A dynamically growing self-organizing tree (DGSOT) for hierarchical clustering gene expression profiles

MOTIVATION: The increasing use of microarray technologies is generating large amounts of data that must be processed in order to extract useful and rational fundamental patterns of gene expression. Hierarchical clustering technology is one method used to analyze gene expression data, but traditional hierarchical clustering algorithms suffer from several drawbacks (e.g. fixed topology structure; mis-clustered data which cannot be reevaluated). In this paper, we introduce a new hierarchical clustering algorithm that overcomes some of these drawbacks. RESULT: We propose a new tree-structure self-organizing neural network, called dynamically growing self-organizing tree (DGSOT) algorithm for hierarchical clustering. The DGSOT constructs a hierarchy from top to bottom by division. At each hierarchical level, the DGSOT optimizes the number of clusters, from which the proper hierarchical structure of the underlying dataset can be found. In addition, we propose a new cluster validation criterion based on the geometric property of the Voronoi partition of the dataset in order to find the proper number of clusters at each hierarchical level. This criterion uses the Minimum Spanning Tree (MST) concept of graph theory and is computationally inexpensive for large datasets. A K-level up distribution (KLD) mechanism, which increases the scope of data distribution in the hierarchy construction, was used to improve the clustering accuracy. The KLD mechanism allows the data misclustered in the early stages to be reevaluated at a later stage and increases the accuracy of the final clustering result. The clustering result of the DGSOT is easily displayed as a dendrogram for visualization. Based on a yeast cell cycle microarray expression dataset, we found that our algorithm extracts gene expression patterns at different levels. Furthermore, the biological functionality enrichment in the clusters is considerably high and the hierarchical structure of the clusters is more reasonable. AVAILABILITY: DGSOT is available upon request from the authors.     

Prediction of protein stability changes for single-site mutations using support vector machines

Accurate prediction of protein stability changes resulting from single amino acid mutations is important for understanding protein structures and designing new proteins. We use support vector machines to predict protein stability changes for single amino acid mutations leveraging both sequence and structural information. We evaluate our approach using cross-validation methods on a large dataset of single amino acid mutations. When only the sign of the stability changes is considered, the predictive method achieves 84% accuracy-a significant improvement over previously published results. Moreover, the experimental results show that the prediction accuracy obtained using sequence alone is close to the accuracy obtained using tertiary structure information. Because our method can accurately predict protein stability changes using primary sequence information only, it is applicable to many situations where the tertiary structure is unknown, overcoming a major limitation of previous methods which require tertiary information. The web server for predictions of protein stability changes upon mutations (MUpro), software, and datasets are available at http://www.igb.uci.edu/servers/servers.html.     

ANGLE: a sequencing errors resistant program for predicting protein coding regions in unfinished cDNA

In the process of making full-length cDNA, predicting protein coding regions helps both in the preliminary analysis of genes and in any succeeding process. However, unfinished cDNA contains artifacts including many sequencing errors, which hinder the correct evaluation of coding sequences. Especially, predictions of short sequences are difficult because they provide little information for evaluating coding potential. In this paper, we describe ANGLE, a new program for predicting coding sequences in low quality cDNA. To achieve error-tolerant prediction, ANGLE uses a machine-learning approach, which makes better expression of coding sequence maximizing the use of limited information from input sequences. Our method utilizes not only codon usage, but also protein structure information which is difficult to be used for stochastic model-based algorithms, and optimizes limited information from a short segment when deciding coding potential, with the result that predictive accuracy does not depend on the length of an input sequence. The performance of ANGLE is compared with ESTSCAN on four dataset each of them having a different error rate (one frame-shift error or one substitution error per 200-500 nucleotides) and on one dataset which has no error. ANGLE outperforms ESTSCAN by 9.26% in average Matthews's correlation coefficient on short sequence dataset (< 1000 bases). On long sequence dataset, ANGLE achieves comparable performance.     

Is it better to combine predictions?

We have compared the accuracy of the individual protein secondary structure prediction methods: PHD, DSC, NNSSP and Predator against the accuracy obtained by combing the predictions of the methods. A range of ways of combing predictions were tested: voting, biased voting, linear discrimination, neural networks and decision trees. The combined methods that involve 'learning' (the non-voting methods) were trained using a set of 496 non-homologous domains; this dataset was biased as some of the secondary structure prediction methods had used them for training. We used two independent test sets to compare predictions: the first consisted of 17 non-homologous domains from CASP3 (Third Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction); the second set consisted of 405 domains that were selected in the same way as the training set, and were non-homologous to each other and the training set. On both test datasets the most accurate individual method was NNSSP, then PHD, DSC and the least accurate was Predator; however, it was not possible to conclusively show a significant difference between the individual methods. Comparing the accuracy of the single methods with that obtained by combing predictions it was found that it was better to use a combination of predictions. On both test datasets it was possible to obtain a approximately 3% improvement in accuracy by combing predictions. In most cases the combined methods were statistically significantly better (at P = 0.05 on the CASP3 test set, and P = 0.01 on the EBI test set). On the CASP3 test dataset there was no significant difference in accuracy between any of the combined method of prediction: on the EBI test dataset, linear discrimination and neural networks significantly outperformed voting techniques. We conclude that it is better to combine predictions.     

SPSSM8: An accurate approach for predicting eight-state secondary structures of proteins

Protein eight-state secondary structure prediction is challenging, but is necessary to determine protein structure and function. Here, we report the development of a novel approach, SPSSM8, to predict eight-state secondary structures of proteins accurately from sequences based on the structural position-specific scoring matrix (SPSSM). The SPSSM has been successfully utilized to predict three-state secondary structures. Now we employ an eight-state SPSSM as a feature that is obtained from sequence structure alignment against a large database of 9 million sequences with putative structural information. The SPSSM8 uses a low sequence identity dataset (9062 entries) as a training set and conditional random field for the classification algorithm. The SPSSM8 achieved an average eight-state secondary structure accuracy (Q8) of 71.7% (Q3, 81.6%) for an independent testing set (463 entries), which had an improved accuracy of 10.1% and 4.6% compared with SSPro8 and CNF, respectively, and significantly improved the accuracy of eight-state secondary structure prediction. For CASP 9 dataset (92 entries) the SPSSM8 achieved a Q8 accuracy of 80.1% (Q3, 83.0%). The SPSSM8 was confirmed as an outstanding predictor for eight-state secondary structures of proteins. SPSSM8 is freely available at http://cal.tongji.edu.cn/SPSSM8.     

Predicting membrane protein types using residue-pair models based on reduced similarity dataset

An algorithm to predict the membrane protein types based on the multi-residue-pair effect in the Markov model is proposed. For a newly constructed dataset of 835 membrane proteins with very low sequence similarity, the overall prediction accuracy has been achieved as high as 81.1% and 71.7% in the resubstitution and jackknife test, respectively, for a prediction of type I single-pass, type II single-pass, multi-pass membrane proteins, lipid chain-anchored and GPI-anchored membrane proteins. The improvement of about 11% in the jackknife test can be achieved compared with the component-coupled algorithm merely based on the amino acid composition (AAC approach). The improvement is also confirmed on a high similarity dataset and the other extrapolating test. The result implies that designing more incisive analysis tools, one should develop algorithms based on the representative dataset with lower sequence similarity. The present algorithm is useful to expedite the determination of the types and functions of new membrane proteins and may be useful for the systematic analysis of functional genome data in a large scale. The computer program is available on request.     

Cascaded deep network systems with linked ensemble components for underwater fish detection in the wild

We propose a fish detection system based on deep network architectures to robustly detect and count fish objects under a variety of benthic background and illumination conditions. The algorithm consists of an ensemble of Region-based Convolutional Neural Networks that are linked in a cascade structure by Long Short-Term Memory networks. The proposed network is efficiently trained as all components are jointly trained by backpropagation. We train and test our system for a dataset of 18 videos taken in the wild. In our dataset, there are around 20 to 100 fish objects per frame with many fish objects having small pixel areas (less than 900 square pixels). From a series of experiments and ablation tests, the proposed system preserves detection accuracy despite multi-scale distortions, cropping and varying background environments. We present analysis that shows how object localization accuracy is increased by an automatic correction mechanism in the deep network's cascaded ensemble structure. The correction mechanism rectifies any errors in the predictions as information progresses through the network cascade. Our findings in this experiment regarding ensemble system architectures can be generalized to other object detection applications.     

多特征融合的植物长链非编码RNA的预测

长链非编码RNA(Long non-coding RNA, lncRNA)是一类被定义为转录本的长度大于200 nt、没有蛋白编码能力的RNA转录本。研究表明,lncRNA在调节植物生长发育、表观遗传反应以及各种胁迫反应中起重要作用。但是与人类和动物相比,植物lncRNA的研究仍然处于起步阶段。目前,如何从大量的转录本中准确地挑选出lncRNA仍然是植物lncRNA研究领域的重要问题之一。本文构建了新的植物lncRNA和mRNA数据集,分析了数据集中植物lncRNA的序列及结构特征,提取了序列的k-mer频数信息、二级结构信息、开放阅读框信息以及序列的几何柔性等特征,基于SVM(Support Vector Machine, SVM)算法,用Jackknife检验对植物lncRNA进行了预测,并且计算了各种特征融合后对植物lncRNA预测结果的影响,准确率达到了96.14%。     

Armadillo: domain boundary prediction by amino acid composition

The identification and annotation of protein domains provides a critical step in the accurate determination of molecular function. Both computational and experimental methods of protein structure determination may be deterred by large multi-domain proteins or flexible linker regions. Knowledge of domains and their boundaries may reduce the experimental cost of protein structure determination by allowing researchers to work on a set of smaller and possibly more successful alternatives. Current domain prediction methods often rely on sequence similarity to conserved domains and as such are poorly suited to detect domain structure in poorly conserved or orphan proteins. We present here a simple computational method to identify protein domain linkers and their boundaries from sequence information alone. Our domain predictor, Armadillo (http://armadillo.blueprint.org), uses any amino acid index to convert a protein sequence to a smoothed numeric profile from which domains and domain boundaries may be predicted. We derived an amino acid index called the domain linker propensity index (DLI) from the amino acid composition of domain linkers using a non-redundant structure dataset. The index indicates that Pro and Gly show a propensity for linker residues while small hydrophobic residues do not. Armadillo predicts domain linker boundaries from Z-score distributions and obtains 35% sensitivity with DLI in a two-domain, single-linker dataset (within +/-20 residues from linker). The combination of DLI and an entropy-based amino acid index increases the overall Armadillo sensitivity to 56% for two domain proteins. Moreover, Armadillo achieves 37% sensitivity for multi-domain proteins, surpassing most other prediction methods. Armadillo provides a simple, but effective method by which prediction of domain boundaries can be obtained with reasonable sensitivity. Armadillo should prove to be a valuable tool for rapidly delineating protein domains in poorly conserved proteins or those with no sequence neighbors. As a first-line predictor, domain meta-predictors could yield improved results with Armadillo predictions.     

Prediction of protein structural class by amino acid and polypeptide composition.

A new approach of predicting structural classes of protein domain sequences is presented in this paper. Besides the amino acid composition, the composition of several dipeptides, tripeptides, tetrapeptides, pentapeptides and hexapeptides are taken into account based on the stepwise discriminant analysis. The result of jackknife test shows that this new approach can lead to higher predictive sensitivity and specificity for reduced sequence similarity datasets. Considering the dataset PDB40-B constructed by Brenner and colleagues, 75.2% protein domain sequences are correctly assigned in the jackknife test for the four structural classes: all-alpha, all-beta, alpha/beta and alpha + beta, which is improved by 19.4% in jackknife test and 25.5% in resubstitution test, in contrast with the component-coupled algorithm using amino acid composition alone (AAC approach) for the same dataset. In the cross-validation test with dataset PDB40-J constructed by Park and colleagues, more than 80% predictive accuracy is obtained. Furthermore, for the dataset constructed by Chou and Maggiona, the accuracy of 100% and 99.7% can be easily achieved, respectively, in the resubstitution test and in the jackknife test merely taking the composition of dipeptides into account. Therefore, this new method provides an effective tool to extract valuable information from protein sequences, which can be used for the systematic analysis of small or medium size protein sequences. The computer programs used in this paper are available on request.     

Identification of helix capping and β-turn motifs from NMR chemical shifts

We present an empirical method for identification of distinct structural motifs in proteins on the basis of experimentally determined backbone and ¹³C^β chemical shifts. Elements identified include the N-terminal and C-terminal helix capping motifs and five types of β-turns: I, II, I′, II′ and VIII. Using a database of proteins of known structure, the NMR chemical shifts, together with the PDB-extracted amino acid preference of the helix capping and β-turn motifs are used as input data for training an artificial neural network algorithm, which outputs the statistical probability of finding each motif at any given position in the protein. The trained neural networks, contained in the MICS (motif identification from chemical shifts) program, also provide a confidence level for each of their predictions, and values ranging from ca 0.7–0.9 for the Matthews correlation coefficient of its predictions far exceed those attainable by sequence analysis. MICS is anticipated to be useful both in the conventional NMR structure determination process and for enhancing on-going efforts to determine protein structures solely on the basis of chemical shift information, where it can aid in identifying protein database fragments suitable for use in building such structures.     

A combined long-range phasing and long haplotype imputation method to impute phase for SNP genotypes

Background

Knowing the phase of marker genotype data can be useful in genome-wide association studies, because it makes it possible to use analysis frameworks that account for identity by descent or parent of origin of alleles and it can lead to a large increase in data quantities via genotype or sequence imputation. Long-range phasing and haplotype library imputation constitute a fast and accurate method to impute phase for SNP data.

Methods

A long-range phasing and haplotype library imputation algorithm was developed. It combines information from surrogate parents and long haplotypes to resolve phase in a manner that is not dependent on the family structure of a dataset or on the presence of pedigree information.

Results

The algorithm performed well in both simulated and real livestock and human datasets in terms of both phasing accuracy and computation efficiency. The percentage of alleles that could be phased in both simulated and real datasets of varying size generally exceeded 98% while the percentage of alleles incorrectly phased in simulated data was generally less than 0.5%. The accuracy of phasing was affected by dataset size, with lower accuracy for dataset sizes less than 1000, but was not affected by effective population size, family data structure, presence or absence of pedigree information, and SNP density. The method was computationally fast. In comparison to a commonly used statistical method (fastPHASE), the current method made about 8% less phasing mistakes and ran about 26 times faster for a small dataset. For larger datasets, the differences in computational time are expected to be even greater. A computer program implementing these methods has been made available.

Conclusions

The algorithm and software developed in this study make feasible the routine phasing of high-density SNP chips in large datasets.     

Statistical prediction of protein chemical interactions based on chemical structure and mass spectrometry data

MOTIVATION: Prediction of interactions between proteins and chemical compounds is of great benefit in drug discovery processes. In this field, 3D structure-based methods such as docking analysis have been developed. However, the genomewide application of these methods is not really feasible as 3D structural information is limited in availability. RESULTS: We describe a novel method for predicting protein-chemical interaction using SVM. We utilize very general protein data, i.e. amino acid sequences, and combine these with chemical structures and mass spectrometry (MS) data. MS data can be of great use in finding new chemical compounds in the future. We assessed the validity of our method in the dataset of the binding of existing drugs and found that more than 80% accuracy could be obtained. Furthermore, we conducted comprehensive target protein predictions for MDMA, and validated the biological significance of our method by successfully finding proteins relevant to its known functions. AVAILABILITY: Available on request from the authors.     

PCA-correlated SNPs for structure identification in worldwide human populations

Existing methods to ascertain small sets of markers for the identification of human population structure require prior knowledge of individual ancestry. Based on Principal Components Analysis (PCA), and recent results in theoretical computer science, we present a novel algorithm that, applied on genomewide data, selects small subsets of SNPs (PCA-correlated SNPs) to reproduce the structure found by PCA on the complete dataset, without use of ancestry information. Evaluating our method on a previously described dataset (10,805 SNPs, 11 populations), we demonstrate that a very small set of PCA-correlated SNPs can be effectively employed to assign individuals to particular continents or populations, using a simple clustering algorithm. We validate our methods on the HapMap populations and achieve perfect intercontinental differentiation with 14 PCA-correlated SNPs. The Chinese and Japanese populations can be easily differentiated using less than 100 PCA-correlated SNPs ascertained after evaluating 1.7 million SNPs from HapMap. We show that, in general, structure informative SNPs are not portable across geographic regions. However, we manage to identify a general set of 50 PCA-correlated SNPs that effectively assigns individuals to one of nine different populations. Compared to analysis with the measure of informativeness, our methods, although unsupervised, achieved similar results. We proceed to demonstrate that our algorithm can be effectively used for the analysis of admixed populations without having to trace the origin of individuals. Analyzing a Puerto Rican dataset (192 individuals, 7,257 SNPs), we show that PCA-correlated SNPs can be used to successfully predict structure and ancestry proportions. We subsequently validate these SNPs for structure identification in an independent Puerto Rican dataset. The algorithm that we introduce runs in seconds and can be easily applied on large genome-wide datasets, facilitating the identification of population substructure, stratification assessment in multi-stage whole-genome association studies, and the study of demographic history in human populations.     

Consensus Data Mining (CDM) Protein Secondary Structure Prediction Server: combining GOR V and Fragment Database Mining (FDM)

One of the challenges in protein secondary structure prediction is to overcome the cross-validated 80% prediction accuracy barrier. Here, we propose a novel approach to surpass this barrier. Instead of using a single algorithm that relies on a limited data set for training, we combine two complementary methods having different strengths: Fragment Database Mining (FDM) and GOR V. FDM harnesses the availability of the known protein structures in the Protein Data Bank and provides highly accurate secondary structure predictions when sequentially similar structural fragments are identified. In contrast, the GOR V algorithm is based on information theory, Bayesian statistics, and PSI-BLAST multiple sequence alignments to predict the secondary structure of residues inside a sliding window along a protein chain. A combination of these two different methods benefits from the large number of structures in the PDB and significantly improves the secondary structure prediction accuracy, resulting in Q3 ranging from 67.5 to 93.2%, depending on the availability of highly similar fragments in the Protein Data Bank.     

A Generalizable Brain-Computer Interface (BCI) Using Machine Learning for Feature Discovery

This work describes a generalized method for classifying motor-related neural signals for a brain-computer interface (BCI), based on a stochastic machine learning method. The method differs from the various feature extraction and selection techniques employed in many other BCI systems. The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge. Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure. The results showed that the average performance of the new algorithm outperformed other published methods using the Berlin BCI IV (2008) competition dataset and was comparable to the best results in the Berlin BCI II (2002–3) competition dataset. The new method was also applied to electroencephalography (EEG) data recorded from five subjects undertaking a hand squeeze task and demonstrated high levels of accuracy with a mean classification accuracy of 78.9% after five-fold cross-validation. Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.     

Identification and application of the concepts important for accurate and reliable protein secondary structure prediction.

A protein secondary structure prediction method from multiply aligned homologous sequences is presented with an overall per residue three-state accuracy of 70.1%. There are two aims: to obtain high accuracy by identification of a set of concepts important for prediction followed by use of linear statistics; and to provide insight into the folding process. The important concepts in secondary structure prediction are identified as: residue conformational propensities, sequence edge effects, moments of hydrophobicity, position of insertions and deletions in aligned homologous sequence, moments of conservation, auto-correlation, residue ratios, secondary structure feedback effects, and filtering. Explicit use of edge effects, moments of conservation, and auto-correlation are new to this paper. The relative importance of the concepts used in prediction was analyzed by stepwise addition of information and examination of weights in the discrimination function. The simple and explicit structure of the prediction allows the method to be reimplemented easily. The accuracy of a prediction is predictable a priori. This permits evaluation of the utility of the prediction: 10% of the chains predicted were identified correctly as having a mean accuracy of > 80%. Existing high-accuracy prediction methods are "black-box" predictors based on complex nonlinear statistics (e.g., neural networks in PHD: Rost & Sander, 1993a). For medium- to short-length chains (> or = 90 residues and < 170 residues), the prediction method is significantly more accurate (P < 0.01) than the PHD algorithm (probably the most commonly used algorithm). In combination with the PHD, an algorithm is formed that is significantly more accurate than either method, with an estimated overall three-state accuracy of 72.4%, the highest accuracy reported for any prediction method.     

The recognition of multi-class protein folds by adding average chemical shifts of secondary structure elements

The recognition of protein folds is an important step in the prediction of protein structure and function. Recently, an increasing number of researchers have sought to improve the methods for protein fold recognition. Following the construction of a dataset consisting of 27 protein fold classes by Ding and Dubchak in 2001, prediction algorithms, parameters and the construction of new datasets have improved for the prediction of protein folds. In this study, we reorganized a dataset consisting of 76-fold classes constructed by Liu et al. and used the values of the increment of diversity, average chemical shifts of secondary structure elements and secondary structure motifs as feature parameters in the recognition of multi-class protein folds. With the combined feature vector as the input parameter for the Random Forests algorithm and ensemble classification strategy, we propose a novel method to identify the 76 protein fold classes. The overall accuracy of the test dataset using an independent test was 66.69%; when the training and test sets were combined, with 5-fold cross-validation, the overall accuracy was 73.43%. This method was further used to predict the test dataset and the corresponding structural classification of the first 27-protein fold class dataset, resulting in overall accuracies of 79.66% and 93.40%, respectively. Moreover, when the training set and test sets were combined, the accuracy using 5-fold cross-validation was 81.21%. Additionally, this approach resulted in improved prediction results using the 27-protein fold class dataset constructed by Ding and Dubchak.     

Validating annotations for uncharacterized proteins in Shewanella oneidensis

Proteins of unknown function are a barrier to our understanding of molecular biology. Assigning function to these "uncharacterized" proteins is imperative, but challenging. The usual approach is similarity searches using annotation databases, which are useful for predicting function. However, since the performance of these databases on uncharacterized proteins is basically unknown, the accuracy of their predictions is suspect, making annotation difficult. To address this challenge, we developed a benchmark annotation dataset of 30 proteins in Shewanella oneidensis. The proteins in the dataset were originally uncharacterized after the initial annotation of the S. oneidensis proteome in 2002. In the intervening 5 years, the accumulation of new experimental evidence has enabled specific functions to be predicted. We utilized this benchmark dataset to evaluate several commonly utilized annotation databases. According to our criteria, six annotation databases accurately predicted functions for at least 60% of proteins in our dataset. Two of these six even had a "conditional accuracy" of 90%. Conditional accuracy is another evaluation metric we developed which excludes results from databases where no function was predicted. Also, 27 of the 30 proteins' functions were correctly predicted by at least one database. These represent one of the first performance evaluations of annotation databases on uncharacterized proteins. Our evaluation indicates that these databases readily incorporate new information and are accurate in predicting functions for uncharacterized proteins, provided that experimental function evidence exists.