蛋白质二级结构预测: 基于词条的最大熵马尔科夫方法

提出了一种新的蛋白质二级结构预测方法. 该方法从氨基酸序列中提取出和自然语言中的“词”类似的与物种相关的蛋白质二级结构词条, 这些词条形成了蛋白质二级结构词典, 该词典描述了氨基酸序列和蛋白质二级结构之间的关系. 预测蛋白质二级结构的过程和自然语言中的分词和词性标注一体化的过程类似. 该方法把词条序列看成是马尔科夫链, 通过Viterbi算法搜索每个词条被标注为某种二级结构类型的最大概率, 其中使用词网格描述分词的结果, 使用最大熵马尔科夫模型计算词条的二级结构概率. 蛋白质二级结构预测的结果是最优的分词所对应的二级结构类型. 在4个物种的蛋白质序列上对这种方法进行测试, 并和PHD方法进行比较. 试验结果显示, 这种方法的Q3准确率比PHD方法高3.9%, SOV准确率比PHD方法高4.6%. 结合BLAST搜索的局部相似的序列可以进一步提高预测的准确率. 在50个CASP5目标蛋白质序列上进行测试的结果是: Q3准确率为78.9%, SOV准确率为77.1%. 基于这种方法建立了一个蛋白质二级结构预测的服务器, 可以通过http://www.insun.hit.edu.cn:81/demos/biology/index.html来访问.     

核酸、氨基酸分类和蛋白质二级结构关系的分析与分子设计

核酸序列中包含一定的蛋白质结构信息。根据通常情况下遗传密码表中密码子中间位的碱基配对时产生的氢键数目,尝试将20种氨基酸划分为两类,并用自编的计算机软件对蛋白质二级结构数据库中两类氨基酸的类聚现象进行了统计分析。结果表明,使用这种方法对氨基酸进行划分后,氨基酸残基具有较大概率与划入同一类的氨基酸残基相邻出现,并且这种聚集体对二级结构具有一定的偏好性。最后按照该方法设计了一段氨基酸序列并给出了预测服务器预测得到的结构。     

核酸、氨基酸分类和蛋白质二级结构关系的分析与分子设计

核酸序列中包含一定的蛋白质结构信息。根据通常情况下遗传密码表中密码子中间位的碱基配对时产生的氢键数目,尝试将20种氨基酸划分为两类,并用自编的计算机软件对蛋白质二级结构数据库中两类氨基酸的类聚现象进行了统计分析。结果表明,使用这种方法对氨基酸进行划分后,氨基酸残基具有较大概率与划入同一类的氨基酸残基相邻出现,并且这种聚集体对二级结构具有一定的偏好性。最后按照该方法设计了一段氨基酸序列并给出了预测服务器预测得到的结构。     

利用氨基酸结构倾向性预测固有不规则蛋白质

为了降低固有不规则蛋白质预测模型中特征矩阵的稀疏性,提高预测模型的性能,提出一种利用氨基酸结构倾向性预测固有不规则蛋白质的方法.利用氨基酸结构倾向性将20种氨基酸进行分类,构建氨基酸简化集合,从间接角度提取氨基酸序列中蕴含的不规则结构特征,利用新的简化集合重新描述氨基酸序列,构建固有不规则蛋白质预测模型.预测结果表明,基于氨基酸结构倾向性的预测模型能够有效地挖掘氨基酸结构倾向性中隐藏的不规则结构特征信息,提高固有不规则蛋白质预测模型的预测精度.     

蛋白质-RNA序列结构界面偏好性及用于对接打分统计势的构建

本文对来自PDB (Protein Data Bank)数据库的蛋白质-RNA复合物结构构建了非冗余非核糖体数据库(694个结构),并对此数据库统计了蛋白质和RNA序列及二级结构的界面偏好性.结果发现,蛋白质β折叠、3₁₀-helix和RNA未配对核苷酸,尤其是未配对中空间排列不规整的核苷酸,具有显著的界面偏好性.据此,对二级结构进行归类,建立了考虑序列和二级结构信息的60×12氨基酸-核苷酸成对偏好势,并将其作为打分函数用于蛋白质-RNA对接中近天然结构的筛选.结果表明,该60×12统计势的打分成功率为65.77%,优于考虑蛋白质或RNA二级结构信息的统计势,及我们小组之前在251个结构上构建的60×8^*统计势.该工作有助于加深对蛋白质-RNA特异性识别的理解,可推动复合物结构预测的进展.     

使用伪氨基酸组成和模糊支持向量机预测蛋白质结构类

蛋白质结构类预测是生物信息和蛋白质科学中重要的研究领域.基于Chou提出的伪氨基酸离散模型框架,从蛋白质序列出发,设计一种新的伪氨基酸组成方法表示蛋白质序列样本.抽取氨基酸组合(10-D)在序列中出现的频率和疏水氨基酸模式(6-D)表示蛋白质序列的附加特征,用和传统的氨基酸组成(20-D)一起构成的36维的伪氨基酸组成向量来表示蛋白质序列的特征.使用遗传算法来优化附加特征的权重系数.伪氨基酸组成向量作为输入数据,模糊支持向量机作为预测工具.使用三个常用的标准数据集来验证算法的性能.Jack-knife检验结果说明本方法具有较高的准确率,有望成为潜在的预测蛋白质功能的工具.     

基于添加模体信息和功率谱密度的组合向量预测27类蛋白质折叠子

以序列相似性低于40%的1895条蛋白质序列构建涵盖27个折叠类型的蛋白质折叠子数据库,从蛋白质序列出发,用模体频数值、低频功率谱密度值、氨基酸组分、预测的二级结构信息和自相关函数值构成组合向量表示蛋白质序列信息,采用支持向量机算法,基于整体分类策略,对27类蛋白质折叠子的折叠类型进行预测,独立检验的预测精度达到了66.67%。同时,以同样的特征参数和算法对27类折叠子的4个结构类型进行了预测,独立检验的预测精度达到了89.24%。将同样的方法用于前人使用过的27类折叠子数据库,得到了好于前人的预测结果。     

基于打分矩阵的多类蛋白质折叠子的预测

依据蛋白质折叠子中氨基酸保守性,以氨基酸、氨基酸的极性、氨基酸的电性以及氨基酸的亲—疏水性为参数,从蛋白质的氨基酸序列出发,采用"一对多"的分类策略,通过构建打分矩阵和选取氨基酸序列模式片断,利用5种相似性打分函数对27类折叠子进行识别,最好的预测精度达到83.46%。结果表明,打分矩阵是预测多类蛋白质折叠子有效的方法。     

蛋白质的二级结构预测——Chou和Fasman方法

蛋白质分子的一切高级结构,都由一级结构即氨基酸残基序列所包含的信息决定。多年来,由蛋白质的氨基酸序列预测二级结构的方法不下十几种。其中,Chou和Fasman的方法自1974年提出,至1978年修正、精化,已得到了很好结果,越益受到重视。此方法的突出优点是简便,无须计算机的复杂分析,就可预测出蛋白质的二级结构,准确性约为80％。目前蛋白质二级结构的测定,当然以X-晶体衍射结果最准确。Chou和Fasman方法正是基于晶体分析的结果,经统计得出的一整套数据     

随机森林方法预测膜蛋白类型

膜蛋白的类型与其功能是密切相关的,因此膜蛋白类型的预测是研究其功能的重要手段,从蛋白质的氨基酸序列出发对膜蛋白的类型进行预测有重要意义。文章基于蛋白质的氨基酸序列,将组合离散增量和伪氨基酸组分信息共同作为预测参数,采用随机森林分类器,对8类膜蛋白进行了预测。在Jackknife检验下的预测精度为86.3%,独立检验的预测精度为93.8%,取得了好于前人的预测结果。     

An Evolution-Based Approach to De Novo Protein Design and Case Study on Mycobacterium tuberculosis

Computational protein design is a reverse procedure of protein folding and structure prediction, where constructing structures from evolutionarily related proteins has been demonstrated to be the most reliable method for protein 3-dimensional structure prediction. Following this spirit, we developed a novel method to design new protein sequences based on evolutionarily related protein families. For a given target structure, a set of proteins having similar fold are identified from the PDB library by structural alignments. A structural profile is then constructed from the protein templates and used to guide the conformational search of amino acid sequence space, where physicochemical packing is accommodated by single-sequence based solvation, torsion angle, and secondary structure predictions. The method was tested on a computational folding experiment based on a large set of 87 protein structures covering different fold classes, which showed that the evolution-based design significantly enhances the foldability and biological functionality of the designed sequences compared to the traditional physics-based force field methods. Without using homologous proteins, the designed sequences can be folded with an average root-mean-square-deviation of 2.1 Å to the target. As a case study, the method is extended to redesign all 243 structurally resolved proteins in the pathogenic bacteria Mycobacterium tuberculosis, which is the second leading cause of death from infectious disease. On a smaller scale, five sequences were randomly selected from the design pool and subjected to experimental validation. The results showed that all the designed proteins are soluble with distinct secondary structure and three have well ordered tertiary structure, as demonstrated by circular dichroism and NMR spectroscopy. Together, these results demonstrate a new avenue in computational protein design that uses knowledge of evolutionary conservation from protein structural families to engineer new protein molecules of improved fold stability and biological functionality.     

Structure-based prediction of DNA target sites by regulatory proteins

Regulatory proteins play a critical role in controlling complex spatial and temporal patterns of gene expression in higher organism, by recognizing multiple DNA sequences and regulating multiple target genes. Increasing amounts of structural data on the protein-DNA complex provides clues for the mechanism of target recognition by regulatory proteins. The analyses of the propensities of base-amino acid interactions observed in those structural data show that there is no one-to-one correspondence in the interaction, but clear preferences exist. On the other hand, the analysis of spatial distribution of amino acids around bases shows that even those amino acids with strong base preference such as Arg with G are distributed in a wide space around bases. Thus, amino acids with many different geometries can form a similar type of interaction with bases. The redundancy and structural flexibility in the interaction suggest that there are no simple rules in the sequence recognition, and its prediction is not straightforward. However, the spatial distributions of amino acids around bases indicate a possibility that the structural data can be used to derive empirical interaction potentials between amino acids and bases. Such information extracted from structural databases has been successfully used to predict amino acid sequences that fold into particular protein structures. We surmised that the structures of protein-DNA complexes could be used to predict DNA target sites for regulatory proteins, because determining DNA sequences that bind to a particular protein structure should be similar to finding amino acid sequences that fold into a particular structure. Here we demonstrate that the structural data can be used to predict DNA target sequences for regulatory proteins. Pairwise potentials that determine the interaction between bases and amino acids were empirically derived from the structural data. These potentials were then used to examine the compatibility between DNA sequences and the protein-DNA complex structure in a combinatorial "threading" procedure. We applied this strategy to the structures of protein-DNA complexes to predict DNA binding sites recognized by regulatory proteins. To test the applicability of this method in target-site prediction, we examined the effects of cognate and noncognate binding, cooperative binding, and DNA deformation on the binding specificity, and predicted binding sites in real promoters and compared with experimental data. These results show that target binding sites for several regulatory proteins are successfully predicted, and our data suggest that this method can serve as a powerful tool for predicting multiple target sites and target genes for regulatory proteins.     

Discovering structural correlations in alpha-helices.

We have developed a new representation for structural and functional motifs in protein sequences based on correlations between pairs of amino acids and applied it to alpha-helical and beta-sheet sequences. Existing probabilistic methods for representing and analyzing protein sequences have traditionally assumed conditional independence of evidence. In other words, amino acids are assumed to have no effect on each other. However, analyses of protein structures have repeatedly demonstrated the importance of interactions between amino acids in conferring both structure and function. Using Bayesian networks, we are able to model the relationships between amino acids at distinct positions in a protein sequence in addition to the amino acid distributions at each position. We have also developed an automated program for discovering sequence correlations using standard statistical tests and validation techniques. In this paper, we test this program on sequences from secondary structure motifs, namely alpha-helices and beta-sheets. In each case, the correlations our program discovers correspond well with known physical and chemical interactions between amino acids in structures. Furthermore, we show that, using different chemical alphabets for the amino acids, we discover structural relationships based on the same chemical principle used in constructing the alphabet. This new representation of 3-dimensional features in protein motifs, such as those arising from structural or functional constraints on the sequence, can be used to improve sequence analysis tools including pattern analysis and database search.     

Statistical measures on residue-level protein structural properties

The atomic-level structural properties of proteins, such as bond lengths, bond angles, and torsion angles, have been well studied and understood based on either chemistry knowledge or statistical analysis. Similar properties on the residue-level, such as the distances between two residues and the angles formed by short sequences of residues, can be equally important for structural analysis and modeling, but these have not been examined and documented on a similar scale. While these properties are difficult to measure experimentally, they can be statistically estimated in meaningful ways based on their distributions in known proteins structures. Residue-level structural properties including various types of residue distances and angles are estimated statistically. A software package is built to provide direct access to the statistical data for the properties including some important correlations not previously investigated. The distributions of residue distances and angles may vary with varying sequences, but in most cases, are concentrated in some high probability ranges, corresponding to their frequent occurrences in either α-helices or β-sheets. Strong correlations among neighboring residue angles, similar to those between neighboring torsion angles at the atomic-level, are revealed based on their statistical measures. Residue-level statistical potentials can be defined using the statistical distributions and correlations of the residue distances and angles. Ramachandran-like plots for strongly correlated residue angles are plotted and analyzed. Their applications to structural evaluation and refinement are demonstrated. With the increase in both number and quality of known protein structures, many structural properties can be derived from sets of protein structures by statistical analysis and data mining, and these can even be used as a supplement to the experimental data for structure determinations. Indeed, the statistical measures on various types of residue distances and angles provide more systematic and quantitative assessments on these properties, which can otherwise be estimated only individually and qualitatively. Their distributions and correlations in known protein structures show their importance for providing insights into how proteins may fold naturally to various residue-level structures.     

Steiner minimal trees, twist angles, and the protein folding problem

The Steiner Minimal Tree (SMT) problem determines the minimal length network for connecting a given set of vertices in three-dimensional space. SMTs have been shown to be useful in the geometric modeling and characterization of proteins. Even though the SMT problem is an NP-Hard Optimization problem, one can define planes within the amino acids that have a surprising regularity property for the twist angles of the planes. This angular property is quantified for all amino acids through the Steiner tree topology structure. The twist angle properties and other associated geometric properties unique for the remaining amino acids are documented in this paper. We also examine the relationship between the Steiner ratio rho and the torsion energy in amino acids with respect to the side chain torsion angle chi(1). The rho value is shown to be inversely proportional to the torsion energy. Hence, it should be a useful approximation to the potential energy function. Finally, the Steiner ratio is used to evaluate folded and misfolded protein structures. We examine all the native proteins and their decoys at http://dd.stanford.edu. and compare their Steiner ratio values. Because these decoy structures have been delicately misfolded, they look even more favorable than the native proteins from the potential energy viewpoint. However, the rho value of a decoy folded protein is shown to be much closer to the average value of an empirical Steiner ratio for each residue involved than that of the corresponding native one, so that we recognize the native folded structure more easily. The inverse relationship between the Steiner ratio and the energy level in the protein is shown to be a significant measure to distinguish native and decoy structures. These properties should be ultimately useful in the ab initio protein folding prediction.     

Predicting the secondary structure of globular proteins using neural network models

We present a new method for predicting the secondary structure of globular proteins based on non-linear neural network models. Network models learn from existing protein structures how to predict the secondary structure of local sequences of amino acids. The average success rate of our method on a testing set of proteins non-homologous with the corresponding training set was 64.3% on three types of secondary structure (alpha-helix, beta-sheet, and coil), with correlation coefficients of C alpha = 0.41, C beta = 0.31 and Ccoil = 0.41. These quality indices are all higher than those of previous methods. The prediction accuracy for the first 25 residues of the N-terminal sequence was significantly better. We conclude from computational experiments on real and artificial structures that no method based solely on local information in the protein sequence is likely to produce significantly better results for non-homologous proteins. The performance of our method of homologous proteins is much better than for non-homologous proteins, but is not as good as simply assuming that homologous sequences have identical structures.     

基于压缩氨基酸和支持向量机进行膜蛋白类型识别

膜蛋白是一类结构独特的蛋白质,是细胞执行各种功能的物质基础。根据其在细胞膜上的不同存在方式,主要分为六种类型。本文利用压缩的氨基酸对原始膜蛋白序列进行信息压缩,再对压缩序列进行氨基酸组成和顺序特征的提取,最后采用支持向量机构建分类模型。通过五叠交叉验证的结果表明,该方法对于六种膜蛋白的分类预测,准确度最高可达98％以上,平均预测准确度在85％以上,可有效实现膜蛋白六种类型的划分,为进一步分析膜蛋白的结构和功能奠定基础。     

Extracting Hydrogen-Bond Signature Patterns from Protein Structure Data

Classification of protein sequences and structures into families is a fundamental task in biology, and it is often used as a basis for designing experiments for gaining further knowledge. Some relationships between proteins are detected by the similarities in their sequences, and many more by the similarities in their structures. Despite this, there are a number of examples of functionally similar molecules without any recognisable sequence or structure similarities, and there are also a number of protein molecules that share common structural scaffolds but exhibit different functions. Newer methods of comparing molecules are required in order to detect similarities and dissimilarities in protein molecules. In this article, it is proposed that the precise 3-dimensional disposition of key residues in a protein molecule is what matters for its function, or what conveys the "meaning" for a biological system, but not what means it uses to achieve this. The concept of comparing two molecules through their intramolecular interaction networks is explored, since these networks dictate the disposition of amino acids in a protein structure. First, signature patterns, or fingerprints, of interaction networks in pre-classified protein structural families are computed using an approach to find structural equivalences and consensus hydrogen bonds. Five examples from different structural classes are illustrated. These patterns are then used to search the entire Protein Data Bank, an approach through which new, unexpected similarities have been found. The potential for finding relationships through this approach is highlighted. The use of hydrogen-bond fingerprints as a new metric for measuring similarities in protein structures is also described.     

Protein metal binding residue prediction based on neural networks

Over one-third of protein structures contain metal ions, which are the necessary elements in life systems. Traditionally, structural biologists were used to investigate properties of metalloproteins (proteins which bind with metal ions) by physical means and interpreting the function formation and reaction mechanism of enzyme by their structures and observations from experiments in vitro. Most of proteins have primary structures (amino acid sequence information) only; however, the 3-dimension structures are not always available. In this paper, a direct analysis method is proposed to predict the protein metal-binding amino acid residues from its sequence information only by neural networks with sliding window-based feature extraction and biological feature encoding techniques. In four major bulk elements (Calcium, Potassium, Magnesium, and Sodium), the metal-binding residues are identified by the proposed method with higher than 90% sensitivity and very good accuracy under 5-fold cross validation. With such promising results, it can be extended and used as a powerful methodology for metal-binding characterization from rapidly increasing protein sequences in the future.     

Bacterioopsin, haloopsin, and sensory opsin I of the halobacterial isolate Halobacterium sp. strain SG1: three new members of a growing family.

The genes coding for bacterioopsin, haloopsin, and sensory opsin I of a halobacterial isolate from the Red Sea called Halobacterium sp. strain SG1 have been cloned and sequenced. The deduced protein sequences were aligned to the previously known halobacterial retinal proteins. The addition of these new sequences lowered the number of conserved residues to only 23 amino acids, or 8% of the alignment. Data base searches with two highly conserved peptides as well as with an alignment profile yielded no significant similarity to any other protein, so the halobacterial retinal proteins should be regarded as a distinct protein family. The protein alignment was used to make predictions about the structure of the retinal proteins as well as about the amino acids in contact with retinal proteins. These results were in excellent agreement with the structural model of bacteriorhodopsin of Halobacterium halobium as well as with mutant studies, indicating that (i) structure predictions based on the sequences of a membrane protein family can be quite accurate; (ii) halorhodopsin and sensory rhodopsin I have tertiary structures similar to that of bacteriorhodopsin; (iii) conserved amino acids do not take part in reactions specific for one group of proteins, e.g., proton translocation for bacteriorhodopsins, but have a crucial role in determining the conformation and reactions of the chromophore; and (iv) the general mode of action (light-induced chromophore and protein movements) is the same for all halobacterial retinal proteins, ion pumps as well as sensors.