Lessons from CASP targets

ShuoYong Shi, Lisa Kinch, Jimin Pei, Ruslan Sadreyev, and Nick V. Grishin

Howard Hughes Medical Institute, Department of Biochemistry,

University of Texas Southwestern

Medical Center at Dallas

http://prodata.swmed.edu/CASP8

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

NF – new fold – historic category in CASP

New fold: were there any?

2008 – where did the new folds go?

176 domains: 2 possibly new folds: ~1%

N-domain of T0397: 3d4r chain A residues -7-82

New fold #1: N-domain of T0397

First models for T0397_1: Gaussian kernel density estimation for GDT-TS scores of the first server models, plotted at various bandwidths (=standard deviations). The GDT-TS scores are shown as a spectrum along the horizontal axis: each bar represents first server model. The bars are colored green, gray and black for top 10, bottom 25% and the rest of servers. The family of curves with varying bandwidth is shown. Bandwidth varies from 0.3 to 8.2 GDT-TS % units with a step of 0.1, which corresponds to the color ramp from magenta through blue to cyan. Thicker curves: red, yellow-framed brown and black, correspond to bandwidths 1, 2 and 4 respectively.

First server models for T0397_1

structure and topology diagrams of ferredoxin fold – fold closest to T0397_1

Most similar: ferredoxin-like fold

N-domain of T0496: 3do9 chain A, residues 4-126

New fold #2: N-domain of T0496

First models for T0496_1: Gaussian kernel density estimation for GDT-TS scores of the first server models, plotted at various bandwidths (=standard deviations). The GDT-TS scores are shown as a spectrum along the horizontal axis: each bar represents first server model. The bars are colored green, gray and black for top 10, bottom 25% and the rest of servers. The family of curves with varying bandwidth is shown. Bandwidth varies from 0.3 to 8.2 GDT-TS % units with a step of 0.1, which corresponds to the color ramp from magenta through blue to cyan. Thicker curves: red, yellow-framed brown and black, correspond to bandwidths 1, 2 and 4 respectively.

First server models for T0496_1

structure and topology diagrams of RNAseH fold – fold closest to T0496_1

Most similar: RNAse H fold

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

E.g.#1: T0460

Know fold: some predicted no better than new!

First models for T0460: Gaussian kernel density estimation for GDT-TS scores of the first server models, plotted at various bandwidths (=standard deviations). The GDT-TS scores are shown as a spectrum along the horizontal axis: each bar represents first server model. The bars are colored green, gray and black for top 10, bottom 25% and the rest of servers. The family of curves with varying bandwidth is shown. Bandwidth varies from 0.3 to 8.2 GDT-TS % units with a step of 0.1, which corresponds to the color ramp from magenta through blue to cyan. Thicker curves: red, yellow-framed brown and black, correspond to bandwidths 1, 2 and 4 respectively.

T0460: very difficult target

Cartoon diagram of 460: 2k4n model 1 residues 1-52,67-10

Jumping through 20 NMR models of 2k4n

Cartoon diagram of 460: 2k4n model 1 residues 1-52,67-10

Cartoon diagram of NADH-quinone oxidoreductase:

2fug chain 5 residues 1-106

T0460 is homologous to Nqo5

This homologous template was NOT FOUND BY ANY SERVER !

Why?

Singleton sequence!

E.g.#2: C-domain of T0407

Know fold: some predicted no better than new!

First models for T0407_2: Gaussian kernel density estimation for GDT-TS scores of the first server models, plotted at various bandwidths (=standard deviations). The GDT-TS scores are shown as a spectrum along the horizontal axis: each bar represents first server model. The bars are colored green, gray and black for top 10, bottom 25% and the rest of servers. The family of curves with varying bandwidth is shown. Bandwidth varies from 0.3 to 8.2 GDT-TS % units with a step of 0.1, which corresponds to the color ramp from magenta through blue to cyan. Thicker curves: red, yellow-framed brown and black, correspond to bandwidths 1, 2 and 4 respectively.

Date: Mon, 2 Jun 2008 23:56:39 -0500 (CDT)From: Nick Grishin <grishin@chop.swmed.edu>To: David Baker <dabaker@u.washington.edu>Cc: Ruslan Sadreyev <sadreyev@chop.swmed.edu>, Robert M Vernon <rvernon@u.washington.edu>Subject: Re: C-terminus of T0407

I liked IG because of 1) length; 2) ~7 strands; 3) many IG are interaction domains in enzymes.

These are very compelling reasons.

Cartoon diagram of 407, C-domain: 3e38 chain A

residues 277-363

Cartoon diagram of VAP-A MSP Homology

Domain: 3z9l

T0407_2 has Immunoglobulin fold

IG-based Baker modelTop GDT server model:

Phyre_de_novo TS1

No server predicted IG fold for T0407_2

Cartoon diagram of 407, C-domain: 3e38 chain A

residues 277-363

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

T0465: who found the template?

HHpred !!!

T0465 is a diverged FYSH domain

FYSH domain of hypothetical protein AF0491: 1t95 chain A

residues 11-94

Cartoon diagram of T0465: 3dfd chain A residues 21-136

T0465 fold is predicted by HHpred

HHpred2 TS1Cartoon diagram of T0465: 3dfd chain A residues 21-136

Falcon TS1

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

T0467: most interesting target !

Bioinfo.pl provides these predictions:

T0467: is bioinfo.pl correct ?

T0467 OB-fold C-terminal fragment:2k5q model 1 residues 64-97

Sso7d SH3-fold C-terminal fragment:2bf4 chain A residues 30-64

You can say so (if you want)

However, only local prediction is correct: extending it to cover the domain results in

a wrong fold prediction !

T0467 OB-fold: 2k5q model 1 residues 7-97

Sso7d SH3-fold: 2bf4 chain A

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

T0510: “server only” target with a twist

 

Cartoon diagram of 510 domains: 3doa, N-, middle and C-domains are shown in blue, green and red,

respectively.

Cartoon diagram of MutM domains: 1ee8_A, N-, middle and C-domains are shown in blue, green and red,

respectively.

Closer look at the N-domains reveals large topological differences

 

N-domain of 510: 3doa residues 1-165N-domain of MutM: 1ee8 chain A

residues 1-121insertion close to the N-terminus is red insertion in the middle of the domain is blue

N-domains are nevertheless homologous

1. New folds: 397_1, 496_1;

2. A few known folds are predicted no better than new folds: 460, 407_2;

3. Short motif recognition = success: 465;

4. Short motif recognition = failure: 467;

5. Structural changes not predicted: 510;

6. Inspect your alignments carefully: 480

T0480: easy alignment with templates

NADH pyrophosphatase intervening domain 1vk6: residues 94-127

Ribbon diagram of 480: 2k4x model 1 residues 17-50. Zinc ion is shown in magenta and side chains of its ligands (four Cys) are displayed.

T0480: most predictions had an error

480 MULTICOM-CLUSTER TS1

Jumping through 20 NMR models of 2k4x

Ribbon diagram of 480: 2k4x model 1 residues 17-50. Zinc ion is shown in magenta and side chains of its ligands (four Cys) are displayed.

T0480: unusual bulge

T0480: bulge could have been predicted

Summary:

1. New folds constitute less than 2% of newly solved non-redundant structures.

2. Many known folds cannot be predicted because templates are impossible to find.

3. Globalization of correct local alignment may or may not yield correct fold prediction.

4. Large structural changes happen in protein cores.

5. Careful inspection of alignments may solve some modeling problems.

Acknowledgement

Our group Collaborators

HHMI, NIH, UTSW,The Welch Foundation

Shuoyong Shi Jing TongRuslan Sadreyev Lisa KinchJimin Pei Ming TangSasha Safronova Yuan QiHua Cheng Jamie WrablIndraneel Majumdar Erik NelsonYong Wang S. Sri KrishnaBong-Hyun Kim Dorothee Staber

David Baker U. WashingtonKimmen Sjölander UC BerkeleyWilliam Noble U. Washington