iViS (In Vivo - In Silico) - Grand Challenge Website

Protein Structure

Bioinformatics based approaches to understanding the relationship between a protein’s structure and its dynamical behaviour. For example, enzymes exhibit conformational change upon substrate binding and product release. Detailed, atomic level information on protein internal motions usually either comes from X-ray crystallography or NMR experiments. In the former, for example, different conformations are observed when an open structure of an enzyme is found in the absence of a bound substrate, or substrate analogue, and the closed structure in the complexed state. In these cases one has detailed information about the internal rearrangements that occur upon substrate binding. More usually is not the case and the conformational states have to be predicted using computational models such as those developed by the group: DynDom. DynDom is the most popular program for analysing domain motions in proteins and is available from Collaborative Computational Project 4, CCP4 (X-ray crystallography) website, a major resource used by structural biologists from all over the world (see http://www.ccp4.ac.uk/main.html). The DyDom database (see http://www.sys.uea.ac.uk/dyndom ), is an online database of protein domain motions that have been analysed using the DynDom program. This site was set up in direct competition to a site on Macromolecular movements in Yale (see http://molmovdb.mbb.yale.edu/MolMovDB/) and is already proving itself to be popular, receiving hits from all over the world.

DOWNLOAD PDF

Pattern Recognition Algorithms for Biology

Kernel learning methods, and in particular Vapnik’s Support Vector Machine (as implemented in our toolbox), currently represent the most vibrant and promising area of research in machine learning, due to a combination of state-of-the art performance, computational efficiency and mathematical elegance. Kernel methods aim to construct very simple linear statistical models in a “feature space” indirectly specified by a kernel function, which essentially measures the similarity between vectors in this feature space. The mathematical tractability and computational efficiency of kernel methods are a result of the underlying linearity of the model. Fortunately a simple linear model constructed in feature space corresponds to a flexible, non-linear model of the original data, allowing state-of-the art performance on a wide range of real world problems. Kernel learning methods are particularly interesting in bioinformatics research as it is possible to construct kernel functions that operate on sequence data, for instance as DNA. In this case the induced feature space consists of the frequencies of all possible substrings of length k, however using a kernel approach it is not necessary to enumerate every substring explicitly (which would be computationally infeasible). Kernel machines have achieved impressive results in a range of bioinformatics applications, including analysis of gene expression microarray data, detecting remote protein homologies, analysis of promoter regions, recognizing translation initiation sites, protein fold classification and protein localization.

DOWNLOAD

Models of Biological Development

The sequencing of the Human Genome, Mouse, Drosophila, Arabidopsis and others opens the way to detailed studies relating genes to function. There have been a number of individual successes in the computational modelling of development patterns in some detail, and the use of L-systems to model visual and other aspects of development has been very successful. Coen [98] has shown the connection between Antirrhinum flower symmetry and gene expression. Sharpe et al have shown [400] how the three dimensional expression pattern of genes visualized using fluorescence can be recorded directly using optical computed tomography of fixed tissue. Clonal analysis can be coupled to mathematical models to infer the pattern of cell division and its relation to shape [381]. The development of markers such as Green Fluorescent Proteins (GFP) now allows a systematic approach to tracking growth through expression and by following division after microinjection. The computational aspects of this work are being developed in close collaboration with Professor Coen in the John Innes Centre (adjacent campus) and Professor Prusinkiewicz (Calgary).

DOWNLOAD