Tuesday 23 October 2007

The drug discovery process and the pharmaceutical industry in general are beginning to face a real challenge: How can we maintain scientifically sound decision making and effective project management as the volume of data burgeons? Today there are few good enterprise solutions to this problem and usually organisations resort to email, Excel, and PowerPoint. In order to address this issue more effectively, better data management tools are needed.
Array is collaborating with General Dynamics VIZ to apply those advances to drug discovery and development command and control. The solution relies on their core technology platform to create an information-centric tool referred to as “CoMotion Discovery”. With this tool, a project’s plan, work progress, scientific results and management decisions across multiple data sources are integrated, updated and disseminated automatically through a single interface. Team members can collaboratively review data and annotate the results with their thoughts and intuitions (soft information), and link out to more details, supporting data or other related information. This presentation focuses on both the technical and cultural issues encountered with the development and deployment of the CoMotion Discovery tool.

Intellectual property management is a critical business process for organisations willing to protect existing products, foster innovation and increase revenue. But prior art analysis has become increasingly difficult in all industries, due to the number and the complexity of documents to analyse. Additionally, monitoring the competition R&D efforts is becoming more difficult and more resistant to manual analysis. This trend has been especially accentuated in the life sciences, with the ever increasing competition from generic and “me-too” drugs and the importance of licensing in and out drugs and molecules within the portfolio of all major players.
Patent specialists need to sift through large collections of patent documents, looking for specific clues. Online databases and visualisation tools provide a first answer to those questions but fail to offer a user-friendly environment to analyse the content of patents in a flexible and precise way. This presentation illustrates some real-life examples of in-depth patent analyses leveraging Text Analytics technology for relevant entities extraction (genes, diseases, chemical compounds, molecular targets).

The basic feature of an IT system that pretends to be “intelligent” is its capability of a specific kind of human-like understanding of the tasks for which it is designed. The challenges of a human-like understanding are described in general as well as their consequences for current scientific retrieval systems. In conclusion, the majority of today’s intelligent systems are classified as “fakes” – which does not necessarily restrict their practical usefulness. This is outlined with an example from intelligent text-mining in Bioinformatics. Finally, intelligent hard-science methods are discussed and demonstrated that circumvent basic problems of fakes but have limitations of their own.

Text mining and content discovery functions are becoming a basic function of enterprise applications. Smaller firms can match the intelligence and analytic functions of the largest organisations at greatly reduced costs. The reason for this development is that text mining functions have been embedded into enterprise applications, from Oracle's database to Microsoft's servers. One licence fee brings multiple tools. This shift is in response to demand for systems that can help manage the large flows of information in chemical or pharma companies.
A related development is that enterprise search system vendors have aggressively adopted text mining as a quick fix to their precision and recall woes. Fierce competition ensures that the licence fees for such systems will continue to drift downward and lead to easier integration among enterprise applications. Commoditisation of advanced text processing creates a number of opportunities for vendors to fill needs for specialised functions.

It is easier to find too many documents on a particular life science topic, than to find the right information inside these documents. With the application of text data mining to biological information, and the explosion of HTS data, it is no surprise that researchers are starting to look at applications to extract chemical information. The mining of chemical entities, both names and structures, brings with it some unique challenges. Ultimately, text data mining applications need to focus on the marriage of biological and chemical information.
Commercial and academic efforts are beginning to address these challenges. Developments in the World Wide Web are changing the way researchers and publishers look at published information and the way researchers from industrial and academic institutions interact with each other and with that information. This presentation explores ways in which all these changes can coalesce into something tangible for a researcher to navigate, and how researchers can infuse their knowledge and experience into the process of literature research. Finally, we consider how this information can be managed over time to find the 'right stuff'

The extensive use of virtual screening techniques and combinatorial chemistry in recent years have brought investigators and researchers large sets of compounds to be tested. The latest trends are focused on the intelligent selection of these compounds for specific targets.
BioEpisteme, a knowledge-based project, was initiated to contribute to the faster discovery of new and safer drugs, as well as the finding of new uses for known molecules. In-house developed datamining algorithms have led to a model that characterises more than 400 different molecular mechanisms of action simultaneously. Millions of molecules are being used to develop the project.

Additionally, the safety of drugs used in clinical practice is under constant scrutiny and the withdrawal of several compounds in recent years confirms the productivity challenges faced by modern biomedical research. In this context, the BioEpisteme technology has been used with success to build models that discriminate between active and inactive compounds for specific pharmacological activities, and to differentiate drug from non-drug compounds.

The development of the BioEpisteme Project and its application in high-throughput virtual screening, drug repositioning and ADMET assessment will be presented.

Retrieval of relevant information in the area of chemistry remains a challenge. Recent progress in the area of information extraction technologies promises to solve a broad spectrum of problems associated with the time consuming tasks of “finding all relevant information”, “finding all relevant mentions and relationships” and “making all this information available to an entire community / organisation”. However, despite the attractive perspectives of fully automated information extraction, there is considerable discrepancy between the promises of vendors of information extraction technology and the true problem solving of real existing problems in the day-to-day work of researchers and patent specialists.

This presentation reviews some of the recent benchmarking activities in the area of information extraction, with a special focus on named entity recognition and relationship mining. It sheds some light on recent research in the area of information extraction from chemical structure depictions, as communication of chemical information through images is one of the preferred routes taken in chemistry and pharmaceutical sciences. The presentation combines an overview on technological approaches with a critical review of performance measures and the outcome of critical assessments and benchmarking activities in the scientific community.

The Search and Business Intelligence markets made some surprisingly swift, significant movements towards each other in the past year. At industry analyst IDC, the Content team led by Sue Feldman and the BI team led by Henry Morris released a series of research reports on the fast emergence of "unified access" applications that are blurring the boundaries between these two fields of information access. Within months, BI giants Business Objects, Cognos, and Oracle all announced major new search features and products.

This convergence opens a rich area of study. At the outset, it touches the antipodes of unstructured versus structured content, atomic views versus aggregate views, and casual users versus trained users. For early adopters of unified access applications, it is already introducing functionality previously unavailable in either market. This presentation shows examples from high seat-count deployments where enterprise users are finding the information they need in unprecedented ways when Search and Business Intelligence converge.