About Desert Scientific Software

• Founded in 2000, based in Sydney, Australia
  
  
• Produces web-based, data-driven software for medicines research
  
  
• Focus on creating scientific software that is useful and easy to use
  
  
• A major goal is the development of new methods for investigating protein-ligand interactions and characterising binding affinities
  
  
• First major product release - Proasis2
  
  
• DesertSci Advantages:
  
  
  • Python, Extreme Programming (XP), code re-use
    
    
  • collaborations with industry
    
    

Current Issues

• The pharmaceutical research community all want to make full use of all available protein structure data in their projects
  
  
• In the past, protein structure database technology has lagged behind developments in small molecule databases
  
  
  • most databases have rigid requirements associated with structure input
    
    
  • most databases have limited scope of output options
    
    
• Proasis2 has solved these deficiencies
  
  

What is Proasis2?

• Proasis2 is a Client-Server, Protein Structure Database and Visualisation System for drug discovery applications
  
  
  
• Proasis2 focuses on
  
  
  • storage and retrieval of 3D structures of proteins and their attributes
    
    
  • bound ligands
    
    
  • details of protein-ligand binding
    
    
  • structure comparisons
    
    
  • customisable
    
    
  • performance and ease-of-use
    
    

Proasis2 Can Benefit a Range of Scientists

• Proasis2 is:
  
  
  • an interoperable relational database system for research informatics
    
    
  • a storage and retrieval system for cystallographers
    
    
  • an aid for molecular modellers
    
    
  • a tool for chemists seeking a better understanding of a drug target and its interactions with small molecules
    
    

More About Proasis2

• Stores in-house and public domain protein crystal structure data, NMR structure data, and results from homology modelling and docking studies in a flexible database system
  
  
• Provides a mapping from the space of crystallographic data to the space of medchem projects - gives a 'medicinal chemistry' view of data
  
  
• Automates routine molecular modeling tasks
  
  
• Access using web browser and popular molecular graphics packages
  
  
• Proasis2 is a system that works the same way a brain works - hiding information that is not relevant to the task
  
  
  • for example, aminoacid residues distal to a bound ligand may be ignored when investigating ligand binding
    
    

Challenges Working With Protein Structure Data

• Protein structure data is complex - structures are very large and usually poorly resolved relative to small molecules
  
  
• Information resources are often disparate
  
  
• PDB format has many limitations
  
  
• Reliable chemical information hard to get
  
  
• Oligomeric systems require special attention
  
  
• Ligand binding modes are challenging to comprehend
  
  
• Legacy software designed for expert users
  
  
  

Proasis2 minimises burden of dealing with each of these

Proasis2: Implementation Details

• Server runs on SGI and Linux, Clients can use Windows, Unix, Macs
  
  
• Protein structures are indexed and annotated in Oracle or MySQL, data files are maintained on file system
  
  
• Server software written in Python and C
  
  
• Front-end typically HTML and Javascript, with all major browsers supported (Unix command line applications also available)
  
  
• Major graphics packages are supported, including RasMol, Chime, Accelrys (WebLab) Viewer; and Grasp on Silicon Graphics workstations
  
  

Proasis2: Data In

A Variety of Ligand Data is Stored

• Ligand attributes stored in databases tables - chemical name, registration ID, etc.
  
  
• 1D structure data stored as a molecular hash code (Ihlenfeldt and Gasteiger, J Comp Chem, 15, 793-813, 1994), for fast structure searching
  
  
• 2D structure data stored as SMILES (www.daylight.com), for sub-structure searching, and 2D Connection Tables (www.mdli.com), for creating depictions
  
  
• 3D Connection Table, for MCS based ligand overlays, and other molecular modelling applications
  
  
• Ligand chemistry obtained from:
  
  
  • in-house corporate database
    
    
  • manual input
    
    
  • public domain resources, e.g., CONECT records, CIF dictionary
    
    
  • predicted from geometry, e.g., using babel (often unreliable)
    
    

Structure Submission

• There are four methods for loading structures into the Proasis2 Database
  
  
  1. Web GUI
     
     
  2. Command line scripts - enabling batch submission
     
     
  3. Automatic structure submission - for legacy structures (and, soon to be implemeted, for weekly updates)
     
     
  4. Load pre-processed XML files - curated public domain data from DesertSci
     
     

Web GUI Structure Submission

• New structures can be individually loaded using a web browser
  
  
• Simple, two-step process:
  
  
  1. A form for uploading and pre-processing a pdb file (in-house or public domain) is provided
     
     
  2. Table data not obtained from the pdb parser is entered into a customised web page
     
     

3D Ligand Handling

• The creation of a valid Connection Table (CTab) for the ligand bound in the active site is the most demanding component of structure submission
  
  
  • This information is required by the modelling community
    
    
• A 3D CTab is usually created from a 2D CTab (eg, from a regno2sdf routine) and ATOM, HETATM and CONECT records in the pdb file
  
  
• When it works properly, a 3D CTab is generated with correct bond orders, and same atom ordering and coordinates as in the pdb file
  
  
• Structure submission provides a report of the success of 3D CTab generation
  
  

3D Ligand Handling (cont.)

• A 3D CTab generation can fail at several points - due to problems with the data (sometimes with the methods)
  
  
• Worst case senarios:
  
  
  • 2D CTab not provided, unable to create 3D CTab -> structure not submitted
    
    
  • 2D CTab provided, unable to create 3D CTab -> load 2D CTab into field meant to hold 3D CTab (having correct chemistry is higher priority)
    
    
  • 2D CTab not provided, automatic geration of 3D CTab -> load 3D CTab, probably with incorrect bond orders, into database
    
    
• 3D CTab stored in database may not even match ligand in protein, eg, missing pdb atoms or covalently bound ligand
  
  

Automatic Structure Submission

• DesertSci provides robust, automated methods that enable Proasis2 to be regularly updated with new public domain structures
  
  
• Proasis2 software can automatically:
  
  
  • parse Header and Coordinate Sections of pdb files
    
    
  • detect to which project a structure belongs
    
    
  • identify any ligand(s) in the binding site using a structure overlay method
    
    
  • identify multimeric systems
    
    
    
• Unclassified structures can be easily classified at a later time
  
  

Database Administration

• Comprehensive database administration facilities
  
  
• Proasis2 database can be updated through web pages, scripts, other connection clients
  
  
• Web pages enable database entries to be easily added, deleted, and modified
  
  

Security Issues

• Proasis2 is an intranet system protected by in-house firewalls and access protocols
  
  
• User and Administration Web pages are delineated enabling control of usage privileges
  
  
• Administration Web pages are password protected - users can only edit their own structures
  
  
  • furthermore, groups can be defined so that all users within a group can edit one-another's structures
    
    
• All structures are backed-up in XML format - minimises risk of losing data
  
  

Proasis2: Data Out

Backing-Up With XML

• XML (Extensible Markup Language) - robust framework for transfer of information
  
  
• All data about a structure, including original pdb file, can be backed-up to disk in xml files
  
  
• Proasis2 xml files can be easily transferred between different database installations
  
  
• XML parsing done using SAX (Simple API for XML) events
  
  

Structure Viewing

• Individual structures can be retrieved based on their identifier
  
  
• Proasis2 delivers graphical representation of protein structure data at the click of a button
  
  
• Fully operational molecular graphics applications are automatically launched on the desk-top
  
  
• Structure views correspond to those most widely used in molecular modelling/rational drug design community
  
  
• The major graphics packages are supported and many visualization modes are available for whole protein or binding site region
  
  
• HTML link to a structure view can potentially be created from any web page on an intranet site
  
  

The Classification of Structures into MedChem Projects

• A key objective of Proasis2 has been to organise protein structure data specifically for life sciences research
  
  
• From the very beginning, structures have been organised into MedChem projects
  
  
• Projects are automatically updated with publics structures from RCSB and manually updated with inhouse structures by crystallographers
  
  
• Once a project is selected, ALL structures relevant to a project can be easily retrieved and explored
  
  
  

Project Hierarchies

• Within large pharma, MedChem projects are typically related to one another, for example, Angiogenesis and Kinase targets
  
  
• Thus, a hierarchical representation scheme was created for Proasis2
  
  
• Database Tables contain project names, project parents, project types
  
  
• Simple DFS searching is used for tree traversal
  
  
• Javascript used for displaying trees in web pages with a MS Windows Explorer display style,
  
  
  

The Bigger Picture - Multiple Hierarchical Classification Schemes

• Proasis2 is able to manage, and maintain, multiple classifications, for example:
  
  
  • EC Number based classification
    
    
  • Human Kinome based classification
    
    
  • Ligand Chemotypes based classification (work in progress)
    
    
  • Fold based, Motif based, ...
    
    
    
• DesertSci provides robust, automated methods that enable Proasis2 to be regularly updated with new public domain structures
  
  
  
• The intention is to provide access to protein structure data to suit the many different approaches to research and to help invent new ones
  
  

Structure Searching

• Text based searching is newly available
  
  
  • Individual database fields can be searched, eg Title, Author, and/or Compound, or the entire Header Section can be searched
    
    
• Sub-structure searching, using DayCart, is newly available
  
  
  • SMARTS searches and similarity searches will be implemented soon
    
  • Alternative searching technologies could be applied instead, eg, using OpenEye software
    
    
• Sequence searching, using Blast, is newly available
  
  
• Searching for recently submitted structures is newly available
  
  
• All search queries can be fine-tuned:
  
  
  • all or any subset of projects can be searched
    
  • retrieve inhouse and/or public structures
    
  • retrieve xray, nmr, and/or model structures
    
  • a variety of sorting methods can be used
    
    
  
  
  
  

  Web-based Downloading

  
  
  • The following data can be easily downloaded from the web interface:
    
    
    • Original PDB
      
      
    • Curated PDB
      
      
    • Ligand SDF with/without DataFields
      
      
    • Structure Factor files
      
      
    • Topology files
      
      
    • Sequence in fasta format
      
      
    • Complete XML
      
      
  
  
  
  

  Align and Overlay Structures

  
  'Science involves discovering the similarities between things that are different and the differences between things that are similar.' A. Schopenhauer
  
  
  • Examining the similarities and differences between protein-ligand complexes is FUNDAMENTAL in rational drug design
    
    
  • Structure superimpositions are usually very time-consuming
    
    
  • Using Proasis2, overlays are made easy
    
    
  
  
  
  

  Protein Structure Alignment and Overlay using Proasis2

  
  
  • The are numerous ways to overlay structures:
    
    
    • sequenced-based alignments and backbone superimpostions
      
      
    • Secondary structure element based overlays
      
      
    • Ligand-based overlays
      
      
    • Multiple alignment versus pairwise
      
      
  • By default, Proasis2 software uses full sequence alignments (or binding site residues) for structures within protein families, and RMS superimpositions of alpha carbon atoms
    
    
  • Multiple structures are overlayed via multiple pairwise superimpositions onto a common reference structure
    
    
  • All monomers within oligomers can be overlayed
    
    
  
  
  
  

  Advanced Protein Structure Overlays

  
  
  • Key features:
    
    
    • Overlays can be done based on:
      
      
      • full sequence alignments
        
        
      • subsets of individually selected aminoacid residues
        
        
      • small molecule similarity, eg kinases JNK3 and LCK (1jnk 1qpc)
        
        
    • Any subset of database structures can be overlayed
      
      
    • Fine-control over size of the binding site displayed
      
      
    • All protein chains can be displayed or just the overlayed ligands
      
      
    • Results can be downloaded to a local file
      
      
  
  
  
  

  Small-Molecule Based Alignments

  
  
  • A fast method for creating protein overlays, can be very useful as a first step in generating fine tuned solutions, eg, for Kinases
    
    
  • Incorporates Spinifex - DesertSci's package for topological similarity based on maximum common substructures
    
    
  • Uses algorithms based upon clique detection (for finding the MOS in an edge induced correspondence graph) and chemical heuristics (for minimising search space)
    
    
  • Raymond et. al., 'Heuristcs for Similarity Searching of Chemical Graphs Using a Maximum Common Edge Subgraph Algorithm', J Comp Chem, 2002
    
    
  • Examples ...
    
    
  
  
  
  

  Small Molecule Alignments - Example 1

  
  
  • A simple example showing the Maximum Overlapping Set for a pair of molecules
    
    
  
  
  
  
  

  Small Molecule Alignments - Example 2

  
  
  • An example showing matching ring systems for COX2 inhibitors
    
    
  
  
  
  
  

  Protein-Ligand Binding Interactions

  
  
  • Proasis2 enables exploration of prt-lig close contacts in a binding site
    
    
  • Interactions computed using a close-packing like approach with atoms treated as non-spherical
    
    
  • Using Proasis2, one can easily explore hydrogen bonding networks, explore hydrophobic interactions, and identify how ligand binding affinity might be improved
    
    
  
  
  
  

  Highlighting of Residues Involved with Protein-Ligand Binding

  
  
  • Using Proasis2, it is possible to highlight residues along a protein sequence that make contact with the ligand
    
    
  • Residues can be differentiated depending on whether the ligand interacts with sidechain atoms, with mainchain atoms, or both mainchain and sidechain atoms
    
    
  • Useful in a range of applications such as: exploring selectivity issues, comparing multiple complexes, analysing the results from docking studies