Deep Learning Framework for Graph-Based Molecular Drug Discovery

Author

Daniel ManuFollow

Publication Date

Summer 6-24-2024

Abstract

This dissertation proposes the GraphGANFed framework, which combines Graph Convolutional Networks (GCN), Generative Adversarial Networks (GAN), and Federated Learning (FL) to generate novel molecules while preserving data privacy. FL allows learning from distributed clients without sharing local datasets, GCN extracts molecular structural properties, and GAN generates new molecules retaining the learned properties. Extensive simulations on three benchmark datasets show GraphGANFed's effectiveness, producing molecules with high novelty (≈ 100) and diversity (> 0.9). Results indicate a trade-off among evaluation metrics, and the right dropout ratio avoids mode collapse. Additionally, Conditional GraphGANFed (cGraphGANFed) incorporates a critic network from Reinforcement Learning (RL) to meet specific client requirements. cGraphGANFed can optimize for particular metrics, producing molecules with higher values in specified metrics or achieving high values across multiple metrics, which may be interdependent. Extensive simulations confirm cGraphGANFed’s efficacy in optimizing specific client objectives and generating molecules with high performance across multiple metrics.

Keywords

Generative adversarial networks, Graph convolutional networks, Federated learning, Reinforcement learning

Sponsors

National Science Foundation under Award under grant no. CNS-2323050 and CNS-2148178, where CNS-2148178 is supported in part by funds from federal agency and industry partners as specified in the Resilient & Intelligent NextG Systems (RINGS) program

Document Type

Dissertation

Language

English

Degree Name

Computer Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Dr. Xiang Sun

Second Committee Member

Dr. Marios Pattichis

Third Committee Member

Dr. Eirini Eleni Tsiropoulou

Fourth Committee Member

Dr. Chonggang Wang

Fifth Committee Member

Dr. Luan Shuang

Recommended Citation

Manu, Daniel. "Deep Learning Framework for Graph-Based Molecular Drug Discovery." (2024). https://digitalrepository.unm.edu/ece_etds/657