Django Website for Prediction of Protein Secondary Structure and Polyproline Helix II

Authors: PAGEAUD Yoann ; LETOURNEUR Quentin.
Université Paris Diderot - Paris 7, France.
Date: 15-04-2017

This document explain every steps of install and configuration necessary to the project.

Prerequesites

• Install Python 3.6:

sudo apt install software-properties-common
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update
sudo apt install python3.6

• Install MySQL and dependencies:

sudo apt install mysql-client
#To run if not installed with mysql-client: sudo apt install mysql-common
sudo apt install mysql-server
sudo apt install mysql-sandbox
sudo apt install libmysqlclient-dev

• Upgrade pip:

pip install --upgrade pip

• Install MySQL-python:

sudo pip install mysqlclient
sudo python -m pip install mysqlclient

• Install Django:

sudo pip install Django
sudo pip install django-autocomplete-light
sudo pip install django-simple-menu
python -m pip install django

• Set your working directory to the git folder containing the script create_db.sql

• Launch Mysql:

mysql -u root -p

Set up Database

• Create Database:

create database M2BI_Projet_PPII;

• Leave Mysql:

quit;

• Load Tables in M2BI_Projet_PPII using the file create_db.sql;

cd Django_for_Prediction_of_Proteins_PPII/
mysql -u root -p M2BI_Projet_PPII < create_db.sql

Once tables are created you can move to the Set up of Django

Set up Django

• cd into the Django project PDBWebsite/PDBWebsite/.

• In settings.py configure project settings for Database:

Replace:

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.sqlite3',
        'NAME': os.path.join(BASE_DIR, 'db.sqlite3'),
    }
}

By:

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.mysql',
        'NAME': 'M2BI_Projet_PPII',
        'USER': 'root',
        'PASSWORD': '<your own root password>',
        'HOST': '',
        'PORT': '',
    }
}

Don't forget to replace <your own root password> by your MySQL password defined during installation.
And save modifications.

• Migrate new parameters to the database:

cd ..
python manage.py makemigrations pdbapp
python manage.py migrate

The database will be updated with the tables necessary for the communication between MySQL and the Django app.

Set up a superuser account:

• Create new Superuser account:

python manage.py createsuperuser

Follow instructions.

Launch the webserver:

• Launch server:

python manage.py runserver

• In your web browser go to localhost:8000 and then localhost:8000/admin/ to sign in with the username and password you just define.
  You can now administrate all informations in the database to be displayed on the website.

Contact

In case of any problems:
Yoann PAGEAUD: yoann.pageaud@gmail.com

Introduction

Secondary structures play a key role in proteins shape and functions. their characteristics permits us to predict and assign them to a given protein sequence based on structural informations. There are three major secondary structures, α-helix, β-sheet and turns. But there are also other patterns like Polyproline II (PPII) which are less frequent and less studied. Although we know there existence since the 50's they were thought to be only present in fibrous proteins and in low frequencies^2,9. Since then, severals studies have shown that PPII are also present in globular proteins and that they can be present even without proline in them^1. They can play a main role for interactions between proteins and there flexibility is one of their interesting trait. What is most surprising about them is that unlike other regular secondary structures they don't seem to have stabilizing interactions (1,fig. 1) . However there is no gold standard prediction yet for there assignment and they are still understudied.
For this project we built a MySQL database containing secondary structure predictions made from 2 prediction tools (DSSP and PROSS)^6,10. The database is accessible through a web interface designed with the Python^14 framework Django.
Thanks to our website, a user can search for any PDB stored into the database to compare their associated secondary structure predictions with emphasis on PPII.

Figure 1: Idealized major periodic structures: β-structure, α-helix, and PPII helix, modeled as the CA-trace helical axis projection (b-1) and perspective projection (b-2), and the 10-Ala polypeptide chain (b-3). PPII helix with n = −3 and d = 3.1 Å is a left-handed narrow and extended helix, the most extended helical structure occurring in proteins, and only slightly less extended than the β-structure. For α-helix, n = +3.6 and d = 1.5 Å; thus, PPII helix covers twice its length per residue. It does not form any regular pattern of local intra- or interchain hydrogen bonds. The CA-trace projection along the helical axis shows the PPII characteristic shape of a triangular prism (b-1).

## Materials and Methods This project has been developped under the Linux distribution **Ubuntu 16.04** and was shared in a git repository hosted on GitLab. Various packages and tools had to be installed. Install steps are explained in the **Prerequesites** part of the **README.md** file. The website is compatible with both latest **Chromium** and **Firefox** versions.

Prediction Tools

To generate predictions, two prediction tools DSSP and PROSS were used.
DSSP Predict secondary structure using mainly H-bond patterns for alpha-helix and beta-strand/bridge, C-alpha positions for bends and backbone dihedral angles for chirality.
Shown below is the description of characters used for secondary structure prediction:

• H = alpha-helix
• E = beta-strand
• B = beta-bridge
• G = 3-helix (310-helix)
• I = 5-helix (pi-helix)
• T = 3-, 4- or 5- turn
• S = bend
• P = PPII
• 'whitespace' = coil

PROSS predict secondary structure using only dihedral angles mesostates.

Figure 2: fine grain grid for mesostate assignment.For two given dihedral angles (phi and psi) of a residue the grid define its mesotate.

• H = alpha-helix
• E = beta-strand
• T = beta-turn
• P = PPII
• C = coil

After launching DSSP and PROSS on the PDBs we downloaded, informations have been extracted from the PDBs and secondary structures predictions were placed to place them in the database.

To do so we generated a bash and a python script : create_pdb_table.sh and create_sspred_table.py.
The first script extracts severals informations from a PDB file like the header, the sequence, the protein size based on the number of the last residue. To make the length of the sequence as close as possible to this number we retrieved and placed in the sequence the missing residues mentioned in the PDB file. To recognize them they were written in lowercase.

The second script extracts informations from secondary structure prediction files from DSSP or PROSS like, for example, the sequence representing the secondary structure assignment, and the computed phi and psi angles by residue.

Both tools detect missing residues (chain breaks) to avoid shift apparition between the sequence and the prediction chain. 'X' were inserted where there are missing residues.

An additional script aligne_sequence_and_prediction.sh has been made to add '-' at the start of the predictions if there is a shift in the sequence and the prediction due to misssing residues at the start of the sequence.

Database and web infrastructure

The choice has been made to use a standard MySQL^8 database better suited for web use.
We decided to create 4 different tables : one for PDB informations, one for predictions informations, and 2 others for prediction methods and resolution methods.
The detailled conceptual framework of our database is available below:

Figure 3: Database conceptual framework.

The database has been created following the steps in the **Setup Database** part in the **README.md** (for details about how to install Python, MySQL, and Django, please refer the **Prerequesites** part in the **README.md** file). Once the database has been created in MySQL, tables were created and filled with entries thanks to the **create_db.sql** file. This file contains queries necessary to the creation of the 4 tables and populate the tables **PDB** and **struct_sec** by calling 2 files: **pdb_table.csv** and **sspred_table.tsv**. **pdb_table.csv** contains all informations about the PDB files, while **sspred_table.tsv** contains all informations about the predictions generated with DSSP and PROSS. After the database has been populated, We had to found a way to access it easily if any further modification would be needed. For this purpose, the choice has been made to use **Django^13**.

Figure 4: Django M.V.T model.

Django is a widely used Python framework thought for developping websites easily with short deadlines. It has made its prooves since it has been used to developped websites of renowed institutions like Instagram, Pinterest or NASA. It's based on the **"Model-View-Template" (M.V.T.)** model (**Fig. 4**) which is slightly different from the common "Model-View-Controller" (M.V.C.) model because Django manages the "Controller" part on its own. It also integrate modules simplifying the creation and configuration of an administration interface to access database entries easily and make small modifications on some of them in case of minor bugs. Moreover, Django is working with its own Python functions to access the database. Administrator can use them to access the database by a **Django shell**. Thus, it has also been used to generate a **models.py** file defining models associated to the MySQL database tables (one class per table) as Python classes, and Django functions can be called in any views contained in the **views.py** file. For all these reasons Django perfectly suited our website project. Using Django, a Django project **PDBWebsite/**:

django-admin startproject PDBWebsite

and a Django application pdbapp/ have been created:

cd PDBWebsite/
python manage.py startapp pdbapp

Multiple applications can be generated in a Django project, following the different needs. For our project, we decided to create only one application pdbapp/ to answer to one biological question.
The structure of a Django project follows some standards defined by the Django Project initiative (Fig. 5).

Figure 5: Structure of the Django project PDBWebsite.

The views "home","pdbinfo","strucinfo","about","detail" and "strdetail" were created in the file views.py. In Django each view is associated to one corresponding HTML template.
The corresponding HTML templates "home.html", "pdbinfo.html", "strucinfo.html", "about.html", "pdbstat.html" and "strucstat.html" were created.
Views and templates are linked together by URLs in the files urls.py. The Django project has a urls.py file and each Django application has its own urls.py.
URLs are listed as followed in the application pdbapp/:

app_name = 'pdbapp'

urlpatterns = [
    url(r'^home/$', views.home, name='home'),
    url(r'^pdbinfo/$', views.pdbinfo, name='pdbinfo'),
    url(r'^strucinfo/$', views.strucinfo, name='strucinfo'),
    url(r'^about/$', views.about, name='about'),
    url(r'^(?P<id_pdb_chain>.{5})/$', views.detail, name='detail'),
    url(r'^(?P<id_struct_sec>[0-9]+)/$', views.strdetail, name='strdetail')
]

and in the project PDBWebsite/:

from pdbapp.views import *

urlpatterns = [
	url(r'^$', home, name='home'),
	url(r'^pdbapp/', include('pdbapp.urls')),
	url(r'^admin/', admin.site.urls)
]

In the views.py file each view return a context containing all variables defined in the view that is called in the corresponding HTML template.
Usually, variables refer to corresponding queries in the database.
HTML templates can host each variable by calling them as followed:

{{ variable }}

If the variable is iterative, a Python for loop with if and else conditions can also be used inside the template. It allows to display any query results dynamically on a web page.

To add some styles to HTML templates, CSS files have been created. All CSS files are stored into a static/pdbapp/ directory. A static/pdbapp/images/ directory has also been created to store all images to be displayed on the web site.

To make the tables generated dynamically, on the page pdbinfo.html and strucinfo.html more user-friendly, javascript files are also stored in the static/pdbapp/ directory. 2 javascript files were used:

• jquery.searchable-1.0.0.min.js^11 to enable the search options.
• jquery.tablesorter.min.js^4 to enable the sorting options.

Both scripts call functions defined in a well known javascript library: JQuery^5 accessible in the file jquery-3.2.0.min.js.

To sum up everything about HTML display: Javascripts files call functions from JQuery.
CSS files call images from the static/pdbapp/images/ directory.
HTML templates call:

• views contained in views.py to display queries results,
• CSS files to get the styles parameters.
• javascript files to enable user actions to search or sort entries.

The website can be navigated thanks to a menu bar on the top of each page.
6 HTML templates were made:

• home.html: this page is an introduction to the website, The biological issue raised, and how it answers to it. Examples of use are also described.
• pdbinfo.html: this page contain a table displaying all entries from the PDB table in the database. Rows can be sorted and filtered, and each PDB Id send to a page displaying every details about a specific PDB.
• strucinfo.html: this page work exactly the same way than the PDBs page, except that from the table, you can access both PDBs details pages and predictions details pages.
• about.html: this page dynamically display interesting statistics about the database, and contains also authors informations.
• pdbstat.html: this page contains PDB details for one PDB. It has an URL ending with the PDB Id concatenated to the PDB chain name.
• strucstat.html: this page contains Predictions details for one prediction. It has an URL ending by an integer corresponding to the prediction id.

Both pdbstat.html and strucstat.html pages contain links that simplifies the navigation between a PDB and its corresponding predictions.
Furthermore, additional options have been added to the PDB details pages to have access to the corresponding RCSB^3 page, to be able to download the PDB file, and to have a quick access to the online tool RAMPAGE^7 which generate multiple Ramachandran plots, one especially representing the distribution of Proline-favorable regions.

A menu bar has been added to the top of each HTML templates so that all pages are easily accessible.

Every details relative to webpages, are available on the Home page when starting the web server in the How to use the website ? part.

Alongside the user accessible part of our website, an administration interface^12 have been configured (to access the admin interface you need to create a superuser account, see details in the README file for more informations).

Figure 6: Administration main page.

Access to entries in all database tables have been made available, with the posibility to remove entries or to modify them. Work groups and users list was already available from the default Django parameters. An history of modifications has also been added (**Fig. 6**). In the admin pages **Pdbs** and **Struct secs** entries have been displayed in tables. they can be accessed by a search bar or by sorting tables rows on both pages. Additionally, the **Pdbs** admin page entries can be filtered following their chain name, and/or the resolution method used to generate the PDBs (**Fig. 7**), while **Struct secs** admin page entries can be filtered following the method of prediction (DSSP or PROSS).

Figure 7: Administration page for PDBs.

Settings of the administration part of the website rely on many different files. Most of them can be found in the **admin.py** file in **pdbapp/**. Each table is defined in it as a Python class and the parameters to access entries in administration tables are defined under 3 differents standard variables:

• list_display: it contains all table fields that should be displayed.
• list_filter: it contains all table fields to be used for filtering.
• search_fields: it contains all table fields to be searched using the search bar.

To modify or delete any entry, just click it, the administrator is redirected to a modification page with all fields of the entry accessible.
The user accessible part of the website has also been made accessible by clicking the link VIEW SITE in the top right corner of the interface, and redirects the administrator to the home page.

Discussion

We found that PROSS was better than DSSP when it comes to PPII detection. This is due to the difference in there way of assigning secondary structure. PROSS is based only on dihedral angles and DSSP mainly on H-bond. Since there are few/none H-bonds in PPII, it may be more difficult for DSSP to detect them. But for other secondary structures, DSSP gives more detailed results because for example it differentiate between different types of helix while DSSP don't.

Using Django framework to develop a website confer many advantages (listed in the Materials and Methods part above) and give convenient tools to easily access the database and make modifications on a website.
With Django, once the web server is launched, you do not need any interruption for maintenance of the website. Moreover, There are now 3 different ways to access the database, modify, delete and create entries:

• The normal way by MySQL in the Terminal.
• The Django way using Django shell (> python manage.py shell) with Django's functions.
• The Django-admin way using the administration interface to make minor modifications on entries.

The Display of Predictions just below a PDB sequence on the PDB details pages, make the interpretation of secondary structure prediction more easy.
Supplementary tools like links to download PDB, display RCSB page, or access RAMPAGE, gave more options to the user to explore any other subsisting questions to which the website does not answer. Next improvments to be added could be a display of a ramachandran plot, for each prediction, coloration of structure prediction, and the support of more PDBs and more precition tools.

Conclusion

Using actual technologies from database to web interface, we made predictions for secondary structure and PPIIs available to a large public to answer.
PROSS and DSSP predictions are stored in a MySQL database and available to any user on the dedicated webpages. MySQL database containing secondary structure predictions made by PROSS and DSSP connected to a web interface using the python framework Django which has been a great help in our project. To able easy detection of PPII and comparison between softwares we designed pages containing the sequence and under its corresponding secondary structure predictions. The database can be search by multiple criteria like the PDB_ID and key words in the protein name. This project can still be improved on many different ways.

References

1- Chebrek, Romain, Sylvain Leonard, Alexandre G. de Brevern, and Jean-Christophe Gelly. “PolyprOnline: Polyproline Helix II and Secondary Structure Assignment Database.” Database: The Journal of Biological Databases and Curation 2014 (November 6, 2014). doi:10.1093/database/bau102.
2- Cowan, Pauline M., Stewart McGAVIN, and A. C. T. North. “The Polypeptide Chain Configuration of Collagen.” Nature 176, no. 4492 (décembre 1955): 1062–64. doi:10.1038/1761062a0.
3- Deshpande, N. “The RCSB Protein Data Bank: A Redesigned Query System and Relational Database Based on the mmCIF Schema.” Nucleic Acids Research 33, no. Database issue (December 17, 2004): D233–37. doi:10.1093/nar/gki057.
4- “jQuery Plugin: Tablesorter 2.0.” Accessed April 10, 2017. http://tablesorter.com/docs/.
5- jquery.org, jquery Foundation-. “jQuery.” Accessed April 10, 2017. https://jquery.com/.
6- Kabsch, Wolfgang, and Christian Sander. “Dictionary of Protein Secondary Structure: Pattern Recognition of Hydrogen-Bonded and Geometrical Features.” Biopolymers 22, no. 12 (décembre 1983): 2577–2637. doi:10.1002/bip.360221211.
7- Lovell, Simon C., Ian W. Davis, W. Bryan Arendall, Paul IW de Bakker, J. Michael Word, Michael G. Prisant, Jane S. Richardson, and David C. Richardson. “Structure Validation by Cα Geometry: Φ, ψ and Cβ Deviation.” Proteins: Structure, Function, and Bioinformatics 50, no. 3 (2003): 437–450.
8- “MySQL.” Accessed April 10, 2017. https://www.mysql.com/.
9- Pauling, Linus, and Robert B. Corey. “The Structure of Fibrous Proteins of the Collagen-Gelatin Group.” Proceedings of the National Academy of Sciences of the United States of America 37, no. 5 (May 1951): 272–81.
10- “PROSS: Dihedral Angle-Based Secondary Structure Assignment.” Accessed April 17, 2017. http://folding.chemistry.msstate.edu/utils/pross.html.
11- “Stidges/jquery-Searchable.” GitHub. Accessed April 10, 2017. https://github.com/stidges/jquery-searchable.
12- “The Django Admin Site | Django Documentation | Django.” Accessed April 10, 2017. https://docs.djangoproject.com/en/1.11/ref/contrib/admin/.
13- “The Web Framework for Perfectionists with Deadlines | Django.” Accessed April 10, 2017. https://www.djangoproject.com/.
14- “Welcome to Python.org.” Python.org. Accessed April 10, 2017. https://www.python.org/.

Encountered difficulties

Quentin:
There may be others but I saw that there is a problem with the prediction of the 1tke.pdb file with DSSP that take into account an heteroatom at the end of the prediction. We could have manually modified the errors in this entry using the Django administration interface.

Yoann:
I had some difficulties in understanding how Django worked since I never developped any website before. It also means understanding how HTML, Python, CSS and Javascript join together to build a functionnal webpage. It was my choice to work on this part, and in the end I really learnt a lot, it was definitely worth it even if I did not had enough time to do everything I wanted.
I also tried to use CanvasJS to generate ramachandran plots from the psi and phi angles, but without any tutorial available I didn't success in this step. I should have also colored the PPII structures in predictions using javascript, but I didn't had enough time to try anything.

Tasks distribution

Project report: Quentin & Yoann
Database conception: Quentin & Yoann
Predictions generation: Quentin
Website conception: Yoann
README: Yoann