MacroFlexEngine

by Natalia Pattarone, Joan Térmens, Miguel Luengo

1. Introduction

MacroFlexEngine is a python application that allows the construction of protein and dna macrocomplexes from pdb information of the dimers that form this complex. Furthermore if a pir file of the whole complex is inputed it will detect which parts of the complex are given and which are missing and complete the missing parts using modeller.

2. Running the application

2.1 Requirements

The application requires Python 3 to be able to run as well as several python modules that will be installed automatically with the setup.

Biopython 1.73
setuptools 1.16.2
numpy
copy
os
sys
re
pylab

The application also uses the modeller software which can be installed from this page.

2.2 Installation

The application can be installed by running the following commands: For linux distributions and Mac ios:

python setup.py sdist
python setup.py install

For windows:

python setup.py bdist_wininst

2.3 Execution

To execute the application the following command can be run

python MFEngine-launch -i absolute_path_to_pdb_folder

As well as the mandatory input flag there are several optional ones:

-o –output: The output file containing the final complex 
-v verbose: More detailed process information
-s -seq: folder with sequence files in pir format for uncompleted models
-p -profile: Show the energy profile of the output model 

By default the program will output a pdb file with the whole macrocomplex. If a folder is specified the output will be saved there.

3. Theoretical Background

The main approach for the building of the complex used in this application is superimposition. The program finds the identical chains in the input dimers by alignment based on their identity value, superimposes them and then moves the remaining chain of the dimer. This way the complex is formed. The program ends when no more chains can be added because of clashes. This means that this program does not need the stoichometry as an input since it will add chains until all the space is explored.

3.1 Data treatment

First of all the input dimer pdbs are read and the chains and interactions are saved separately. With these variables saved they can now be compared to the pir file if it is inputed to check if there is a missing chain.

3.2 FlexEngineModeller (using Modeller)

In cases where there are chains missing, we developed an Engine supported on Modeller to obtain and complete those structures missing a complete PDB. The modelling process are as follow:

1. Search for potentially related sequences for known structures, utilizing a non-redundant PDB file with sequences at 95% sequence identity from which it creates a bin file
2. Creates a Modeller profile (which are compacter and better for sequence database searching) containing similar information as alignments do
3. Searches the PDB database bin file and adds the matches to the profile
4. Builds the profile using the query sequence and the homologue candidates and selects the best templates.

Note: The PDB code in each line of the profile file built is the representative of a group of PDB sequences that share 95% or more sequence identity to each other and have less than 30 residues or 30% sequence length difference. One of the column reports the percentage sequence identities between the query sequence and a PDB sequence normalized by the lengths of the alignment (indicated in the previous column). In general, a sequence identity value above approximately 25% indicates a potential template unless the alignment is short (i.e., less than 100 residues). A better measure of the significance of the alignment is given in the another column by the e-value of the alignment, in which we based our criteria selection.

6. To assess a better template selection we compare structural and sequence similarity between the possible templates, using malign3d command then performs an iterative least-squares superposition of the 3D structures, using the multiple sequence alignment as its starting point. It does not make an alignment, but it calculates the RMS and DRMS deviations between atomic positions and distances, differences between the mainchain and sidechain dihedral angles, percentage sequence identities, and several other measures. Nextly, creates a dendrogram which calculates a clustering tree from the input matrix of pairwise distances, helping visualize differences among the template candidates.

Dedrogram example

7. It requires the user a selection from the dendrogram and table created, for further assessment on template selection

8. Then, performs the alignment with the target sequece and the template based on a dynamic programming algorithm, different from standard sequence-sequence alignment methods as it takes into account structural information from the template when constructing an alignment.

Note: This task is achieved through a variable gap penalty function that tends to place gaps in solvent exposed and curved regions, outside secondary structure segments, and between two positions that are close in space. As a result, the alignment errors are reduced by approximately one third relative to those that occur with standard sequence alignment techniques. This improvement becomes more important as the similarity between the sequences decreases and the number of gaps increases

9. Finally, creates (5 in our case) a 3D model of the target automatically, outputig a log file which reports warnings, errors and other useful information including the input restraints used for modeling that remain violated in the final model.

3.3 Alignment

The alignment process consists of two steps to know which chains of the two structures must be superimposed. In the first place a pairwise sequence alignment of all of the chains is done. If the identity between two chains is higher than a certain score it is considered an homolog and will be superimposed. The identity is calculated with a score of 1 for matches and a value of 0 otherwise with no special gap penalty. Afterwards when the structural alignment for the superimposition is carried out the rmsd is calculated and if the normalised value is lower than 1 Å the chain is discarded and the following is tested. The normalization is done considering 100 residues as in Caraguo, 2001:

Where RMSD₁₀₀ is the normalized RMSD to 100 redues and N is the number of residues of the shortest chain.

3.4 Superimposition

The superimposition is the main part of the program. The process will start with two input dimers. The common chain found in the alignment process will be superimposed and the remaining chain will be moved to form a complex with three chains. Then the process will be repeated with the complex that has been created and another one of the inputed dimers. Each time a new chain is added to the model the program checks if it causes a clash. If there are no clashes the chain is added. This allows the formation of macro-complexes with many repeating units without the need of a stoichometry and the program will always input the most complete protein possible. The algorithm for the clash calculation is explained below.

3.5 Clash Checking

Before moving a chain and adding it to the model the program will check if the new position creates a clash with the existing structure. This is done by calculating the distance betweeb the center of mass of the chain being added and its furthest atom and an imaginary sphere is drawn between the two points plus two Å. Then, the distance between each of the atoms from the other chains that are within the sphere and the atoms from the moved chain will be calculated. If the distance between two of these atoms is greater than 1.1 Å will be considered a clash. If more than 10 clashes are found for a specific chain being added to the complex this chain will be discarded and the following one will be evaluated. If there are no more chains to be evaluated the program ends.

3.6 Output Analysis

Once the superimposition process is complete and a final complex has been found, modeller can be used to evaluate the model. If the flag is added to the input the program will calculate the energy profile of the resulting model using the Discrete Optimized Protein Energy (DOPE) method.

4. Examples

RNA polymerase with promoter DNA and transcription activator

PDB id: 5nss

This is an example of a complex protein structure with dna interaction. It has 10 protein chains and a DNA double helix. The input for this complex is the chains that form the complex separated in dimers. The execution time for this complex is around 40 seconds. The model is represented in light brown and the original pdb in light blue.The result is the following.The rmsd was normalised using the formula developed by Carugo and Pongor (2001). The normalised rmsd is calculated as follows, rmsd_100 = rmsd / 1+ln(sqrt(N/100)) where the N is the length of the shortest of the chains.

Fig. 1: 5nss model with original complex. RMSD₁₀₀:0

This result was obtained without modifying the input chains of the intial complex but even when they were moved and rotated randomly the program was still able to obtain very good results with a very low rmsd. The following figure shows that even with added noise the algorithm is able to replicate the reference complex perfectly.

Fig. 2: 5nss model with original rotated chains. RMSD₁₀₀:0,001

Virus Capsid

Example of virus capsid built from the repetition of few input chains and their interactions. Here we can see one of the strengths of the program. With only a few chains a macro complex can be formed by adding the same structure until a clash happens when the sphere is already completed.

Fig. 3: Virus capsid model

Microtubule

Example of protein microtubule. The original model obtained is very large(around 670000 atoms) so the chimera software could not handle it. The version shown is a segment of the whole complex. Similarly to the virus model this large complex is formed from only a few initial dimers.

Fig. 3: Fragment of the microtubul model

About this document

This document can be compiled into a PDF with the following command:

pandoc -f gfm -t latex                  \
       --variable papersize:a4paper     \
       --variable geometry:margin=1.5cm \
       --variable fontsize=11pt         \
       --highlight-style pygments       \
       -o report2.pdf	\
          report2.md

References

• Carugo O, Pongor S. A normalized root-mean-square distance for comparing protein three-dimensional structures. Protein Sci. 2001;10(7):1470–1473. doi: 10.1110/ps.690101