|
Informatics & Data Visualisation Conference
Intech Science Center
Winchester 10/11th March 2008
ABSTRACTS
Dr Paul J. Skipp -Centre for Proteomic Research, School of Biological Sciences, University of Southampton
Informatics in Proteomics - Finding Diagnostic Needles in Clinical Haystacks
Proteomics - the systematic study of the expression, functions and interactions of proteins - has important applications that span the life sciences but poses significant challenges. In addition to its dynamic nature, the complexity of the proteome is several orders of magnitude greater than that of the genome, due in large part to the post-translational processing of proteins. The range of expression levels and the variety of protein variants in the proteome have spurred efforts to develop new strategies for separation and quantification. Equally, however, they also create major opportunities for the discovery of new biomarkers of pathological conditions.
To derive biomarkers that are useful in the diagnosis, prognosis or monitoring of disease it is not only necessary to have access to high quality clinical samples but also to put in place sophisticated informatic strategies to deal with ‘noisy’ datasets with large number of variables (n = >500,000). This presentation will describe some of the challenges that arise from large-scale proteomic datasets and will show how multivariate statistics and data visualisation tools can be used to interrogate proteomic data to identify novel biomarkers for a variety of diseases.
Prof Richard G Brereton - Centre for Chemometrics, University of Bristol
Applications of Pattern Recognition in Chromatography
Pattern recognition is an important area of chemometrics. In modern chromatography it is possible to obtain large quantities of complex data rapidly, which are often too large to interpret manually. A typical example involves acquiring 1000 GCMSs of extracts of human sweat each with around detectable 500 peaks, or half a million peaks overall. Computational approaches are essential for interpretation. The aim is to relate data to properties of the samples being studied. It can be used both to study hypotheses, for example "can we determine whether by analysing the urine of a patient he or she has diabetes?" or to make predictions "how confident are we that a process is under control?"
The lecture will illustrate the power of pattern recognition in three areas. The first involves metabolic profiling, using extracts from bodily secretions e.g. urine and sweat to determine the provenance of the donor. It is necessary to analyse very large GCMS datasets using automated approaches. The importance of large sample sizes, well designed experiments, quality control in quantitative chromatography, and considering the null hypothesis are stressed. The second involves looking at pharmaceutical manufacturing processes. LCMS can be employed to determine the origins of a batch of tablets, by studying patterns in small impurity peaks which are highly diagnostic of manufacturing routes and so could be employed in patent protection. The third involves the use of on-line HPLC together with chemometrics, to monitor processes, allowing the potential of highly diagnostic real-time assessments of faults in processes together with interpretation of their origins.
The potential applicability of pattern recognition to chromatography is very broad, and has an especially important modern role as instruments have the ability to generate large quantities of data rapidly. There is surprisingly limited public domain software as there are requirements for peak detection and matching together with a broad range of modern approaches for pattern recognition involving handling very large datasets.
Dr Michael O’Shea: Tessella Plc
A Software System for Automated HTA in Heterogeneous Empower, ChemStation, and MassLynx Controlled Analytical Systems Using a Single Common User Interface
The technology for LC and MS instrumentation control, including the processing of resultant run information, has existed for decades and modern analytical laboratories most often use commercial off the shelf (COTS) systems for these purposes. The advantages of COTS are numerous and include validation, support, cost, instrumentation feature set, and add-on options. COTS does however tie-in the laboratory to a specific vendor and analytical scientists often become overly familiar with the functionality, GUI, and idiosyncrasies of a specific system at the expense of alternatives, possibly even in the same laboratory.
The presentation will describe a software system where analytical sample registration, analytical LC, MS, and data processing selection and configuration, processing, and reporting are implemented through a single unified interface, irrespective of the vendor of the underlying instrumentation control software. The software architecture extensibility will be outlined and demonstrate that the instrumentation components can be swapped in or out with little change to the common GUI or underlying software subsystem. Detail of how the software system accommodates custom post acquisition processing requirements, for example for mass spec/UV peak correlation and mass spec CODA processing, will also be discussed. The presentation will focus on an LC/MS implementation using instrumentation controlled by MassLynx,ChemStation, and Empower software at Pfizer Global Research & Development laboratories.
Ian Peirson1 & Michael McBrien2
1Advanced Chemistry Development, Inc., Bracknell, England, 2Advanced Chemistry Development, Inc. Toronto, Ontario, Canada
Strategies in automated Method Development
The iterative and inherently time-consuming nature of chromatographic method development logically leads the chromatographer to the conclusion that this is an area where automation must play a key role. Previous efforts in automated method development have thus focused considerable effort in the instrument control aspects of development, generating time savings in terms of chromatographers transcribing method information from modeling software to chromatography data systems. However, the system must also focus on the other aspects of method development: clear communication of what was done; effective, accurate automated data extraction; a clear mechanism for manual review and correction of results; and efficient reporting. A complete solution in automated method development should begin with an effective, efficient system for method development, and link this to instrument control.
Practically all of the features associated with a complete solution in automated method development are equally applicable to the computer-assisted method development process. This presentation will discuss the current state-of-the-art in Method Development Suite LC/MS 11.0, detailing the new capabilities that are to be enabling technologies for the upcoming release of the fully automated AutoChrom for ChemStation and AutoChrom for Empower.
Dr Chris Luscombe - GSK
Random Forests - Can we see the Wood for the Trees?
QSAR modelling tools have various advantages and disadvantages dependant on the modelling technique deployed. Ideally a modelling tool should be highly predictive, yet easily interpretable, two constraints that are often at odds with one another. In addition we need to be able to handle in many cases large numbers of input descriptors which may or may not be ultimately relevant to the response of interest. Decision trees are one technique appear to deliver these requirements in that we can produce predictive models in which irrelevant descriptors are ignored, and we are able to 'see' how the classification has been obtained. One drawback often cited regarding Decision trees is that they are sensitive to the primary cut and that the fact that small changes to the make up of the training set can lead to very different trees. It has recently been recognised that using multiple trees, a so called ensemble approach, can improve robustness of performance and Random Forests is the latest attempt at exploiting the advantages of using Decision trees in this manner. This presentation will cover some of the potentially unique features of RF and describe there application to particular ADME models within GSK.
Dr David Joyce - Thermo Fisher Scientific
DataManager - How XML Stands Aid Universal Storage, Recall and Reporting
Although we all applaud diversity and ‘variety is the spice of life’ proprietary standards for data storage add unnecessary friction to the pursuit of scientific research. Simplifying access to data and allowing rapid searching across multiple sources can significantly reduce the time wasted in administration (34%1) and duplication of effort across the organisation. (a further 80%2!).
In the ideal world, our instruments would all speak the same native language - and heroic efforts have been made to achieve this goal - notably JCAMP, NetCDF and lately AnIML. However none of these formats have achieved universal adoption.
Data Manager imposes order on the chaotic world of instrument data by converting over 180 proprietary formats into a single XML standard. The data is then stored securely, tagged and indexed to be accessed throughout the enterprise - without the need for the original software package.
The same conversion technology allows the ‘stand alone’ Envision desktop viewer to read, review and report directly from any one of these vendor formats providing a single tool to access all of your data.
Rapid data Access can be further enhanced by exposing the XML data to index based ‘Google’ like search tools via published standards like Microsoft IFilter. This allows federated searching across the organization and the possibility of introducing ontologies to further increase the relevance of search results.
By adopting such informatics techniques now common in other industries we can remove many of artificial obstructions imposed on our day to day working practices.
1 R.L. Harmon 1996 & Holmes and Yadani 1999
2 A.D. Little
Prof Jeremy Frey - School of Chemistry University of Southampton
"Blogs, Logs and Pods"
A wide ranging look at how we can use all the modern web type applications like blogs and pods to record data in the laboratory so that we can keep it, find it. talk about it, and re-use it.
Dr. Christoph Nickel – Agilent Technologies, Germany
Data Explosion and the Paperless Lab - How new Software Solutions can help Cope with these Challenges
Key drivers in the pharmaceutical industry have been the need for overall efficiency improvements in particular in R&D (the "drying pipeline" phenomena) and the need to comply with the continuously changing US and European regulations. Combined with the major trends of the ever-growing amount of data created in and around the lab and the attempt to eliminate paper documentation for decision-relevant information this presents a significant challenge.
Suppliers to the industry have reacted to their customers' needs with new tools and services. Two new software categories have been establish in reply to these needs: The "Electronic Lab Notebook (ELN)" and "scientific Data Management or lab-related Enterprise Content Management. This talk will discuss where these new applications help in meeting the needs mentioned above. It will provide an overview introduction of the problems ELN and SDMS/ECM solve and present some real-life customer case studies showing how specific issues of Research efficiency, intellectual property or compliance improvements can be solved with these tools. This talk will be given from a vendor's perspective and is intended to foster discussions around new trends and applications for existing software tools.
Mrs. Madeleine Ashley - British American Tobacco R&D.
Minitab as a Tool to Visualise Data
Minitab can be used to lighten the task of validating data ready for statistical analysis. British American Tobacco uses a technique called Filter Analysis to estimate smokers’ mouth level exposure to tar and nicotine using the tar and nicotine content of human smoked filter tips. Examples of validation of these data will be used to show how data can be investigated using Minitab, with an emphasis on graphical/visual analysis.
Mrs Heather Longden - Waters
Visualizing and Reviewing Data Using an Independent Data Viewer from Waters
In an ideal world there would be a single software package for all your analytical data review. While progress is being made to create single repositories for PDF, printed reports or raw data, you still need to take the data back to the original application to review it in detail or reprocess it
Waters now offers comprehensive data viewing and reviewing software for all generally available analytical methods with data read directly from many open and manufacturer-specific binary formats.
Using InSpector you will be able to compare, annotate and report all kinds of analytical data, without conversion, at your desktop. Annotations may be text, graphical or even chemical structures and reactions. If required, you can use InSpector to convert your data to both secure and open data formats.
Inspector is a stand alone visualization and reviewing application, however you will learn how this same capability is embedded in our other informatics products SDMS Vision Publisher and AWM, as well as working seamlessly with Empower and MassLyn
TUESDAY 11th March
Dr.John Tetteh - DiKnow Technologies
Avoiding Informatics Traps- From Data Capture to Analysis
This paper will present some informatics issues that impact on the quality of analytical output. The interplay between sample quality and matrix complexity as well as the correct choice of instrumentation and informatics tools will be presented. This paper aims to show how to avoid some informatics and visualisation traps that can lead to inaccurate analytical conclusions. Cases examples will be drawn from the life and material sciences.
Dr Andy Vine GSK
Letting the Data Speak - Understanding Processes through Visualisation
Previously the domain of manufacturing environments, 6 Sigma methods and tools are now being used in more complex environments such as Drug Discovery and elsewhere. Processes and Workflows within Drug Discovery can be complex. However, it is possible to measure and improve these processes by the appropriate application of Lean and Six Sigma tools and methods. These methods rely heavily upon visualisation and some statistical analyses, in order to allow the data to "speak". By using these methods, it is possible to 'see' and understand various discovery processes and to action improvements that ultimately lead to benefits and improved efficiencies that are sustainable. This talk will provide an insight into how this can be achieved in real world discovery environments, and the important factors that ensure pragmatic use of these tools, thus avoiding "analysis paralysis" or "metrics for metrics sake".
Dr Andrzej M. Kierzek - Faculty of Health and Medical Sciences, University of Surrey
Analysis of High Throughput Measurements in Molecular Biology in the Context of Molecular Interaction Networks of Living Cells.
A century of research in biochemistry and molecular biology resulted in the collection of voluminous datasets describing details of the molecular interactions in the living cell. Moreover, recent advances in experimental techniques allow monitoring of the state of the living cell at an unprecedented scale and level of detail. However, application of this knowledge in medical diagnostics, design of drugs, therapies and industrial organisms requires analysis of the complex, global network of molecular interactions. I will briefly review current computational approaches to predictive modelling of biochemical reaction networks. I will present results of the analysis of cDNA microarray data in the context of the global cellular metabolism model to demonstrate that interaction network models can be used to significantly enhance interpretation of multidimensional datasets obtained by contemporary high throughput measurement methods. I will also talk about the need for quantitative measurements in biological systems which are necessary to further increase predictive power of the computational cell biology models.
Dr Robert Lott - Varian
A Quality by Design Approach to Automated Method Development
There are many different ways that chromatographer have tried to systematically approach HPLC methods development. The statistical design of experiments (DOE) method is one of the best approaches. GalaxieTM Fusion AETM automates the statistical experiment design approach by determining experimental conditions and then creates methods and sequences which are automatically run on the instrumentation. The results are then sent back to Galaxie Fusion AE where they re-processed and displayed. Examples of rapid method optimizations and benchmarking of isocratic and gradient method development studies are discussed.
Richard Girdler - Pfizer Global R&D
Automation of HPLC purification
This presentation will talk about the issues and resolution of providing a single global solution to automating the HPLC purification process at Pfizer. It will describe the steps we took to reaching agreement, the business process analysis, and the final software solution. This software solution is based on the Teranode platform with in-house written adapters to communicate with Pfizer systems and other third party tools. It also integrates the in-house written QUARC software for quantitation by Evaporative Light Scattering Detector (ELSD)
Dr Steve Yemm - Symyx Technologies
Use of Laboratory Execution and Analysis Software for Automated Polymorphism Studies
The talk will describe an approach which uses commercial off the shelf software to capture data from diverse analytics, allowing users to mine, view, analyze, and apply automated ‘data to decision’ initiatives. This approach may be applied to diverse automation environments in the pharmaceutical and chemical industries and in this case is applied to the specific example of salt selection and polymorphism studies in drug development.
Dr Michael Cauchi - Cranfield University, Miss Jaspreet Babrah & Dr Nick Stone Gloucestershire Hospitals NHS Trust, and Dr Conrad, Bessant Cranfield University
Use of Principal Components Analysis (PCA) and False Colour Mapping (FCM) for Application in Clinical Diagnostics
In the UK, 284,560 persons are diagnosed with cancer each year. Breast, colorectal, prostate and blood cancers account for over 50% of all new cases of cancer. Patients who have not been diagnosed early enough tend to have a low survival rate. It is therefore imperative that early diagnoses are made in order to improve the chances of survival especially as the diseases can be curable if diagnosed at an early stage.
Generally, diagnoses are made via a host of biological processes which involve the sampling of blood and tissue from the patient, the aim being to assess the changes in pathology. However these assessments tend to be prone to delay, have a high risk of an unrepresentative sample being taken, and are costly in time, manpower and materials. It would therefore be beneficial if a more cost-effective and efficient process could be developed, and to this end we have been working on the novel spectroscopic methods.
This presentation discusses the application of the exploratory technique of principal components analysis (PCA) coupled with false colour mapping (FCM) to better enhance the hyperspectral images of cancer cell lines acquired from a mercury cadmium telluride (MCT) linear array detector coupled to an FT-IR spectrometer. In short, the scores of the first three principal components are used as RGB (red green blue) colour values for each pixel in the image from the MCT linear array detector, allowing it to be re-produced in the form of an FCM based on biochemical composition. It was observed that in most cases, there was a clear distinction between healthy and diseased tissue samples.
Dr John Hollerton –GSK
Data Rich, Information Poor
Pressures on the pharmaceutical industry have forced the discovery cycle to speed up dramatically. To deal with this, analytical systems have also improved in speed of data acquisition and data quality. This, in turn, has led to the generation of vast amounts of electronic data. The challenge now facing us is to convert this data into information of the same (or better) quality than the traditional "low-throughput" methods. Traditional methods relied on human interaction, this presentation identifies some of the challenges facing the automation of information derivation from raw data.
Dr Sabine McNiell - 3d Metrics
New Measures for Science through Quantifying Images
“Visual Data Intelligence” is software based on innovative proprietary mathematical techniques. When applied to analysing digital images, it can reveal new measures about the objects portrayed.
It can also be used to comparing imaging technologies in terms of resolution. And finally it forms the basis for discerning patterns such that images can be processed in a newly automated way This allows for selections within certain threshold values or by other ranking or sorting criteria.
Chris Taylor - EMBL-EBI (Cambridge) and NEBC (Oxford)
Standards Development: Necessarily a Two-Way Street
In recent years great strides have been made toward the establishment of public standards with which to capture and manage data generated by diverse bioscience workflows; this includes those employing one of several kinds of separation technologies. Latterly, several projects have attempted to integrate such standards to better support systems biology.
However, (or perhaps, as a result...) for 'non-involveds' the overall picture can be confusing, and motivations unclear.
Equally, for 'involveds' there is a concern that lack of input from the wider community will hinder development, and ultimately, uptake. Dialogue between standards developers and those they seek to support is therefore crucial if we are to have appropriate, useful standards with which to store, share and integrate data.
In this presentation, the range of public standards now available from various standards-generating bodies, and their relevance to the public and private sectors, will be described.
|
