Efficient remote homology detection using local structure

Abstract

I-sites server is accessible through the web at http://www.bioinfo.rpi.edu. Programs are available upon request for academics. Licensing agreements are available for commercial interests. The framework of encoding local structure into feature vector is available upon request.

Remote homolog detection using local sequence–structure correlations

“…For example, the relative performance of SVM-Fisher and SAM agrees with the results given in Jaakkola et al, 10 as does the relative performance of SAM and PSI-BLAST with the results given in Park et al 28 and the relative performance of SVM-I-sites and SVM-pairwise given in Hou et al 16 One surprise is the magnitude of the difference between SVM-pairwise and the 2 methods, SVM-Fisher and SAM, that directly or indirectly utilize SAM-T99 model. It is reported in Liao and Noble 11 that SVM-pairwise significantly outperforms these 2 methods, which is not the case in this work.…”

“…SVM-pairwise uses the pairwise Smith-Waterman sequence similarity algorithm in place of the gradient vector in the SVM-Fisher method that we described earlier. In contrast, SVM-I-sites encodes the local structure composition of a protein as the sum of I-sites motif confidence scores, 16,17 where each motif defines one feature. After the vectorization step, all of the SVM-based methods will define a similarity score for 2 proteins based on the feature vectors and use that similarity as the kernel of the classifier.…”

“…In our earlier work, we noted that similar structures contain similar local structure motifs. Our first SVM method, SVM-I-sites, 16 demonstrated that the local structure content of a protein improves remote homolog detection, even without knowing the locations of the local features within the sequence. Local structure motifs were found using the I-sites library of sequence-structure correlations.…”

Protein Structure Evolution and the Scop Database

“…These classifications were then extensively used for integrative structural data mining to develop predictive methods and structure comparison tools (Chou and Maggiora, 1998;Przytycka et al, 1999;Lackner et al, 2000;Bertone and Gerstein, 2001;Bukhman and Skolnick, 2001;Pasquier et al, 2001;Stambuk and Konjevoda, 2001;Torshin, 2001;Aung and Tan, 2006); (ii) SCOP classified proteins were extensively used in understanding evolution of protein enzymatic functions (Konin et al, 1998;Murzin, 1998;Powlowski and Godzik, 2001;Todd et al, 2001;George et al, 2004;Glasner, et al, 2006), evolutionary change of protein folds (Caetano and Caetano, 2005;Panchenko et al, 2005;Grishin, 2001;Lupas et al, 2001;Zhang and DeLisi, 2001), and hierarchical structural evolution (Dokholyan and Shakhnovich, 2001;Paoli, 2001); (iii) SCOP classification of proteins at superfamily and fold levels were used to study distantly related proteins with the same fold (Grigoriev et al, 2001;Teichmann et al, 2001;Thornton, 2001;Hou et al, 2003;Sandhya et al, 2005;Melo and Marti, 2006); (iv) SCOP is used to study sequence and structure variability and their dependence in homologous proteins (D'Alfonso et al, 2001;Gowri et al, 2003;Panchenko et al, 2005); (v) SCOP families are used to derive amino acid similarity matrices and substitution tables useful for sequence comparison and fold recognition studies (Dosztanyi and Torda, 2001;Shi et al, 2001;Dunbrack, 2006); (vi) SCOP is helpful in studying the structural anatomy of folds and doma...…”

Using distances between Top-n-gram and residue pairs for protein remote homology detection

“…Spectrum Kernel [21] searched all possible subsequences of length k from a alphabet to form a feature map. SVM-I-sites [22] encoded structure information into the feature vectors. Mismatch kernel [23] was calculated based on shared occurrences of ( k , m )-patterns in the data.…”