Skip to content

Using Deep Learning to predict properties of Chemicals

Notifications You must be signed in to change notification settings

evijit/DeepChem

Repository files navigation

DeepChem

Using Deep Learning to predict properties of Chemicals from their 3D structures. This is one of the projects in KWoC 2017!

Discussion:

Registered Research Project :

Fig1

Objective:

To Predict a difficult to obtain property of a compound using only the 3D structure. The property that I have selected for the initial proof-of-concept is Solubility.

Methodology:

The first phase of the project will consist of data collection. For this, we will be generating a huge database of 3D structures vs Solubility.

For this purpose, there is a Handbook of solubility which the student will be using to obtain the solubility. The 3D structure will be parsed from SDF files, which can be obtained from NIST: Benzene SDF

The second phase of the project will be to try and use the 3D structure to predict the solubility. For this, I aim to apply Vector Models to convert the 3D SDF to a high dimensional vector, and then build a Deep Neural Network to predict the solubility.

© 2017 Avijit Ghosh for the Department of Chemical Engineering, IIT Kharagpur.