There is a variety of public resources on the Internet which contain information about various aspects of chemical, biological and pharmaceutical domains. The quality, maturity, hosting organizations, team sizes behind these data resources vary wildly and as a consequence content cannot be always trusted and the effort of extracting information and preparing it for reuse is repeated again and again at various levels. This problem is especially serious in applications for QSAR, QSPR and QNAR modeling. On the other hand authors of this poster believe, based on their own extensive experience building various types of chemical, analytical and biological databases for decades, that the process of building such knowledgebase can be systematically described and automated. This poster will outline the work performed on text and data-mining various public resources on the Web, data curation process and making this information publicly available through a portal and a RESTful API. We will also demonstrate how such knowledgebase can be used for real-time QSAR and QSPR predictions.

1. Using publicly available resources
to build a comprehensive knowledgebase of chemical information
by B. Sattarov, R. Zakharov and V.Tkachenko
Science Data Software
Abstract
There is a variety of public resources on the Internet which contain information about
various aspects of chemical, biological and pharmaceutical domains. The quality,
maturity, hosting organizations, team sizes behind these data resources vary wildly and
as a consequence content cannot be always trusted and the effort of extracting
information and preparing it for reuse is repeated again and again at various levels.
This problem is especially serious in applications for QSAR, QSPR and QNAR modeling.
On the other hand authors of this poster believe, based on their own extensive
experience building various types of chemical, analytical and biological databases for
decades, that the process of building such knowledgebase can be systematically
described and automated tool for building a comprehensive knowledgebase of
chemical information.
We have developed data mining workflow to collect and standardize chemical data
from open sources, using several simple python scripts which will be included in open
source library. Data collection was carried out by HTML parsing and by using
ChemSpider API. We also used python version of Chemical Validation and
Standardization Platform developed by us to standardize collected data.
Our ChemScrapper allowed us to resolve 19.85% names of
biologically active compounds from MESH 2017 dataset and to save
this data into json and handy sdf format.
Chemical Validation and Standardization Platform
(CVSP),
which we used to standardize chemical structures, can also be used as
stand-alone platform for SMIRKS-based standardization of any dataset,
thanks to the visual implementation of its python version functionality on
Jupyter.
You can see every standardization
rule applied as SMIRKS string simply
by clicking on SMIRKS button as
well as download standardized
dataset as *.sdf file by checking
corresponding folder.
Example json—output with mol block, properties and synonyms
Example CLI
Example input
One of the most productive data mining tools we have created works with ChemSpider web
API. It allows user, looking for a chemical structures/data, to deal only with convenient com-
mand line interface written in Python, in order to resolve chemicals names and identifiers or
to find new data for QSAR/QSPR analysis or any other purpose that requires .
API
HTML Parsing
CVSP
Standardization
Data collection
Comprehensive
knowledgebase of
chemical information
Open Science Data Repository (OSDR)
Comprehensive distributed semantic knowledgebase of scientific information
with built-in Machine Learning capabilities