Searching for Patterns in Activity across Multiple Targets

Introduction

Searching for Patterns in Activity across Multiple Targets is a BBSRC funded project. This project addresses priority area 3 of the Bioinformatics Initiative: novel computer science techniques applied to biology. We will apply advanced neural computing machine learning techniques to the visualization and analysis of high throughput screening data. The aims of the research project are to find meaningful structural and molecular properties that are good predictors of activity, and to use these methods to search for patterns of activity across multiple targets. The project is led by Ian Nabney working with Peter Tino (the postdoctoral RA) and Yi Sun (who is a PhD student). Additional funding for the work is provided from Pfizer Central Research Ltd.

Research Funding

BBSRC:     136,636 GBP

Industrial: 36,000 GBP

   Total:    172,636 GBP

Duration

Project Background

This project will extend our current knowledge and understanding of biological activity through the analysis of large HTS databases relating to many different targets.

The aim of HTS is to determine the few `lead' compounds (or `hit') which are active for a given biological target so that they can be considered for development into a product of commercial value. The high degree of automation in HTS means that extremely large databases of biological information can be built very quickly. A more detailed explanation of HTS can be found here. HTS can only be useful if the data is collected is accurate. The problem of quality control has been recognised by BBSRC as a key issue for biological databases, and it is the fifth of the priority areas in the Bioinformatics Initiative. Over the past two years, Pfizer has sponsored two students on the MSc by Research in Pattern Analysis and Neural Networks at Aston to investigate a solution to this problem for HTS data .

Objectives

• Develop viable training methods for the very large quantities of data typical of HTS.
• Explore molecular and structural properties that are good predictors of biological activity using ILP and ARD to build up and select attributes.
• Investigate the clustering of compounds using structure and screening information.
• Determine whether the 'inverse map' is possible to predict molecular properties from a desired pattern of activity over several targets.

Achievements

• We have developed a hierarchical Generative Topographic Mapping. Details can be found in the technical report Constructing localized non-linear projection manifolds in a principled way: hierarchical Generative Topographic Mapping.
• Matlab software for the hierarchical Generative Topographic mapping can be found here. This requires the Netlab toolkit.

Further Information

We attempt to produce online versions of all relevant publications arising from this project. You can see a list of all the available technical reports by following the link below:

Online Publications

This page is maintained by Yi Sun (suny@aston.ac.uk)