15-comparative-modelling.md

1.3 Comparative Modelling

18.05.2017 | Slides | Lecture Recording

---

• 1. Recap

• Profile-Sequence comparisons are more accurate than sequence-sequence aligments

• Profile-Profile alignments gain even more accuracy

Question: How do you build up a family (profile) of sequences?

1. Find proteins of similar sequence with BLAST
2. Use the found proteins to build a PSSM (profile)
3. Use profile-sequence alignment with the calculated PSSM to retrieve more distant family members
4. Add the newly found proteins to the family by recalculating the PSSM

• When building up a profile, start with a high threshold (only very similar sequences are taken), so the profile is not wrong from the beginnig

2. Goal of structure prediction

Sequence uniquely determines structure! ➡ Thus, from a sequence it should be possible to predict 3D structure and function

How would you assess prediction performance?

CASP: Critical Assessment of Structure Prediction

• Yearly event
• Submit predictions for structures, which will be experimentally predicted before a deadline
• Compare (after release of experimental structures) how the methods performed

Current State

• Only Homology Modeling is good
• No general prediction of 3D structure from sequence yet
• BUT: Important improvement in many fields

3. Structure by Experiment

Different Methods to determine 3D structure

• 90% - X-Ray Crystallography
• 09% - Nuclear Magnetic Resonance Spectroscopy (NMR)
• 01 % - Cryo Electron Microscope (Cryo-EM)

**X-Ray Crystallography **

1. Grow Crystal: Force the protein to grow a crystal
2. Observe Diffraction Pattern: Shoot x-rays onto crystal and observe the diffraction pattern
3. Compute Electron Density Map
4. Fit observations to atomic model

**NMR **

1. Protein has to be in similar solution as naturally
2. Massive Magnets required

**Cryo-EM **

• worse resolution than other methods
• cheaper than other methods
• Pushing the boundaries of resolution of Cryo-EM is the future

Question: Which methods to experimentally determine the structure of a protein exist? How much are they used?

Fraction of proteins in the PDB by experimental method:

• 90% - X-Ray Crystallography
• 09% - Nuclear Magnetic Resonance Spectroscopy (NMR)
• 01 % - Electron Microscope (EM)

Question: How does X-Ray Crystallography Work

1. Grow Crystal: Force the protein to grow a crystal
2. Observe Diffraction Pattern: Shoot x-rays onto crystal and observe the diffraction pattern
3. Compute Electron Density Map
4. Fit observations to atomic model

Hydrogen Bond Formation

💡 Idea: Secondary structure is completely explained by hydrogen bond formation.Helix: Hydrogen-Bond between residue i and residue i+4, which stabilize the helix.
Sheet: Two strands come together to form a sheet by forming hydrogen bonds between them

Question: How to get 1D secondary structure from 3D coordinates?

Two methods where used to annotate 3D coordinates:

1) DEFINE, based on geometry (not used anymore)
2) DSSP, based on hydrogen bond pattern (coulomb energy)

4. Comparative Modeling (=Homology Modeling)

Assumption: Sequence uniquely determines structure and therefore, from similar sequence follows similar structure.

How can we use this to predict 3D structure?

Target: Protein to model
Template: Protein to model from

1. Identify Template: Query the PDB for similar sequences to your Target
2. Align Target / Template: Select the best match as **Template **and assume the Target has the same structure
3. Build Model
4. Assess Model
5. Refine Model


Question: How does Homology Modeling (Comparative Modeling) work?

Target: Protein to model
Template: Protein to model from

1. Identify Template: Query the PDB for similar sequences to your Target
2. Align Target / Template: Select the best match as **Template **and assume the Target has the same structure
3. Build Model
4. Assess Model
5. Refine Model

Question: Which tradeoff does comparative modeling face? What are the limiting factors based on PSI (Percentage Sequence Identity)?

Tradeoff: Accuracy vs Coverage

Limiting factor in homology modeling:
75% - 100% - Speed of Modeling
50% - 75% - Quality of Model
25% - 50% - Alignment Accuracy
0% - 25% - Detection of Homology

5. Comparative Modeling Methods

5.1 MODELLER

**Summary: **lots of whistles and bells, downloadable, very accurateConstraint Satisfaction: use a set of objective functions to check whether the model is plausible

• $$C_{\alpha} - C_{\alpha}$$ distance
• Molecular dynamics
• Langevin dynamics
• Rigid bodies
• Rigid molecular dynamics
• ...

Optimization Steps (run repeatedly)

• explore different local minima

Typical Errors

• side chain packing
• misalignment
• wrong template

Pick the right solution:

• DOPE score (Discrete Optimized Protein Energy)
• based on knowledge based pair potentials

Question: How to handle a missing loop in comparative modeling?

• One way would be to find similar loops and compute the average over them.
• Another solution would be to apply molecular dynamics on the loop sequence. (only for shot loops)

5.2 SWISS-Model

**Summary: **automated, increasingly comprehensive and flexible

Underlying 'Philosophy'

• fully automated
• for non-expert users / experimental biologists
• do less, make less mistakes

Original

1. alignment by BLAST / PSI-BLAST
2. copy to coordinates
3. end

Today: More complicated ...