Development of a structure based protein function prediction method: Calcium binding protein

Abstract

Three-dimensional local structures located near active sites are responsible for protein functions. To predict protein functions, we developed a method to extract three-dimensional structural characteristics of functions. We chose EF-hand calcium binding proteins as the model target. We defined a local structure as the structure within 12Å from an arbitrary Asp(C α). Analysis of the amino acid propensities in the active sites showed that there were clear preferences for certain amino acids to locate at the binding sites. Furthermore, the analysis of the distributions of the distances between Asp(C α) and any other amino acids(C α) showed that there were preferred distances for respective amino acids. Using such structural data, we derived a DB (distance based) score and a DPB (distance and propensity based) score for the prediction of functional sites. A search for EF-hand calcium binding sites was performed to test the method. EF-hand calcium binding sites were successfully predicted.

GFP Tagging of Sieve Element Occlusion (SEO) Proteins Results in Green Fluorescent Forisomes

“…We are far from excluding the possibility that essential component(s) of forisomes, perhaps the one(s) responsible for cation binding, might still await detection. However, we also note that with an increasing number of well-characterized proteins, variants of calcium-binding structures have become known that would not have been classified correctly before their first publication (Asaoka et al 2003 , Rigden and Galperin 2004 , Zhou et al 2006 ). Quite conceivably, SEO proteins might possess Ca 2+ -binding sites which cannot be identified through sequence similarities to known motifs at this time.…”

タンパク質立体構造情報を利用した機能予測法fcanal：酵素・結合タンパク質への応用

“…We made a score matrix for each functional site (Figure 2) mostly according to the previous report [1]. The abstract of the flow chart of the FCANAL with slight modification is as follows:…”

“…In the previous report, we have developed a method which extracts 3D structural characteristics of functions, the FCANAL (Fast Calculable protein function ANALyzer) [1]. It used protein motif sequences for identifying the key residues for extracting 3D features, which limited the number of predictable proteins.…”

Combining Molecular Dynamics and Machine Learning to Improve Protein Function Recognition

“…These methods employed artificial neural networks [13], graph theory and geometric similarity [14], bond-valence calculations [15,16] and distribution of distances between C α atoms of residues [17]. We have previously described FEATURE [18], a machine learning algorithm which employs Bayesian scoring scheme, and its ability to identify Ca 2+ binding sites [19].…”