Shuichi Hirose

Predicting mostly disordered proteins by using structure-unknown protein data

Predicting intrinsically disordered proteins is important in structural biology because they are thought to carry out various cellular functions even though they have no stable three-dimensional structure. We know the structures of far more ordered proteins than disordered proteins. The structural distribution of proteins in nature can therefore be inferred to differ from that of proteins whose structures have been determined experimentally. We know many more protein sequences than we do protein structures, and many of the known sequences can be expected to be those of disordered proteins. Thus it would be efficient to use the information of structure-unknown proteins in order to avoid training data sparseness. We propose a novel method for predicting which proteins are mostly disordered by using spectral graph transducer and training with a huge amount of structure-unknown sequences as well as structure-known sequences. When the proposed method was evaluated on data that included 82 disordered proteins and 526 ordered proteins, its sensitivity was 0.723 and its specificity was 0.977. It resulted in a Matthews correlation coefficient 0.202 points higher than that obtained using FoldIndex, 0.221 points higher than that obtained using the method based on plotting hydrophobicity against the number of contacts and 0.07 points higher than that obtained using support vector machines (SVMs). To examine robustness against training data sparseness, we investigated the correlation between two results obtained when the method was trained on different datasets and tested on the same dataset. The correlation coefficient for the proposed method is 0.14 higher than that for the method using SVMs. When the proposed SGT-based method was compared with four per-residue predictors (VL3, GlobPlot, DISOPRED2 and IUPred (long)), its sensitivity was 0.834 for disordered proteins, which is 0.052–0.523 higher than that of the per-residue predictors, and its specificity was 0.991 for ordered proteins, which is 0.036–0.153 higher than that of the per-residue predictors. The proposed method was also evaluated on data that included 417 partially disordered proteins. It predicted the frequency of disordered proteins to be 1.95% for the proteins with 5%–10% disordered sequences, 1.46% for the proteins with 10%–20% disordered sequences and 16.57% for proteins with 20%–40% disordered sequences. The proposed method, which utilizes the information of structure-unknown data, predicts disordered proteins more accurately than other methods and is less affected by training data sparseness.BackgroundPredicting intrinsically disordered proteins is important in structural biology because they are thought to carry out various cellular functions even though they have no stable three-dimensional structure. We know the structures of far more ordered proteins than disordered proteins. The structural distribution of proteins in nature can therefore be inferred to differ from that of proteins whose structures have been determined experimentally. We know many more protein sequences than we do protein structures, and many of the known sequences can be expected to be those of disordered proteins. Thus it would be efficient to use the information of structure-unknown proteins in order to avoid training data sparseness. We propose a novel method for predicting which proteins are mostly disordered by using spectral graph transducer and training with a huge amount of structure-unknown sequences as well as structure-known sequences.ResultsWhen the proposed method was evaluated on data that included 82 disordered proteins and 526 ordered proteins, its sensitivity was 0.723 and its specificity was 0.977. It resulted in a Matthews correlation coefficient 0.202 points higher than that obtained using FoldIndex, 0.221 points higher than that obtained using the method based on plotting hydrophobicity against the number of contacts and 0.07 points higher than that obtained using support vector machines (SVMs). To examine robustness against training data sparseness, we investigated the correlation between two results obtained when the method was trained on different datasets and tested on the same dataset. The correlation coefficient for the proposed method is 0.14 higher than that for the method using SVMs. When the proposed SGT-based method was compared with four per-residue predictors (VL3, GlobPlot, DISOPRED2 and IUPred (long)), its sensitivity was 0.834 for disordered proteins, which is 0.052–0.523 higher than that of the per-residue predictors, and its specificity was 0.991 for ordered proteins, which is 0.036–0.153 higher than that of the per-residue predictors. The proposed method was also evaluated on data that included 417 partially disordered proteins. It predicted the frequency of disordered proteins to be 1.95% for the proteins with 5%–10% disordered sequences, 1.46% for the proteins with 10%–20% disordered sequences and 16.57% for proteins with 20%–40% disordered sequences.ConclusionThe proposed method, which utilizes the information of structure-unknown data, predicts disordered proteins more accurately than other methods and is less affected by training data sparseness.

Prediction of protein motions from amino acid sequence and its application to protein-protein interaction

BackgroundStructural flexibility is an important characteristic of proteins because it is often associated with their function. The movement of a polypeptide segment in a protein can be broken down into two types of motions: internal and external ones. The former is deformation of the segment itself, but the latter involves only rotational and translational motions as a rigid body. Normal Model Analysis (NMA) can derive these two motions, but its application remains limited because it necessitates the gathering of complete structural information.ResultsIn this work, we present a novel method for predicting two kinds of protein motions in ordered structures. The prediction uses only information from the amino acid sequence. We prepared a dataset of the internal and external motions of segments in many proteins by application of NMA. Subsequently, we analyzed the relation between thermal motion assessed from X-ray crystallographic B-factor and internal/external motions calculated by NMA. Results show that attributes of amino acids related to the internal motion have different features from those related to the B-factors, although those related to the external motion are correlated strongly with the B-factors. Next, we developed a method to predict internal and external motions from amino acid sequences based on the Random Forest algorithm. The proposed method uses information associated with adjacent amino acid residues and secondary structures predicted from the amino acid sequence. The proposed method exhibited moderate correlation between predicted internal and external motions with those calculated by NMA. It has the highest prediction accuracy compared to a naïve model and three published predictors.ConclusionsFinally, we applied the proposed method predicting the internal motion to a set of 20 proteins that undergo large conformational change upon protein-protein interaction. Results show significant overlaps between the predicted high internal motion regions and the observed conformational change regions.

Feature Selection Based on Physicochemical Properties of Redefined N-term Region and C-term Regions for Predicting Disorder

The prediction of intrinsic disorder from amino acid sequence has been gaining increasing attention because these have come to be known as important regions for protein functions. The most common way of predicting disorder is based on binary classification with machine learning. Since amino acid composition has different propensities in the N-term, C-term, and internal regions, the accuracy of prediction increases by dividing training data into these three regions and predicting them separately. However, previous work has lacked discussion about a concrete definition of the N-term and C-term regions, and has only used the heuristic length from the terminal. Other previous work has shown that general physicochemical properties rather than specific amino acids are important factors contributing to disorder, and a reduced amino acid alphabet can maintain excellent precision in predicting disorder. In this paper, we redefine a suitable length and position for the N-term and C-term regions for predicting disorder. Moreover, we show that each region has different physicochemical properties, which are important factors contributing to disorder. We also suggest a region-specific-reduced set of amino acid and modified PSSM based on that for predicting disorder. We implemented our method and (1) compare it with the conventional division method, (2) compare our feature selection with all physicochemical features, on casp6 benchmark, PDB dataset, and DisProt. The result supports that the method of new data separation is effective, and indicates each region has different physicochemical properties that are important factors for predicting protein disorders.