DectICO: an alignment-free supervised metagenomic classification method based on feature extraction and dynamic selection
By Xiao Ding, Fudong Cheng, Chang-Chang Cao, Xiao Sun. 2015
11 citations recorded by the source corpus.
Continual progress in next-generation sequencing allows for generating increasingly large metagenomes which are over time or space. Comparing and classifying the metagenomes with different microbial communities is critical. Alignment-free supervised classification is important for discriminating between the multifarious components of metagenomic samples, because it can be accomplished independently of known microbial genomes. We propose an alignment-free supervised metagenomic classification method called DectICO. The intrinsic correlation of oligonucleotides provides the feature set, which is selected dynamically using a kernel partial least squares algorithm, and the feature matrices extracted with this set are sequentially employed to train classifiers by support vector machine (SVM). We evaluated the classification performance of DectICO on three actual metagenomic sequencing datasets, two containing deep sequencing metagenomes and one of low coverage. Validation results show that DectICO is powerful, performs well based on long oligonucleotides (i.e., 6-mer to 8-mer), and is more stable and generalized than a sequence-composition-based method. The classifiers trained by our method are more accurate than non-dynamic feature selection methods and a recently published recursive-SVM-based classification approach. The alignment-free supervised classification method DectICO can accurately classify metagenomic samples without dependence on known microbial genomes. Selecting the ICO dynamically offers better stability and generality compared with sequence-composition-based classification algorithms. Our proposed method provides new insights in metagenomic sample classification.
Subfields: Molecular Biology
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Molecular Biology
- Casey S. Greene
- Yifan Peng
- Brett K. Beaulieu-Jones
- Macha Nikolski
- Hayssam Soueidan
- G. Way
- Amr M. Alexandari
- Zhiyong Lu
Citations in the corpus
- Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “Several groups have proposed methods to determine host or environment phenotypes from the organisms that are identified [234–237] or overall sequence composition [238].”
- Machine learning for metagenomics: methods and tools (2015, methodology): “This measure has been recently employed in the DectICO method (Ding et al., 2015) (see section 7.”