Discovering Rational Properties of Chemical Molecules by Rationalization and Use of Causal View in AI

Please enable javascript in your browser.

Saeidi Kelishami, Amin | 2024

0 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 57279 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Hossein Khalaj, Babak
Abstract:
The exploration of chemical molecule properties and their underlying reasons, along with the interpretability of machine learning models and artificial intelligence, constitutes a pivotal concern in their respective domains. Numerous research endeavors have been undertaken to address these challenges. Uncovering the existence and rationales behind the chemical properties in molecules enhances our comprehension of diverse substances and instills greater confidence. Among various algorithms and computational approaches, graph neural networks emerge as a significant solution for modeling and computation in molecular research. Moreover, these networks offer modern interpretability methods. This research focuses on investigating the rational properties of chemical molecules through the application of rationalization in graph neural networks. Additionally, our approach seeks to devise a generalized method by employing different splitting approaches. Benchmarking is a core objective, and our primary aim is to reduce the black-box nature of machine learning models
Keywords:
Chemical Properties ; Interpretability ; Graph Neural Network ; Rationalization ; Artificial Intelligence ; Machine Learning