and El Allali, 2019;Wen, et al, 2019).…”