gacVAE: a multi-approach model to propose new drug candidates for Non-Small Cell Lung Cancer

Non-Small Cell Lung Cancer


NSCLC is one of the fatal cancers that is well studied and is subjected to many drug discovery efforts, yet consisting of a majority of lung cancer cases and deaths. Large molecular search space and complexity of molecular structures have always been challenging.


Here we combined the QSAR method and genetic algorithm to elucidate the molecular search space. NCI60 dataset of the Growth Inhibition effect of more than 50 thousand molecules is encoded as 166-bit MACCS fingerprints using molecular descriptors. Subsequently, a cVAE network is configured to decipher the structure-activity relationship between MACCS FP arrays and their corresponding inhibitory effect on NSCLC cells. Once the cVAE network is configured and optimized, hidden space is searched effectively using GA, and newly generated molecules, corresponding to optimal hidden coordinates are compared to ones existing in the PubChem database based on their similarity and, three molecules are proposed as new drug candidates for NSCLC.

Results and Conclusion.

Results indicate that proposed molecules are shown to be bioactive and exhibiting growth inhibitory behaviors, affirming desired performance of the network.

0 0 votes
Your rating to this post
Inline Feedbacks
View all comments