Antony Browne

Biological data mining with neural networks: implementation and application of a flexible decision tree extraction algorithm to genomic problem domains

In the past, neural networks have been viewed as classification and regression systems whose internal representations were extremely difficult to interpret. It is now becoming apparent that algorithms can be designed which extract understandable representations from trained neural networks, enabling them to be used for data mining, i.e. the discovery and explanation of previously unknown relationships present in data. This paper reviews existing algorithms for extracting comprehensible representations from neural networks and describes research to generalize and extend the capabilities of one of these algorithms. The algorithm has been generalized for application to bioinformatics datasets, including the prediction of splice site junctions in Human DNA sequences. Results generated on this datasets are compared with those generated by a conventional data mining technique (C5) and conclusions drawn.

Automatic Relevance Determination for Identifying Thalamic Regions Implicated in Schizophrenia

There have been many theories about and computational models of the Schizophrenic disease state. Brain imaging techniques have suggested that abnormalities of the Thalamus may contribute to the pathophysiology of Schizophrenia. Several studies have found the Thalamus to be altered in Schizophrenia, and the Thalamus has connections with other brain structures implicated in the disorder. This paper describes an experiment examining thalamic levels of the metabolite N-acetylaspartate (NAA), taken from schizophrenics and controls using in vivo proton magnetic resonance spectroscopic imaging. Automatic relevance determination was performed on neural networks trained on this data, identifying NAA group differences in the pulvinar and mediodorsal nucleus, underscoring the importance of examining thalamic subregions in schizophrenia.

Filtering normal retinal images for diabetic retinopathy screening using multiple classifiers

Diabetic retinopathy is a complication of diabetes and early detection is essential for effective treatment. In this paper, a novel technique for the separation of normal and abnormal retinal images is described. Various features are extracted from local sub images and then fed through multiple classifiers to categorise them into interim classes followed by a reasoning process to give a more reliable and robust result. This is then followed by a global analysis to decide the normality of the whole image.

Using Ensembles of Neural Networks to Improve Automatic Relevance Determination

Automatic relevance determination (ARD) is an efficient technique to infer the relevance of input features with respect to their ability to predict the target output for a task. ARD optimizes the hyperparameters to maximize the evidence. This optimization can cause some hyperparameters of relevant features tends towards infinity and therefore these features are inferred as irrelevant by an ARD model. The overfitting of relevance parameters cause feature relevance determinations to be not stable and reliable. Neural network ensemble methods can utilize the diversity between ensemble members to reduce the uncertainty in order to generate a more reliable determination of input feature relevancies. Input features were properly grouped based on their relevance level by ensemble relevance prediction.

Use Of Neural Networks To Predict And Analyse Membrane Proteins In The Proteome

proteinsNeural NetworksKnowledge DiscoverySecondary Structure Prediction There have been several att mpts over th last 20 years to develop tools for predicting membrane-spanning regions, but the problem of prediction is made topologically more complex by the presence of several transmembrane domains in many proteins, and current tools are far away from achieving 95% reliability in prediction. Though neural networks have been considered as classification and regression systems whose inner working principles were very difficult to interpret, it is now becoming apparent that algorithms can be designed which extract understandable representations from trained neural networks that might be a powerful tool for biological data mining. In this research construction of novel neural network architectures/algorithms, amino acid representations to the neural networks with appropriate encodings and understanding of the relationship between structure and function of transmembrane proteins were studied.

Classifying Membrane Proteins in the Proteome by Using Artificial Neural Networks Based on the Preferential Parameters of Amino Acids

1 MRC Clinical Sciences Centre, Faculty of Medicine, Imperial College London, UK subrata.bose@ic.ac.uk 2 Department of Computing, School of Engineering and Physical Sciences, University of Surrey, Guildford, UK a.browne@eim.surrey.ac.uk 3 Intelligent Systems Research Centre, Department of Computing, Communication Technology and Mathematics, London Metropolitan University, London, UK h.kazemian@londonmet.ac.uk 4 Institute for Health Research and Policy, Departments of Health and Human Sciences, London Metropolitan University, London, UK kenneth.white@londonmet.ac.uk

Investigating the influence of feature correlations on automatic relevance determination

Feature selection is the technique commonly used in machine learning to select a subset of relevant features for building robust learning models. Ensemble feature relevance determination can properly group the most relevant features together and separate the relevant features from the irrelevant and redundant features. However, it cannot provide reliable local feature relevance rank. In this paper, we demonstrate that the predicted local relevance rank for the relevant features could be influenced by their highly correlated redundant features, according to the strength of their correlations.

Presenting a Novel Neural Network Architecture for Membrane Protein Prediction

In the early seventies, it was clear that primary amino acid sequence and its local solution environment hold most of the information necessary for protein folding. Since then, scientists have been trying to solve the bioinformatics problem by constructing the tertiary three-dimensional structure of protein from the primary amino acid sequences by using computational biology. Success of several genome sequencing projects put considerable momentum in an effort to analyze these bio-chemically uncharacterized sequence. A handful of methods are developed to solve the problem of globular proteins prediction because of the easy availability of the data but the prediction of membrane protein structures is a key area that remains mainly unsolved. The problem of prediction is made topologically more complex by the presence of several transmembrane domains in many proteins, and current tools are far away from achieving significant reliability in prediction. But from a pharmaeconomical perspective, though it is the fact that membrane proteins constitute ~ 75% of possible targets for novel drugs but MPs are one of the most understudied groups of proteins in biomedical research. In this paper we present novel neural networks (NNs) architecture and algorithms for predicting membrane spanning regions from primary amino acids sequences by using their preference parameters

Modelling the Stroop Effect: Dynamics in Inhibition of Automatic Stimuli Processing

In this study, we simulate the dynamics of suppressing an automatic stimulus processing which interferes with a different non-automatic target task. The dynamics can be observed in terms of interference and facilitation effects that influence target response processing time. For this purpose, we use Hopfield neural network with varying attention modulation in a colour-word Stroop stimuli processing paradigm. With the biologically realistic features of the network, our model is able to model the Stroop effect in comparison to the human performance.

Using Neural Networks with Automatic Relevance Determination to Identify Regions of the Thalamus Implicated in Schizophrenia

Abnormalities of the thalamus may contribute to the pathophysiology of schizophrenia. Several studies have found the thalamus to be altered in schizophrenia, and the thalamus has connections with other brain structures implicated in the disorder. This work examined thalamic levels of the metabolite N-acetylaspartate (NAA), taken from schizophrenics and controls using in-vivo proton magnetic resonance spectroscopic imaging. Neural networks were trained on this data, and automatic relevance determination (ARD) identified NAA group differences in the pulvinar and mediodorsal nucleus, underscoring the importance of examining thalamic subregions in schizophrenia.