Current Status and Future Directions for TEXTAL

1
Current Status and FutureDirections for TEXTAL

• March 2, 2003
• The TEXTAL Group at Texas AM
• Thomas R. Ioerger
• James C. Sacchettini
• Tod Romo
• Kreshna Gopal
• Reetal Pai
• Vinod

2
Current model-building capabilities

• targeting medium-res maps (2.5-3.5A)
• builds 80-90 of models in medium-quality maps
• typically 2-10 chains
• RMSD error 1A for C-alphas and side-chains
• a.a. identity accuracy
• near 80-100 for good maps
• recommended make 2.8A map around 1 molecule
• time several hours (depends on size of map)

3
Progress Update

• Integration with PHENIX
• Sequence Alignment
• almost there (Tod)
• 80-100 acc. for good maps (IF5a, CzrA hi-res
  data)
• still face challenges with lower-quality density
• incorrect chain connections cause confusion
• truncated/unrepresentative side-chain density
• New algorithms
• connecting chain-breaks using fragment lookup
  (Reetal)
• new density-scaling technique (experimental)
• feature-weighting algorithm (Kreshna)

4
Current Developments

• Trying different databases
• real vs. back-transformed maps (phase error?)
• different resolutions!
• clustering (rotamers, Kmeans, SVD, reduce size)
• have to recalculate feature vectors and feature
  weights
• iterating between model-building and phase
  refinement
• (not much progress since Sept...)

5
Graphics

• Desired an interactive interface to TEXTAL
• semi-automated, rather than black-box
• give advice to user, but get their help
• generate alternatives, but let them decide
• Approach small targeted apps
• special-purpose, customized graphical tools
• e.g. link Capra chains, select better residues...
• Extend to VR and other interfaces

6
Textal Assistance
7
Virtual Benefits

• Vestibular cues reinforce stereo cues
• Immersive environment fosters rapid understanding
  of complex topologies
• Enabling technology for low spatial people
• Collaborative tool
• Shared virtual environment
• CAVElib support for connecting remote cave
  sessions

8
Longer-term Ideas

• Take structure factors as input, generate maps
• Extensions of pattern-recognition
• apply to nucleotides, carbohydrates?
• can use same features capturing shape of density?
• train neural network to recognize phosphodiester
  bonds same as C-alphas
• use linear/helical geometric constraints to build
  backbone then fit bases like LOOKUP
• how to distinguish DNA/RNA from protein?

9
Textal on-line