FiltRest3D help

Filtering protein models by fuzzy restraints

Server   Documentation   Example   Queue status   Download standalone version

Guide to ranking protein models with FiltRest3D web server

Introduction

FiltRest3D allows to find 3D protein models which best match experimental restraints. It scales well and could be applied to hundreds of thousands of homology models or de novo decoys.

Processing of your data proceeds in three steps:

  1. Packing your data for upload (needed only for the huge number of decoys).
  2. Entering your restraints.
  3. Interpreting the ranking given by FiltRest3D

Features

Restraint types available in the current version are:

Data preparation

If you have just few models, you can upload PDB files directly:

Screenshot of file input table

If you have many decoys, it is advised to pack your PDB files into .ZIP file,

bash$ zip /tmp/PVI.zip PvuII/*.pdb

and download whole file. Then one may use pre-filled form with restraints to see the results.

Entering restraints

Distance restraints

Distance restraints may be for example derived from crosslinking experiments. Let us assume 22 Å long cross-linking agent, so we assume that C-β atoms of bound residues are at most 24 Å apart. Below is an example restraint between surface lysines 78 and 118.

Screenshot of distance restraints table
It is not strictly necessary to enter residue names here, but it is a recommended habit to check that residues in all models are properly numbered. It is not uncommon to slightly modify sequence during modelling procedure.)
Atom name is a PDB atom code inside residue. PDB structure format demands that there is only one atom with each name inside any residue.
Some modelling programs may omit non-backbone atoms or place them incorrectly. It is recommended to measure expected distance between backbone (C-α or C-β) atoms and the constrained atom to derive restraint on the backbone (C-α or C-β) atoms. Another solution is to use rotamer optimization programs to refine models.

Restraint weights

If experimental evidence for particular restraint is weak and uncertain, it is possible to assign a weight <1 to this restraint with a weight field. When particular restraint is strongly supported by experiments (like active center of the protein with a known biochemical function), the restraint weight above 1 may be used.

Secondary structure restraints

Partial knowledge of secondary structure motifs can be entered using secondary structure restraints. Below is simple restraint preferring models with β-strand from 18th to 24th residue.

Secondary structure entry table

Aminoacid burial or exposition to solvent

Assuming that aminoacid is known to be on a protein surface, we can use crude relative solvent accessibility estimation above 30% for a residue 67 that we believe is fully exposed or below 5% for a residue 98 that is fully buried.

burial restraints entry table

Comparing aminoacid burial with Sable prediction

When most decoys contain non-protein-like conformations, then it may help to compare aminoacid burial scores in the model with a prediction from Sable solvation prediction method. Usually it is switched off, as other restraints provide enough discriminatory data. sable burial comparison switch

Knots in proteins

As far as it is known today, real protein structures do not have topological knots except for one protein family, but it is common for modelling programs to produce such knotted models. This option is time consuming but allows to eliminate all these knotted protein models.

knots in proteins switch
Rosetta eliminates these decoys using internal filters. Please switch this option off, when ranking Robetta decoys.

Output interpretation

After ranking the data you are given output like the screenshot below. The most important is the table of all decoys with filename in the first column and total score in the second. Following columns contain decomposition of scores to individual restraints. When restraint is satisfied, then the score is 0.0, when it is completely violated (or cannot be checked because of incomplete data), the value of 100.0 is assigned.

output with important sections marked in red

To facilitate error-checking, the web server also includes full copy of standard output and standard error streams from FiltRest3D engine. Standard output contains the table of results in Comma Separated Values format with colon ':' used as a record separator. It can be easily parsed or imported into spreadsheet. Standard error contains all errors and warnings, including these which cause assignment of maximum ranking penalties because of missing data.

Examples of methods for model-building with the use of spatial restraints

MolProbity is a web server recommended as a complementary resource for testing models for the presence of high-resolution features.

Relevant papers


Developed by Marta Kaczor, Anastasia Yu. Bakulina and MichaƂ J. Gajda<mgajda@genesilico.pl>

Currently maintained by Michal J. Gajda <mgajda@genesilico.pl>.