Back to "Biological Language Modeling Seminar Topics"

Ivet's slides

Protein Structure Prediction

1. Secondary Structure Prediction

max. accuracy average 78%, best methods use Neural Networks, and multiple sequence alignment

2. Tertiary Structure Prediction

The types of predictions in use:

Overall survey of structure prediction: Baker and Sali 2001 Science Paper

 

A. When no information but sequence and physical principles are  used

= ab initio structure prediction

- Blue Gene IBM 

 

B. When other information is used  (Survey of "ab initio" methods that use pdb information and their relation to protein folding)

 

        "fold recognition":

        requires a method for evaluating the compatibility of a given sequence with a given folding pattern

B0. 3D profiles

B1. Rosetta: conformations from short segments in pdb

B2. Including experimental structural constraints

B3. Threading (=sequence-structure alignment),

B4. Inverse threading and folding experiments Reference Ivet

    B4a. using short-range information

    B4b. using short- and long-range information

 

B4. Predicting structural class only    Reference Ivet

B5. Predicting active site only?

B6. Predicting protein-protein interaction sites?

B7. Predicting surface shape?

 

C. When a template with known structure must be available

homology modeling

 

D. Modeling structures based on experimental data

Both NMR and X-ray underdetermine the protein structure. To solve a structure one must minimize a combination of the deviation from the experimental data and the conformational energy:

D1. NMR (set of constraints on distances and angles)

D2. X-ray crystallography (Fourier transform of the electron density)

Evaluating structure prediction ability:

Use rmsd to known structures (from D) - defines structural similarity

Critical Assessment of Structure Predictions (CASP) competitions

EVA, EVA submits sequences automatically to different prediction servers shortly before structures are published in pdb, see links