https://github.com/aalto-ics-kepaco/onto-chem-compounds

Predicting the ontology of chemical compounds

https://github.com/aalto-ics-kepaco/onto-chem-compounds

Repository

Predicting the ontology of chemical compounds

https://github.com/aalto-ics-kepaco/onto-chem-compounds/blob/master/

onto-chem-compounds
===================

**About**
--------------------
Code for classification of a molecular species in an ontology from molecular fingerprints.  Fingerprints are used to represent chemical compounds, and for each of the compounds there exists an ontological classification in form of a directed acyclic graph (DAG). 

**Methods**
----------------------
We use two methods in this project. First we use structured prediction method developed for network response named [SPIN (Structured Prediction of Network Response)](http://jmlr.org/proceedings/papers/v32/su14.pdf). The input is a set of fingerprint vectors of some chemical compounds and the output is a DAG.  Then we use SVM as a single target classifier. 

**Data preparation**
------------------------
The ontology data are downloaded from [ChEBI](http://www.ebi.ac.uk/chebi/aboutChebiForward.do). We have the fingerprints for all the compounds in PubChem. We use 1141 compounds from the [metlin ontology](metlin_onto/) (ask author for complete directory) for which we already have the outputs (DAGs). 

To get the input fingerprint vectors, several mappings are needed: first PubChem substance Id is retrieved for each CHEBI Id from the meta-info file, [CHEBI_complete.sdf](http://www.ebi.ac.uk/chebi/downloadsForward.do).  For the compound with id  4194 I couldn't find links to Pubchem substance Id, neither from the meta-info file nor from the CHEBI database, therefore I use 1140 compounds for the classification task. 

Next, PubChem compound Id can be retrieved from the PubChem substance Id by using the [service](https://pubchem.ncbi.nlm.nih.gov/idexchange/idexchange.cgi) provided by PubChem. The file  [all_ids_sorted](onto/all_ids_sorted) contains all the ids; CHEBI Id, PubChem  substance Id and PubChem compound Id.

From here, by using the PubChem compound Ids we can retrieve the fingerprint vectors, there are [openbabel](onto/onto.babel.phix) 
fingerprints (FP3, FP4 and MACCS) all together 528 bits and [PubChem](onto/onto.pubchem.phix) fingerprints (998 bits). 

**Contents**
-------------

* README.md 

    This file  

* classification 

    A directory containing the matlab scripts for running SPIN and SVM. 
    
* metlin_onto 

    The folder contains the sub-ontology (Each compound ontology will be a subset of the underlying ontology) 
    for 1141 compounds in metlin ontology.    
    
* onto

    The directory contains the input data used for the classification task and the scripts to generate the data. 

Owner

GitHub Events