Microarray technology allows simultaneous measurements of expression levels for thousands of genes. An important aspect of microarray studies includes the prediction of patient survival based on their gene expression profile. This naturally calls for the use of a dimension reduction procedure together with the survival prediction model. In this study, a new method based on wavelet transform for survival-relevant gene selection is presented. Cox proportional hazard model is typically used to build prediction model for patients' survival using the selected genes. The prediction model will be evaluated with the R2, concordance index, likelihood ratio statistic and Akaike information criteria. The results proved that good performance of survival prediction is achieved based on the selected genes. The results suggested the possibility of developing more advanced tools based on wavelets for gene selection from microarray data sets in the context of survival analysis.

The existence of confounders such as population structure in genome-wide association study makes it difficult to apply machine learning methods directly to solve biological problems. It is still unclear how to effectively correct confounders. In this work, we propose an Orthogonal Projection Correction (OPC) method to correct confounders. This is achieved by orthogonally decomposing each feature to a confounding component and a non-confounding component, such that the original data can be best reconstructed by only the non-confounding components of features. The confounder space is built based on prior knowledge, and each feature is projected to its orthogonal complement space. This OPC procedure is shown to be kernelisable. We then propose a ProSVM method by integrating the OPC method and support vector machine for classification. In the experiments, our OPC method for confounder correction improves the tumour diagnosis based on samples from different labs and phenotype prediction in the presence of population structure.

The MCR-Miner algorithm is aimed to mine all maximal high confident association rules form the microarray up/down-expressed genes data set. This paper introduces two new algorithms: IMCR-Miner and PMCR-Miner. The IMCR-Miner algorithm is an extension of the MCR-Miner algorithm with some improvements. These improvements implement a novel way to store the samples of each gene into a list of unsigned integers in order to benefit using the bitwise operations. In addition, the IMCR-Miner algorithm overcomes the drawbacks faced by the MCR-Miner algorithm by setting some restrictions to ignore repeated comparisons. The PMCR-Miner algorithm is a parallel version of the new proposed IMCR-Miner algorithm. The PMCR-Miner algorithm is based on shared-memory systems and task parallelism, where no time is needed in the process of sharing and combining data between processors. The experimental results on real microarray data sets show that the PMCR-Miner algorithm is more efficient and scalable than the counterparts.

Prediction of different classes within the human leukocyte antigen (HLA) gene family can provide insight into the human immune system and its response to viral pathogens. Therefore, it is desirable to develop an efficient and easily interpretable method for predicting HLA gene class compared to existing methods. We investigated the HLA gene prediction problem as follows: (a) establishing a dataset (HLA262) such that the sequence identity of the complete HLA dataset was reduced to 30%; (b) proposing a feature set of informative physicochemical properties that cooperate with SVM (named HLAPred) to achieve high accuracy and sensitivity (90.04% and 82.99%, respectively) compared with existing methods; and (c) analysing the informative physicochemical properties to understand the physicochemical properties and molecular mechanisms of the HLA gene family.

In this paper, a hybrid approach based on clustering and Particle Swarm Optimisation (PSO) is proposed to perform gene selection and classification for microarray data. In the new method, firstly, genes are partitioned into a predetermined number of clusters by K-means method. Since the genes in each cluster have much redundancy, Max-Relevance Min-Redundancy (mRMR) strategy is used to reduce redundancy of the clustered genes. Then, PSO is used to perform further gene selection from the remaining clustered genes. Because of its better generalisation performance with much faster convergence rate than other learning algorithms for neural networks, Extreme Learning Machine (ELM) is chosen to evaluate candidate gene subsets selected by PSO and perform samples classification in this study. The proposed method selects less redundant genes as well as increases prediction accuracy and its efficiency and effectiveness are verified by extensive comparisons with other classical methods on three open microarray data.

Identification of differential gene regulators with significant changes under disparate conditions is essential to understand complex biological mechanism in a disease. Differential Network Analysis (DiNA) examines different biological processes based on gene regulatory networks that represent regulatory interactions between genes with a graph model. While most studies in DiNA have considered correlation-based inference to construct gene regulatory networks from gene expression data due to its intuitive representation and simple implementation, the approach lacks in the representation of causal effects and multivariate effects between genes. In this paper, we propose an approach named Differential Gene Regulatory Network (DiffGRN) that infers differential gene regulation between two groups. We infer gene regulatory networks of two groups using Random LASSO, and then we identify differential gene regulations by the proposed significance test. The advantages of DiffGRN are to capture multivariate effects of genes that regulate a gene simultaneously, to identify causality of gene regulations, and to discover differential gene regulators between regression-based gene regulatory networks. We assessed DiffGRN by simulation experiments and showed its outstanding performance than the current state-of-the-art correlation-based method, DINGO. DiffGRN is applied to gene expression data in asthma. The DiNA with asthma data showed a number of gene regulations, such as ADAM12 and RELB, reported in biological literature.

Rsp5, E3 ligases conserved from yeast to mammals, plays a key role in diverse processes in yeast. However, many of Rsp5 substrates are still unclear. Therefore we proposed an in silico method to recognise new substrates of Rsp5. To investigate the molecular determinants that affect the interaction between Rsp5 and its substrate, we have systematically analysed many features that perhaps correlated with the Rsp5 substrate recognition. It is found that PPxY motif, transmembrane region, disorder region and N-linked glycosylation modification are the most important features for substrate recognition. We have constructed an SVM-based classifier to recognise Rsp5 substrates, obtaining 81.5% sensitivity and 74.1% specificity averagely on ten independent testing dataset. We also applied the model on the whole yeast proteome, and identified -66 new Rsp5 substrates. Functional annotation reveals that half of these novel substrates function in the Rsp5 involved cell processes as Rsp5-interacting proteins.

Analysing and classifying sequences based on similarities and differences is a mathematical problem of escalating relevance and importance in many scientific disciplines. One of the primary challenges in applying machine learning algorithms to sequential data, such as biological sequences, is the extraction and representation of significant features from the data. To address this problem, we have recently developed a representation, entitled Multi-Layered Vector Spaces (MLVS), which is a simple mathematical model that maps sequences into a set of MLVS. We demonstrate the usefulness of the model by applying it to the problem of identifying signal peptides. MLVS feature vectors are generated from a collection of protein sequences and the resulting vectors are used to create support vector machine classifiers. Experiments show that the MLVS-based classifiers are able to outperform or perform on par with several existing methods that are specifically designed for the purpose of identifying signal peptides.

In this paper, an algorithm is proposed that detects the existence of a common ancestor gene sequence for non-overlapping transposition metric given two input DNA sequences. We consider two cases: fixed length transposition and all length transposition. For the first one, the algorithm has the time complexity of O(n3), where n is the length of input sequences. In case of all length transposition, theoretical worst case time complexity of the algorithm is proven to be O(n4). However, practically the worst case and the average case time complexity for all length transposition are found to be O(n3) and O(n2) respectively. This work is motivated by the purpose of diagnosing unknown genetic disease that shows allelic heterogeneity, a case where a normal gene mutates in different orders resulting in two different gene sequences causing two different genetic diseases. The algorithm can be useful as well in the study of breed-related hereditary to determine the genetic spread of a defective gene in the population.

In silico prediction of novel miRNAs from genomic sequences remains a challenging problem. This study presents a genome-wide miRNA discovery software package called GenoScan and evaluates two hairpin classification methods. These methods, one ensemble-based and one using logistic regression were benchmarked along with 15 published methods. In addition, the sequence-folding step is addressed by investigating the impact of secondary structure prediction methods and the choice of input sequence length on prediction performance. Both the accuracy of secondary structure predictions and the miRNA prediction are evaluated. In the benchmark of hairpin classification methods, the regression model achieved highest classification accuracy. Of the structure prediction methods evaluated, ContextFold achieved the highest agreement between predicted and experimentally determined structures. However, both the choice of secondary structure prediction method and input sequence length had limited impact on hairpin classification performance.

In this paper, we develop a novel regularisation method for MVAR via weighted fusion which considers the correlation among variables. In theory, we discuss the grouping effect of weighted fusion regularisation for linear models. By virtue of the probability method, we show that coefficients corresponding to highly correlated predictors have small differences. A quantitative estimate for such small differences is given regardless of the coefficients signs. The estimate is also improved when consider empirical approximation error if the model fit the data well. We then apply the proposed model on several time series data sets especially a time series dataset of human gut microbiomes. The experimental results indicate that the new approach has better performance than several other VAR-based models and we also demonstrate its capability of extracting relevant microbial interactions.

The scarcity of large labelled datasets comprising clinical text that can be exploited within the paradigm of supervised machine learning creates barriers for the secondary use of data from electronic health records. It is therefore important to develop capabilities to leverage the large amounts of unlabelled data that, indeed, tend to be readily available. One technique utilises distributional semantics to create word representations in a wholly unsupervised manner and uses existing training data to learn prototypical representations of predefined semantic categories. Features describing whether a given word belongs to a certain category are then provided to the learning algorithm. It has been shown that using multiple distributional semantic models, each employing a different word order strategy, can lead to enhanced predictive performance. Here, another hyperparameter is also varied--the size of the context window--and an experimental investigation shows that this leads to further performance gains.

Identification of non-coding RNAs (ncRNAs) has been significantly improved over the past decade. On the other hand, semantic annotation of ncRNA data is facing critical challenges due to the lack of a comprehensive ontology to serve as common data elements and data exchange standards in the field. We developed the Non-Coding RNA Ontology (NCRO) to handle this situation. By providing a formally defined ncRNA controlled vocabulary, the NCRO aims to fill a specific and highly needed niche in semantic annotation of large amounts of ncRNA biological and clinical data.

MicroRNAs (miRNAs) are a class of short RNA molecules that regulate gene expression by binding directly to messenger RNAs. Conventional approaches to miRNA target prediction estimate the accessibility of target sites and the strength of the binding miRNA by finding optimums of some energy models, which involves O(n3) computations. Alternatively, we narrow down potential binding sites of miRNAs to suboptimal hits of a pairwise alignment algorithm called Fitting Alignment in O(n2). We invoke a same algorithm, once for all candidate sites to measure the site accessibilities. These features are applied to a binary classifier being learned to predict true associations between miRNAs and target genes. Training the classifier requires the negative samples indicating non-affected genes. The experiments verifying such negative associations have been rarely performed, so we exploit tissue-specific gene expression data to impute the negative associations. The recall rate of our method is above 70% (at precision 85%).

Knowledge of metabolic processes is collected in easily accessable online databases which are increasing rapidly in content and detail. Using these databases for the automatic construction of metabolic network models requires high accuracy and consistency. In this bipartite study we evaluate current accuracy and consistency problems using the KEGG database as a prominent example and propose design principles for dealing with such problems. In the first half, we present our computational approach for classifying inconsistencies and provide an overview of the classes of inconsistencies we identified. We detected inconsistencies both for database entries referring to substances and entries referring to reactions. In the second part, we present strategies to deal with the detected problem classes. We especially propose a rule-based database approach which allows for the inclusion of parameterised molecular species and parameterised reactions. Detailed case-studies and a comparison of explicit networks from KEGG with their anticipated rule-based representation underline the applicability and scalability of this approach.

The computational framework used the traditional similarity measures to find out the significant relationships in biological annotations. But its prerequisites that the biological annotations do not cooccur with each other is particular. To overcome it, in this paper a new method Improved Algorithm for Maximal Information Coefficient (IAMIC) is suggested to discover the hidden regularities between biological annotations. IAMIC approximates a novel similarity coefficient on maximal information coefficient with generality and equitability, by bettering axis partition through quadratic optimisation instead of violence search. The experimental results show that IAMIC is more appropriate for identifying the associations between biological annotations, and further extracting the novel associations hidden in collected data sets than other similarity measures.

In the last decade, biology and medicine have undergone a fundamental change: next generation sequencing (NGS) technologies have enabled to obtain genomic sequences very quickly and at small costs compared to the traditional Sanger method. These NGS technologies have thus permitted to collect genomic sequences (genes, exomes or even full genomes) of individuals of the same species. These latter sequences are identical to more than 99%. There is thus a strong need for efficient algorithms for indexing and performing fast pattern matching in such specific sets of sequences. In this paper we propose a very efficient algorithm that solves the exact pattern matching problem in a set of highly similar DNA sequences where only the pattern can be pre-processed. This new algorithm extends variants of the Boyer-Moore exact string matching algorithm. Experimental results show that it exhibits the best performances in practice.

Many clustering methods have been developed to identify functional modules in Protein-Protein Interaction (PPI) networks but the results are far from satisfaction. To overcome the noise and incomplete problems of PPI networks and find more accurate and stable functional modules, we propose an integrative method, bipartite graph-based Non-negative Matrix Factorisation method (BiNMF), in which we adopt multiple biological data sources as different views that describe PPIs. Specifically, traditional clustering models are adopted as preliminary analysis of different views of protein functional similarity. Then the intermediate clustering results are represented by a bipartite graph which can comprehensively represent the relationships between proteins and intermediate clusters and finally overlapping clustering results are achieved. Through extensive experiments, we see that our method is superior to baseline methods and detailed analysis has demonstrated the benefits of integrating diverse clustering methods and multiple biological information sources.

Cuckoo Search (CS) optimisation algorithm is used for feature selection in cancer classification using microarray gene expression data. Since the gene expression data has thousands of genes and a small number of samples, feature selection methods can be used for the selection of informative genes to improve the classification accuracy. Initially, the genes are ranked based on T-statistics, Signal-to-Noise Ratio (SNR) and F-statistics values. The CS is used to find the informative genes from the top-m ranked genes. The classification accuracy of k-Nearest Neighbour (kNN) technique is used as the fitness function for CS. The proposed method is experimented and analysed with ten different cancer gene expression datasets. The results show that the CS gives 100% average accuracy for DLBCL Harvard, Lung Michigan, Ovarian Cancer, AML-ALL and Lung Harvard2 datasets and it outperforms the existing techniques in DLBCL outcome and prostate datasets.

High-dimensional data and a large number of redundancy features in bioinformatics research have created an urgent need for feature selection. In this paper, a novel random forests-based feature selection method is proposed that adopts the idea of stratifying feature space and combines generalised sequence backward searching and generalised sequence forward searching strategies. A random forest variable importance score is used to rank features, and different classifiers are used as a feature subset evaluating function. The proposed method is examined on five microarray expression datasets, including leukaemia, prostate, breast, nervous and DLBCL, and the average accuracies of the SVM classifier in these datasets are 100%, 95.24%, 85%, 91.67%, and 91.67%, respectively. The results show that the proposed method could not only improve the classification accuracy but also greatly reduce the computation time of the feature selection process.

Gene regulatory networks provide comprehensive insights and indepth understanding of complex biological processes. The molecular interactions of gene regulatory networks are inferred from a single type of genomic data, e.g., gene expression data in most research. However, gene expression is a product of sequential interactions of multiple biological processes, such as DNA sequence variations, copy number variations, histone modifications, transcription factors, and DNA methylations. The recent rapid advances of high-throughput omics technologies enable one to measure multiple types of omics data, called 'multi-omics data', that represent the various biological processes. In this paper, we propose an Integrative Gene Regulatory Network inference method (iGRN) that incorporates multi-omics data and their interactions in gene regulatory networks. In addition to gene expressions, copy number variations and DNA methylations were considered for multi-omics data in this paper. The intensive experiments were carried out with simulation data, where iGRN's capability that infers the integrative gene regulatory network is assessed. Through the experiments, iGRN shows its better performance on model representation and interpretation than other integrative methods in gene regulatory network inference. iGRN was also applied to a human brain dataset of psychiatric disorders, and the biological network of psychiatric disorders was analysed.