Molecular structure of Bio-Molecules possess valuable information about their properties such as inhibition / binding capacity to specific receptors, solubility in water or other liquids etc. The molecular structure can also be represented in the form of Graph data-structures for ease of storage & computation which also provides additional insights into how the various atoms in a molecule interact with each other.
SMILES (simplified molecular-input line-entry system) is used to describe structure of molecules in short strings. This project takes advantage of SMILES nomencalture of molecules to obtain graphical representations of respective molecular structure & estimate relevant molecular properties
This project explores Deep Learning techniques specifically Graph Neural Networks & associated algorithms to assess molecular properties from the Graphical Representations of molecules, which are derived from their respective SMILES representations.
The prediction tasks explored in this project are as follows:
-
Estimate the binding capacity of inhibitors of human β-secretase 1(BACE-1) from the corresponding inhibitor molecular structure, as in - BACE_GNN.ipynb.
-
Estimate Water Solubility from molecular structure as explored in - ESOL_GNN.ipynb
-
Predict the Hydration Free Energy from respective moleccular structure as performed in - FreeSolv_GNN.ipynb