School of Engineering Department of Computer Science and Engineering 120 Graph Machine Learning Methods for Scientific Discovery Supervisor: SONG Yangqiu / CSE Student: CHEN Rulin / MATH-GM Course: UROP 1100, Summer Graph machine learning methods are important for scientific research since using the structure graph can help to analyze more complex relationships and patterns within the data. For chemistry, as a common structure to represent molecules or reactions, graph can easily store the features and make predictions to atoms (node-level tasks), bonds or the relationships between reactants and products (edge-level tasks), and the whole molecules or reactions (graph-level tasks). This research focuses on the application of graph machine learning methods in predicting some kinetic properties like the activation energy (Ea) or the rate constant (k) of some chemical reactions, which is a graph-level task of graph machine learning methods or graph neural networks (GNNs) and each reaction will be regarded as an input to GNNs models. Therefore, a proper way to abstract the chemical reactions to graphs, including the featurization of nodes and edges, is essential to this research. Graph Machine Learning Methods for Scientific Discovery Supervisor: SONG Yangqiu / CSE Student: CHUA Shawn Darren Siytiu / DSCT Course: UROP 1100, Spring Graph Neural Networks (GNNs) have emerged as a powerful tool for representing and analyzing graphstructured data, with applications in various domains, including chemistry. This report presents the work conducted by a student at the Hong Kong University of Science and Technology (HKUST) as part of the Undergraduate Research Opportunities Program (UROP) on GNNs, with a focus on Atom-to-Atom Matching (AAM) in chemical reactions. The research involved studying state-of-the-art methods, notably RXNMapper, a tool which uses transformers and attention-guided maps for AAM. Emphasis was placed on exploring techniques for extracting and mapping token embeddings. Additionally, alternative approaches such as combinatorial algorithms and optimal transport were considered. Graph Machine Learning Methods for Scientific Discovery Supervisor: SONG Yangqiu / CSE Student: HUANG Hao / COMP Course: UROP 1100, Summer This report investigates the machine learning method of ReactionOOD, with a specific focus on the GOOD (Graph Object-Object Directories) dataset and the Chemprop model. The investigation aims to enhance the understanding of chemical reactivity using advanced computational tools. By exploring the intricacies of these components, this study focuses on the intersection of computational chemistry and machine learning. Through the exploration of the GOOD dataset and the Chemprop model, this report showcases the potential of interdisciplinary collaboration in advancing reactivity understanding and shaping the future of computational chemistry and machine learning in the realm of chemical informatics.
RkJQdWJsaXNoZXIy NDk5Njg=