Andrew Starkey

Differential Activity by Polymorphic Variants of a Remote Enhancer that Supports Galanin Expression in the Hypothalamus and Amygdala: Implications for Obesity, Depression and Alcoholism

The expression of the galanin gene (GAL) in the paraventricular nucleus (PVN) and in the amygdala of higher vertebrates suggests the requirement for highly conserved, but unidentified, regulatory sequences that are critical to allow the galanin gene to control alcohol and fat intake and modulate mood. We used comparative genomics to identify a highly conserved sequence that lay 42 kb 5′ of the human GAL transcriptional start site that we called GAL5.1. GAL5.1 activated promoter activity in neurones of the PVN, arcuate nucleus and amygdala that also expressed the galanin peptide. Analysis in neuroblastoma cells demonstrated that GAL5.1 acted as an enhancer of promoter activity after PKC activation. GAL5.1 contained two polymorphisms; rs2513280(C/G) and rs2513281(A/G), that occurred in two allelic combinations (GG or CA) where the dominant GG alelle occurred in 70-83 % of the human population. Intriguingly, both SNPs were found to be in LD (R2 of 0.687) with another SNP (rs2156464) previously associated with major depressive disorder (MDD). Recreation of these alleles in reporter constructs and subsequent magnetofection into primary rat hypothalamic neurones showed that the CA allele was 40 % less active than the GG allele. This is consistent with the hypothesis that the weaker allele may affect food and alcohol preference. The linkage of the SNPs analysed in this study with a SNP previously associated with MDD together with the functioning of GAL5.1 as a PVN and amygdala specific enhancer represent a significant advance in our ability to understand alcoholism, obesity and major depressive disorder.

Evidence of uneven selective pressure on different subsets of the conserved human genome; implications for the significance of intronic and intergenic DNA.

BackgroundHuman genetic variation produces the wide range of phenotypic differences that make us individual. However, little is known about the distribution of variation in the most conserved functional regions of the human genome. We examined whether different subsets of the conserved human genome have been subjected to similar levels of selective constraint within the human population. We used set theory and high performance computing to carry out an analysis of the density of Single Nucleotide Polymorphisms (SNPs) within the evolutionary conserved human genome, at three different selective stringencies, intersected with exonic, intronic and intergenic coordinates.ResultsWe demonstrate that SNP density across the genome is significantly reduced in conserved human sequences. Unexpectedly, we further demonstrate that, despite being conserved to the same degree, SNP density differs significantly between conserved subsets. Thus, both the conserved exonic and intronic genomes contain a significantly reduced density of SNPs compared to the conserved intergenic component. Furthermore the intronic and exonic subsets contain almost identical densities of SNPs indicating that they have been constrained to the same degree.ConclusionOur findings suggest the presence of a selective linkage between the exonic and intronic subsets and ascribes increased significance to the role of introns in human health. In addition, the identification of increased plasticity within the conserved intergenic subset suggests an important role for this subset in the adaptation and diversification of the human population.

Using a Lumped Parameter Dynamic Model of a Rock Bolt to Produce Training Data for a Neural Network for Diagnosis of Real Data

Ground anchorage systems are used extensively throughout the world as supporting devices for civil engineering structures such as bridges and tunnels. The condition monitoring of ground anchorages is a new area of research, with the long term objective being a wholly automated or semi-automated condition monitoring system capable of repeatable and accurate diagnosis of faults and anchorage post-tension levels. The ground anchorage integrity testing (GRANIT) system operates by applying an impulse of known force by means of an impact device that is attached to the tendon of the anchorage. The vibration signals that arise from this impulse are complex in nature and require analysis to be undertaken in order to extract information from the vibrational response signatures that is relevant to the condition of the anchorage. Novel artificial intelligence techniques are used in order to learn the complicated relationship that exists between an anchorage and its response to an impulse. The system has a worldwide patent and is currently licensed commercially.A lumped parameter dynamic model has been developed which is capable of describing the general frequency relationship with increasing post-tension level as exhibited by the signals captured from real anchorages. The normal procedure with the system is to train a neural network on data that has been taken from an anchorage over a range of post-tension levels. Further data is needed in order to test the neural network. This process can be time consuming, and the lumped parameter dynamic model has the potential of producing data that could be used for training purposes, thereby reducing the amount of time needed on site, and reducing the overall cost of the systems operation.This paper presents data that has been produced by the lumped parameter dynamic model and compares it with data from a real anchorage. Noise is added to the results produced by the lumped parameter dynamic model in order to match more closely the experimental data. A neural network is trained on the data produced by the model, and the results of diagnosis of real data are presented. Problems are encountered with the diagnosis of the neural network with experimental data, and a new method for the training of the neural network is explored. The improved results of the neural network trained on data produced by the lumped parameter dynamic model to experimental data are shown. It is shown how the results from the lumped parameter dynamic model correspond well to the experimental results.

Use of neural networks in the condition monitoring of ground anchorages

The GRANIT system is a non-destructive integrity testing method for ground anchorages. It has won two major UK Awards--the Design Council Millennium Product Status in 1999 and the John Logie Baird Award in 1997. It makes use of novel artificial intelligence techniques in order to learn the complicated relationship that exists between an anchorage and its frequency response to an impulse. The GRANIT system has a worldwide patent and is currently licensed to AMEC plc.It is widely recognised that non-destructive testing methods for ground anchorages need to be developed as a high priority [A National Agenda for Long-term and Fundamental Research for Civil Engineering in the United Kingdom (1992)], with only between 1 and 5% of anchorages currently being monitored in service [British Standard Code of Practice for Ground Anchorages (8081) (1989)] using currently available techniques. The GRANIT system is a solution for this requirement, and this paper describes how the use of artificial intelligence techniques enabled, for the first time, the cross-anchorage diagnosis of ground anchorages, where data taken from one anchorage was used to train a neural network, which was then used to diagnose the condition of an adjacent anchorage.The results presented in this paper describe the training of a neural network on data taken from a bolt anchorage, and the diagnosis, using this neural network, of further test data taken from the same anchorage. Data taken from an adjacent anchorage of similar construction is also presented to the neural network, and the cross-anchorage diagnosis of the load level of the second anchorage is achieved.

Condition Monitoring of Ground Anchorages by Dynamic Impulses: GRANIT system

The GRANIT system operates by applying an impulse of known force by means of an impact device that is attached to the tendon of the anchorage. The vibration response signals resulting from this impulse are complex in nature and require analysis to be undertaken in order to extract information from the vibrational response signatures that is relevant to the condition of the anchorage. In the system, the complicated relationship that exists between characteristics of an anchorage and its response to an impulse is identified and learned by a novel artificial intelligence network based on artificial intelligence techniques.The results presented in this paper demonstrate the potential of the GRANIT system to diagnose the integrity of ground anchorages at a site near Stone, England, by using a trained neural network capable of diagnosing the post-tension level of the anchorage. This neural network was used for the diagnosis of load in a second ground anchorage adjacent to the original anchorage used for the training of the neural network. Further tests were taken with a different anchor head configuration of the anchorage and a different relationship between the signature response of the anchorage to an applied impulse and its post-tension level was found.Problems encountered during the diagnosis of this second set of test signatures by the trained neural network are investigated with the use of a lumped parameter dynamic model. This model is able to identify the parameters in the anchorage system that affect this change in response signature. The results from the investigation lead to a new form of classification for the installed anchorages, based on their anchor head configuration.Laboratory strand anchorage tests were undertaken in order to compare with and validate the results obtained from the field tests and the lumped parameter dynamic model.

Answer Type Identification for Question Answering

Question Answering research has long recognised that the identification of the type of answer being requested is a fundamental step in the interpretation of a question as a whole. Previous strategies have ranged from trivial keyword matches, to statistical analyses, to well-defined algorithms based on shallow syntactic parses with user-interaction for ambiguity resolution. A novel strategy combining deep NLP on both syntactic and dependency parses with supervised learning is introduced and results that improve on extant alternatives reported. The impact of the strategy on QALD is also evaluated with a proprietary Question Answering system and its positive results analysed.

Making Test Corpora for Question Answering More Representative

Despite two high profile series of challenges devoted to question answering technologies there remains no formal study into the representativeness that question corpora bear to real end-user inputs. We examine the corpora used presently and historically in the TREC and QALD challenges in juxtaposition with two more from natural sources and identify a degree of disjointedness between the two. We analyse these differences in depth before discussing a candidate approach to question corpora generation and provide a juxtaposition on its own representativeness. We conclude that these artificial corpora have good overall coverage of grammatical structures but the distribution is skewed, meaning performance measures may be inaccurate.

Application of feature selection methods for automated clustering analysis : a review on synthetic datasets

The effective modelling of high-dimensional data with hundreds to thousands of features remains a challenging task in the field of machine learning. This process is a manually intensive task and requires skilled data scientists to apply exploratory data analysis techniques and statistical methods in pre-processing datasets for meaningful analysis with machine learning methods. However, the massive growth of data has brought about the need for fully automated data analysis methods. One of the key challenges is the accurate selection of a set of relevant features, which can be buried in high-dimensional data along with irrelevant noisy features, by choosing a subset of the complete set of input features that predicts the output with higher accuracy comparable to the performance of the complete input set. Kohonen’s self-organising neural network map has been utilised in various ways for this task, such as with the weighted self-organising map (WSOM) approach and this method is reviewed for its efficacy. The study demonstrates that the WSOM approach can result in different results on different runs on a given dataset due to the inappropriate use of the steepest descent optimisation method to minimise the weighted SOM’s cost function. An alternative feature weighting approach based on analysis of the SOM after training is presented; the proposed approach allows the SOM to converge before analysing the input relevance, unlike the WSOM that aims to apply weighting to the inputs during the training which distorts the SOM’s cost function, resulting in multiple local minimums meaning the SOM does not consistently converge to the same state. We demonstrate the superiority of the proposed method over the WSOM and a standard SOM in feature selection with improved clustering analysis.

Automated Feature Identification and Classification Using Automated Feature Weighted Self Organizing Map (FWSOM)

This paper investigates the application of a novel method for classification called Feature Weighted Self Organizing Map (FWSOM) that analyses the topology information of a converged standard Self Organizing Map (SOM) to automatically guide the selection of important inputs during training for improved classification of data with redundant inputs, examined against two traditional approaches namely neural networks and Support Vector Machines (SVM) for the classification of EEG data as presented in previous work. In particular, the novel method looks to identify the features that are important for classification automatically, and in this way the important features can be used to improve the diagnostic ability of any of the above methods. The paper presents the results and shows how the automated identification of the important features successfully identified the important features in the dataset and how this results in an improvement of the classification results for all methods apart from linear discriminatory methods which cannot separate the underlying nonlinear relationship in the data. The FWSOM in addition to achieving higher classification accuracy has given insights into what features are important in the classification of each class (left and right-hand movements), and these are corroborated by already published work in this area.

Comparison of Methods for Automated Feature Selection Using a Self-organising Map

The effective modelling of high-dimensional data with hundreds to thousands of features remains a challenging task in the field of machine learning. One of the key challenges is the implementation of effective methods for selecting a set of relevant features, which are buried in high-dimensional data along with irrelevant noisy features by choosing a subset of the complete set of input features that predicts the output with higher accuracy comparable to the performance of the complete input set. Kohonen’s Self Organising Neural Network MAP has been utilized in various ways for this task. In this work, a review of the appropriate application of multiple methods for this task is carried out. The feature selection approach based on analysis of the Self Organising network result after training is presented with comparison of performance of two methods.