Protein fold recognition using geometric kernel data fusion

Abstract

Motivation: Various approaches based on features extracted from protein sequences and often machine learning methods have been used in the prediction of protein folds. Finding an efficient technique for integrating these different protein features has received increasing attention. In particular, kernel methods are an interesting class of techniques for integrating heterogeneous data. Various methods have been proposed to fuse multiple kernels. Most techniques for multiple kernel learning focus on learning a convex linear combination of base kernels. In addition to the limitation of linear combinations, working with such approaches could cause a loss of potentially useful information.Results: We design several techniques to combine kernel matrices by taking more involved, geometry inspired means of these matrices instead of convex linear combinations. We consider various sequence-based protein features including information extracted directly from position-specific scoring matrices and local sequence alignment. We evaluate our methods for classification on the SCOP PDB-40D benchmark dataset for protein fold recognition. The best overall accuracy on the protein fold recognition test set obtained by our methods is ∼86.7%. This is an improvement over the results of the best existing approach. Moreover, our computational model has been developed by incorporating the functional domain composition of proteins through a hybridization model. It is observed that by using our proposed hybridization model, the protein fold recognition accuracy is further improved to 89.30%. Furthermore, we investigate the performance of our approach on the protein remote homology detection problem by fusing multiple string kernels.Availability and implementation: The MATLAB code used for our proposed geometric kernel fusion frameworks are publicly available at http://people.cs.kuleuven.be/∼raf.vandebril/homepage/software/geomean.php?menu=5/Contact: pooyapaydar@gmail.com or yves.moreau@esat.kuleuven.beSupplementary information: Supplementary data are available at Bioinformatics online.

Complete fold annotation of the human proteome using a novel structural feature space

“…Each fold has at least 11 training examples. The accuracy of our classifiers are shown in Table 1 along with the results reported by several other published methods121516171819. Since some of the sequences in these benchmarks are similar to the templates that make up our feature space (that is, similar to the sequences of the domains from which the templates are derived), we ran our classifiers both with and without filtering of these template-similar sequences (Table 1; “filtered” versions correspond to benchmarks where sequences with >25% pairwise identity with any template were removed; see Methods).…”

In silico toxicity prediction of chemicals from EPA toxicity database by kernel fusion-based support vector machines

“…From our previous studies, the kernel in kernel methods mainly acts as an interface between the input data module and the learning algorithms [13]. Selecting the best kernel, which is rich in the classification information, is the key step for the application to kernel methods [45]. However, it is very difficult to select the best kernels in practice, because the selection of the kernel largely depends on the encoding of our prior knowledge about the data and the types of patterns we can expect to identify.…”

Determining Microbial Products and Identifying Molecular Targets in the Human Microbiome

“…Hess et al (Hess et al, 2009), for example, used an ensemble method combining three different Bayesian approaches that integrated sequence features, coexpression, protein-protein interactions, binding sites, and subcellular localization to improve protein function prediction. Other studies have combined tertiary structure with primary structural homology and secondary structure biochemistry (Wang et al, 2014); sequence data with gene expression, protein-protein interactions, and evolutionary conservation (Cozzetto et al, 2013); sequence with interaction profiles and domain co-occurrence (Wang et al, 2013); and sequence features with active site motifs and structural alignment for determination of protein fold activities (Zakeri et al, 2014). Purely technical issues have impeded the application of these methods to microbial genomes and metagenomes.…”