Science fiction often predicts the direction of cutting-edge research and suggests a future in which humans live and interact in physical realities modelled in silico (e.g. James Cameron’s download of the human consciousnesses into foreign bodies in Avatar). A key element in such models would be a complete description of all molecular interactions within living organisms.

We are fascinated by complex computational methods in information processing that can address data interpretation problems as we encounter them in structural biology. Indeed, recognising patterns in experimental data that describe macromolecules is itself an application of artificial intelligence. Structure determination provides essential data for integrative modelling of the basis of life: DNA, RNA, proteins, macromolecular complexes and assemblies. Current approaches, largely based on macromolecular X-ray crystallography, are static in nature and concentrate on a reductionist view of a single structure from a single method or experiment. Future applications (e.g. a quantitative description of the living cell) will necessitate radically different approaches where a wider context of information, using data from complementary tools, is implemented in computational methods serving as an integrated platform for a model of life.

ARP/wARP is a software suite to build macromolecular models in X-ray crystallography electron density maps. Structural genomics initiatives and the study of complex macromolecular assemblies and membrane proteins all rely on advanced methods for 3D structure determination. ARP/wARP meets these needs by providing the tools to obtain a macromolecular model automatically, with a reproducible computational procedure. ARP/wARP 7 tackles several tasks: iterative protein model building including a high-level decision-making control module; fast construction of the secondary structure of a protein; building flexible loops in alternate conformations; fully automated placement of ligands, including a choice of the best-fitting ligand from a ‘cocktail’; and finding ordered water molecules. All protocols are easy to handle by a nonexpert user through a graphical user interface or a command line. The time required is typically a few minutes although iterative model building may take a few hours.

Group members involved in project:

Related Websites:

Today’s highly automated structure solution tools – such as ARP/wARP – do most of the routine part of modelling alone. Immediate graphical feedback is lost however, viewing and manual work has moved towards the end when models get completed and ligands built – with other software.

Programs for semi-automatic modelling and validation are very popular and are always involved in all stages of modelling. To bridge this gap between automation and user control and feedback from ARP/wARP a graphical frontend has been developed, which allows interactive execution of modelling tasks.

The program brings with it a number of standard graphical tools and settings to adapt to a user’s visual habits. It allows to pick atoms, molecules or density regions and perform specific actions on them.

Currently a beta version is included in the ARP/wARP 7.1 release. These actions include ligand modelling, the tracing of alpha-helices and beta-strands and solvent building. Actions can be triggered from the graphics or via filling task-specific forms.

Group members involved in project:

The EMBL-Hamburg automated crystal structure determination platform is a system which contains several distinct decision-makers which utilize a number of macromolecular crystallographic software programs to produce a software pipeline for automated and efficient crystal structure determination. A large number of possible structure solution paths are encoded in the system and the optimal path is selected by the decision-makers as the structure solution evolves. The processes have been optimised for speed so that the pipeline can be used effectively for validating the X-ray experiment at a synchrotron beamline. Currently, the platform offers SAD, SIRAS, 2W-MAD, 3W-MAD or 4W-MAD phase determination, molecular replacement (MR) and MRSAD phasing. Recently it has been extended to include RIP and MRRIP phasing.

Group members involved in project:

Related Websites:

FREC software has been designed for automated image analysis of crystallization experiments using fluorescence from trace amounts of a nonspecific dye. The fluorescence images obtained strongly contrast protein crystals against other phenomena, such as precipitation and phase separation. FREC is able to quantitatively evaluate the crystallization outcome based on a biophysical metric correlated with voxel protein concentration.

Group members involved in project:

Related Websites:

Pattern recognition algorithms based on the interpretation of a shape are used ubiquitously in the current implementation of ARP/wARP. For example, the identification of planar objects in an electron density map produced by nucleotides, achieved with the use of third order moment invariants that parametrise such shapes as a series of numbers, aids the placement of the backbone base pairs in DNA and RNA. The automated tracing of a protein's backbone, meanwhile, is achieved by the recognition of atomic arrangements that correspond to the typical shape around a residue's alpha carbon atom. The automated recognition of ligands in electron density maps is highly desirable - endogenous substrates and inhibitors, as well as xenobiotics, are often found bound to proteins after crystallisation and their identification often aids in the interpretation of the macromolecule's function and mechanism of action. Furthermore, cocktails of potential drug molecules can be soaked into target proteins as a means of lead identification in structure based drug design, and the automated identification of the bound molecule clearly has the potential to speed up such efforts. However the task is difficult and complex owing to the chemical diversity of the ligands that can bind macromolecules and the conformational variability of many of these ligands - thus, the search space for ligand identification is huge relative to that of proteins or nucleotides for which certain shapes and structures are always found. Successful identification of ligands and their binding modes in a protein thus requires more sophisticated and complex methods than those implemented to date.

We are using a battery of shape-recognition algorithms in order to tackle ligand identification and building. By making use of already-implemented technologies and complementing them with novel features and descriptors, we can now recognise the compounds that are bound to proteins given a series of candidates, and furthermore, automatically build the ligands into the structure with a high level of success. We are now applying the same techniques in the converse direction, and thus identifying ligand binding sites in electron density maps given the bound ligand. By making use of surface description techniques developed by our collaborators at the European Bioinformatics Institute in the UK (Janet Thornton and Roman Laskowski), we should soon be able to identify binding sites from the proteins themselves, without the need for experimental data from crystallographic experiments.

A natural follow-on to the above work was to make use of similar methods for lead identification in structure based drug design (SBDD). We can now rapidly identify compounds from large databases that are complementary in shape to various clefts and cavities of interest in a protein structure. Current targets of interest include some malarial proteins isolated and identified by Matthew Groves as part of his study of potential malarial drug targets (as described further on this page) and various beta-lactamase isoforms (a collaboration with Alexei Igorov and Vitaly Grigorenko at Moscow State University, Russia). Future work in this regard will involve the incorporation of information related to protein residues at the binding site into the algorithms - for example, potential electrostatic interactions and hydrogen bonds that may facilitate binding will be sought.

Group members involved in project:

We are examining proteins involved in the generation and transport of oxaloacetate from the malarial parasite P. falciparum in collaboration with Dr. I. Mueller (BNI) and Prof. C. Wrenger (Sao Paulo). We apply a combination of techniques, including in vivo imaging, biochemical characterisation and structural biology.

Group members involved in project:

Formation of amyloid fibrils is characteristic for human diseases, such as Alzheimer's disease, type II diabetes and different forms of amyloidosis. Non-desease related amyloid proteins have been identified from bacteria and fungi and termed functional amyloids. Both the proteins of disease-related and functional origin are divers in size, fold and function, however, capable of forming a similar fibrillar array. We are fascinated by this phenomena and wish to establish the pathway through witch the formation of these fibrils occurs. Our research builds on class I hydrophobin, a functional amyloid, and protein crystallography and extends towards other types of amyloid forming proteins and lower resolution imaging techniques.

Group members involved in project:

Current development of methods for solving macromolecular structures is largely focused on high-resolution data. However, many of the most challenging questions, which have to be addressed by modern structural biology, require an analysis of large macromolecular complexes, for which only in rare cases high-resolution data can be obtained.

The interpretation of very low-resolution data (>10Ĺ) usually starts with the segmentation of the map (e.g., Watershed algorithm), which does not always give satisfactory results. Overall, fitting of known structures currently requires a lot of human expert knowledge and interaction. Instead, an automated objective procedure is highly desirable.

In this context our research is focussed on two scenarios: On the one hand on the model building in a low-resolution density map using the known high-resolution structures of the domains, and on the other hand on the interpretation of such a map, when it is not known how the structures of the individual subunits look like. Our main interest is the development of pattern recognition methods to extract the information which is present even in a noisy low resolution map, to build models of the protein structures.

We use 3rd-order moment invariants to identify regions in density maps of macromolecular complexes that match known structures or their fragments. Third-order moment invariants give a concise but comprehensive description of 3D objects, providing convenient means for fast searches through a large amount of 3D data. The placement of known domain structures is guided by the detection of similar patterns in the low-resolution map and the known domain structures. If th domain structures are not known the derived features are used for an exhaustive search for comparable patterns within the PDB. The placed candidate models are used for subsequent phase extension to higher resolution.

Group members involved in project:

Solving structures of large macromolecular structures and complexes is a challenging task in macromolecular crystallography (MX). Such crystals rarely diffract to high resolution - resulting in a scarce amount of data for generating electron density maps. Automated modelling approaches have been focused on high-resolution and their application to data of less than 3.0 Ĺ typically results in incomplete and highly fragmented models. Therefore a method using complementary and intrinsic information for structure completion is highly desirable for robust and automated solution of low-resolution MX structures.

At least 50% of all protein structures in the PDB contain fragments related by non-crystallographic symmetry (NCS). By using an all-vs-all least squares superposition of the built fragments, we identify NCS relations in the intermediate (partial) models. Typically, fragments are built differently in different NCS-related copies, thus relations are used to extend and connect fragments in subsequent building cycles. Exploiting NCS provides a significant improvement in many cases, often in less model-building cycles. Application of the method on real data improves model completeness by up to 10%, while 30% more of the sequence is docked and the fragmentation decreases by 25%. In case no NCS is present, we currently develop methods that will exploit the high abundance of structural information in the Protein Data Bank (PDB). By identifying experimental fragments that are very close in proximity and have a reasonable amount of unused density between them, fragments from the PDB can be used to close the gaps. Hence further increasing the structural information by complementary / a priori information. The iterative nature of ARP/wARP helps to avoid model bias in both cases.

Group members involved in project:

XREC is a software suite for automated crystal recognition. Currently the software is applicable for automated centring of crystal on a beamline goniostat. XREC processes a series of images, which display different orientations of a crystal flash-cooled in a loop, determines the crystal centre and provides the estimate of accuracy of the results in terms of their 'reliability'. Please refer to the XREC Manual for details. The functionality of the software is being extended for automated scoring of crystallisation conditions.

Group members involved in project:

Related Websites: