1. RNA Helical Structures Prediction and Comparison
Faculty of Life Sciences, Computer Science Department
Bar-Ilan University
Nussinov-Wolfson Structural Bioinformatics Group, Faculty of Exact Sciences
Tel Aviv University
RNA Helical Structures Prediction
and Comparison
Liron Barzilay and Tom Susel
Project Advisor: Prof. Ruth Nussinov

2. Background: ncRNAs
DNA mRNA Protein 3%
~30K ?
DNA ncRNA
97%
>60K

3. Bacteriophage MS2 Capsid
Hepatitis delta virus Ribozyme
E. Coli Initiator tRNA
receptor 3 ectodomain complex with double-stranded RNA
Haloarcula marismortui
50s ribosomal unit

4. Structure Importance 1: Function

5. Structure Importance 2: Drugs Anti-Biotic Targeting:
RNA Helixes:
Shallow groove
Deep groove
Mixture of RNA Helixes
and phosphate groups:

6. Background: Structuring Main Idea 2D 3D ?

7. Background: Reduction to Helixes
WHY Helixes?
 >50% of all ncRNA nucleotides
 Most conserved (even more than loops)
 Usually Small (<15 BP, Common 2 BP)
 1D≈2D

8. Idea: RCSB PDB  ~800 ncRNAs.  ~14500 Helixes.
~ BP Helixes (Window 2).
1D D
1D D

9. Modified Residues (Nucleosides): Intra-Chain Direction Reverse:
Analysis: Complexity
Modified Residues (Nucleosides):
 mRNAs: 4-Letter Language.
 ncRNAs: ~110-Letter Language.
Intra-Chain Direction Reverse:
 mRNAs: Standard 3’-5’/5’-3’.
 ncRNAs: ~3% Reversed 3’-5’/3’-5’.

10.  Analysis: Complexity Broken Linkage:
 mRNAs: Complete Phosphate Skeleton.
 ncRNAs: ~80% Have One or More Phosphate Missing.
Non-wc BPs 
 mRNAs: Non-WC BPs Are Rare.
 ncRNAs: ~40% Non-WC BPs.

11. Analysis 1: Find Helixes
Manual selection of ncRNA
find_pairs 2D
Only Valid Helixes 1D 2D BP
Manager
Strand A
Strand B
Positions A
Positions B
Initial DB 3D

12. Analysis 2: Clustering Distance Matrix Engine All Against All RMSD
Strand A
Strand B
Positions A
Positions B
All Against
All
RMSD
Distance Matrix
Engine
Window 2
Window 3
Window 6
9240X9240
6848X6848
2273X2273
CLUTO

13. Output: Helix and Cluster Information
Analyzing Results
Input:
_Strand A_
Clusters DB
_Strand b_
Finding
The Best
Fitted Cluster:
Levenshtein Distance
(Edit Distance)
Finding
The Representing
Helix:
Inner Cluster RMSD (for closest members)
Output: Helix and Cluster Information

14. DEMO!

15. Some Statistics Code: Running Time: Requirements: C++: ~2200 lines.
Perl: ~1200 lines.
Html/Java script: ~200 lines.
Memory and
Multi-Processing Optimized.
Include Wrap for 3 external tools.
Running Time:
Started with 1.2gb of ncRNA PDBs
BP Manager: 2 hours .  mb of Helix-Only PDBs
DM Engine: 8h x 5 = 40 hours.
Perl Scripts: 30m x 5 = 2.5 hours.
305mb of small PDBs gb of Matrices
Requirements:
C++ code is for Ubuntu (Multi- Processing). Perl is OS-Independent.
DM Engine and the associated Perl scripts are very demanding.
For window 2 we needed Quad 3.0 GHz with 4gb of Memory to complete the run before memory overload.