page 1

These notes are taken from the presentation slides for CSE 648, November 14, 2000, by Vinhthuy Phan (phan@cs.sunysb.edu).

Threading

Many apparently unrelated sequences have been found to have very similar 3-dimensional structures.
Threading is based on fold-similarity; detect the fold of a sequence from a library of known folds.

NCBI CASP3 Assessment

Counts of "correct models" and "overpredictions":

Easy Medium Hard

Team T54 T79 T46 T53 T63 T71 T81 T83 T43 T44 T59 T67 T80 sum

003 + 1/0

005 + - - + + + + - - - - 5/6

009 - - + + - - - - - - - - - 2/11

017 + - + - - 2/3

019 - - - - - + + - - - 2/8

028 - + - - - - + - - - - 2/9

035 - + - - - - - - - 1/8

061 + - - - - - - - - - 1/9

066 - - + - - - 1/5

074 + - - - - - - 1/6

076 - + - - - - - - - 1/8

084 - + - - - - 1/5

156 + - - - - 1/4

162 + - - 1/2

166 + - - + - - - 2/5

176 - - + - - + - 2/5

212 + + + + + + 6/0

217 + - - + - - - 2/5

273 - - + - - - - 1/6

"Team Bryant"

Anna Panchenko,
Aron Marchler-Bauer,
Stephen H. Bryant,
from National Center for Biotechnology Information (NCBI).
Got 6 out of 6. Though, each is not the best.
Alignment calculated using the Gibbs-Sampling method. Other techniques that are used include variations of dynamic programming, search heuristics, HMMs.

Measurements

Global Distance Test (GDT)
- Percentage of residues in the model that are close to those in the template. "Closeness" means within 0.5A, or 1.0A, or 1.5A, ... , or 10.0A.
- Angstrom is the unit of distance in the C-alpha coordinate.

Percent of the structure predicted: 43.97.
Largest set of residues from the Model that can fit under DISTANCE cutoff :
13 percent of residues ( 15 residues ) fits under 1A with local RMSD 0.50
28 percent of residues ( 33 residues ) fits under 3A with local RMSD 1.88
37 percent of residues ( 43 residues ) fits under 6A with local RMSD 3.04

Detailed analysis of the model.
Each residue in the prediction is assigned to the largest set of residues containing that residue and deviating from the target by no more than a specified CA distance cutoff.
For a given residue the distance cutoff under which the set of 5, 10 and 50 percent of the modeled structure can be fitted is used for graphical presentation (three blue lines).

Target (top row): helices - PURPLE , strands - GREEN
Model (second row): helices - PURPLE , strands - GREEN
Target-Model GDT(5,10,50): BLUE
Bad model or residues not predicted: BLACK
Percent of the structure predicted: 43.97

Longest Continuous Segments (LCS)
- For a given residue the size of the longest segment that can be fitted under a specified CA RMSD cutoff of 1A, 2A, and 5A.
- LCS measures the longest segment in the model that is close to a segment in the template. "Closeness" means RMSD of 1A, 2A, or 5A.

Target (top row): helices - PURPLE , strands - GREEN
Model (second row): helices - PURPLE , strands - GREEN
LCS(1,2,5): BLUE
Bad model or residues not predicted: BLACK
Percent of the structure predicted: 43.97.
Longest Continuous Segments under RMSD cutoff : 1A 2A 5A : 14 14 18 residues

The Gibbs-Sampling Threader

Is a free software package provided by the National Center for Biotechnology. Developed by Bryant et al.
Setup Steps:
1. Getting target/probe sequences.
2. Defining a threading core.
3. Calculating Threading Alignments and Examining Results
4. Setting up a Database Search
5. Adjusting Confidence Values, Constraints

Getting target/probe sequences:

     gicA <- READ.FASTA("test.tfa")

Defining a Threading Core

a threading core -- minimum structure that is thought to be conserved through evolution.

a core description consists of 2 components:

Core Elements

csp.core <- CORE.DEFINITION("1CSP",chn=" ")
csp.core
$Core.Elements:
     Ctr NMn NMx CMn CMx 
[1,]   4   3   3   2   2
[2,]   9   2   2   1   3
[3,]  17   2   4   2   6
[4,]  28   3   4   1  11
[5,]  48   2   8   2   4
[6,]  60   2   7   1   2
[7,]  64   1   1   1   3

the structure has 7 core elements, with residues 4, 9, 17, etc. at the centers.
NMn, NMx : minimum and maximum n-terminal extensions of each core.
CMn, CMx : minimum and maximum c-terminal extensions of each core.

Loop Constraints

$Loop.Constraints:
     LMn LMx RMx 
[1,]   0   0   0
[2,]   0   5   5
[3,]   0   8   8
[4,]   0   7   7
[5,]   0  20  20
[6,]   0  11  11
[7,]   0   6   6
[8,]   0   0   0

LMn, LMx : minimum and maximun loop length.

Display using Rasmol:
```
RASMOL.CORE("1CSP  ",cdf=csp.core)
```
- red : minimum core elements.
- white : n-terminal extensions.
- green : c-terminal extensions.

Calculating Alignments and Examining Results

Do the actual threading:

 
THREAD(sqn="gicA",str="1CSP    ",cdf=csp.core,LRS=50)

Now look at the "Scores":
```
Thread.Set$Scores
              G0(s)  G(s) %core %id 
gicA 1CSP     -31.7 120.4   48  45.28
```
G0(s) : the reference "energy" for randomized sequences, used in the calculation of the composition corrected score.
G(s) : the raw threading "energy" of the native sequence.
%core : the fraction of the structural domain that has to be aligned (because of the minimum core size).
%id : the percentage of identical residues in the top threading alignment.

Alignment Result

THREAD.ALIGN.PRINT("gicA","1CSP    ",tnm="Thread.Set",cdf=csp.core)

Text Output:

gicA      MSKIKGnVKWFneskGFGFITpedgsKDVFVHFSAIQTNGFkTLAEGQRVEFEITngakg   60
identity  |   |                      ||||||||||  || || ||| | |||         23
1CSP      MLEGKV.KWFNsekgFGFIEVegq..DDVFVHFSAIQGEGFkTLEEGQAVSFEIVegnr.   56
coredefn  [--#-].[-#] <> [-#-]   <..>[--#]         <>     [-#-] <>   .    5

gicA      PSAANVIAL..   69
identity     |          24
1CSP      GPQAANVTKea   67
coredefn   [-#]<[#] <    7

[----#---] : core definition.
vertical bar : a residue match.

Using RasMol:
```
RASMOL.THRD("gicA","1CSP    ",tnm="Thread.Set",cdf=csp.core,modelnum=1)
```
Backbone colored in red for the core elements, and sidechains shown for residues identical with template residues.
Various other plots.

Setting Up a Data Search

Thread your sequence against hundreds of different cores. The cores are either pre-existing or predefined by an automatic algorithm.

Batch running. Example: search against a mini-databse of 5 proteins.

THREAD("gicA","1MJC    ",tnm="gicA.set")
THREAD("gicA","1CTF    ",tnm="gicA.set")
THREAD("gicA","1CRN    ",tnm="gicA.set")
THREAD("gicA","1BOV A  ",tnm="gicA.set")
THREAD("gicA","1MBN    ",tnm="gicA.set")

Results:
gicA.set$Scores G0(s) G(s) dG(s) Z(s) %seq %str %core Rand.Start rec %id gicA 1MJC -40.2 108.2 148.5 5.7657 81 81 42 125 5 67.86 gicA 1CTF -26.4 75.0 101.4 4.4370 72 68 50 125 3 8.00 gicA 1CRN -14.1 59.3 73.3 4.2978 55 83 46 125 4 15.79 gicA 1BOV A -33.6 101.8 135.5 5.5817 78 78 62 125 5 11.11