Remote homolog detection using local sequence–structure correlations

Abstract

Remote homology detection refers to the detection of structural homology in proteins when there is little or no sequence similarity. In this article, we present a remote homolog detection method called SVM-HMMSTR that overcomes the reliance on detectable sequence similarity by transforming the sequences into strings of hidden Markov states that represent local folding motif patterns. These state strings are transformed into fixeddimension feature vectors for input to a support vector machine. Two sets of features are defined: an order-independent feature set that captures the amino acid and local structure composition; and an order-dependent feature set that captures the sequential ordering of the local structures. Tests using the Structural Classification of Proteins (SCOP) 1.53 data set show that the SVM-HMMSTR gives a significant improvement over several current methods. Proteins 2004;57:518 -530.

Protein Structure Prediction using String Kernels

“…In particular, a number of different methods have been developed that use support vector machines (SVM) [56] to produce results that are generally superior to those produced by either pairwise sequence comparisons or approaches based on generative modelsprovided there is sufficient training data. [19,35,33,34,17,18,52,31].…”

“…All three of these methods represent a sequence £ as a vector in this simpler feature space, but differ in the scheme they employ to actually determine if a particular dimension is above a user-supplied threshold. An entirely different feature space is explored by the SVM-Isites [17] and SVM-HMMSTR [18] methods that take advantage of a set of local structural motifs (SVM-Isites) and their relationships (SVM-HMMSTR).…”

“…The SVM-Fisher, SVM-Pairwise, LA-Kernel, and Mismatch results were obtained from [52]. The SVM-HMMSTR results were obtained from [18]and correspond to the best-performing scheme (the authors did not report ROC values). The Profile results were obtained locally by running the publicly available implementation of the scheme obtained from the authors.…”

“…The same dataset and classification problems 2 have been used in a number of earlier studies [35,18,52] Employing the same dataset and overall methodology as in remote homology detection, we simulate fold detection by formulating it as a fold classification problem within the context of SCOP's hierarchical classification scheme. In this setting, protein domains within the same superfamily are considered positive test examples, and protein domains within the same fold but outside the superfamily are considered positive training examples.…”

“…Table 3 and Table 4 compare the performance of the various kernel functions developed in this paper against that achieved by a number of previously developed schemes for the superfamily-and fold-level classification problems, respectively. In the case of the superfamily-level classification problem, the performance is compared against SVM-Fisher [19], SVM-Pairwise [35], and different instances of the LA-Kernel [52], SVM-HMMSTR [18], Mismatch [34], and Profile [31].…”