School of Science Department of Chemistry 6 Data Analytics of Homogeneous Transition Metal Catalyzed Reactions Supervisor: SU Haibin / CHEM Student: LI Changwen / CHEM Course: UROP 1100, Fall This UROP project focuses on constructing a citation network to support our team's research paper on homogeneous nickel catalyzed (HoNiCa) reactions. Using Python-based tools for graphic analysis, such as main path analysis, we aim to gain deeper insights into the relationships among our team’s reference papers in the field and track knowledge flows in the HoNiCa reaction field. Cytoscape is the primary tool used for building and visualizing the citation network to show papers relationships. By implementing these methods, we obtained an overview of our team’s reference papers, and have a deeper understanding of the research focus and popular topics. Data Analytics of Homogeneous Transition Metal Catalyzed Reactions Supervisor: SU Haibin / CHEM Student: WONG Jemson Jun Peng / COMP Course: UROP 2100, Fall UROP 3100, Spring The advent of sophisticated tools has revolutionized the deployment and sharing of machine learning models, enabling developers to concentrate more on crafting these models rather than their distribution. In the realm of chemistry, groundbreaking advancements such as reaction extractors and molecule recognition tools have emerged, propelled by the swift advancements in machine learning, NLP, AI, and GPT technologies. Currently, the focus is on deploying machine-learning models more efficiently and delving into the effectiveness of chemistry paper extractors. In this pursuit, I plan to utilize YOLOv8 as a core component of my toolkit, significantly enhancing the process of document recognition and streamlining the journey of information extraction. Developing Deep Learning Enabled Nucleic Acid Structural Generator Supervisor: SU Haibin / CHEM Student: WANG Martin Shen / CHEM-IRE Course: UROP 3200, Summer The requirement for accurate structure predictions of novel nucleic acid structures has become everincreasingly important as the discovery of more diverse nucleic acid families have become important in understanding key biological interactions besides the storage of genetic information, and the feasibility of developing de novo nucleic acid structures like aptamers and nanostructures to serve as drug molecules or carriers, biosensors, or nanofabrication applications. While existing nucleic acid structure predictors that use an attention-based architecture to decipher features of nucleic acid tertiary structures have recently become popular, their large computational requirements, long training time, and narrow scope in terms of what species of nucleic acids the model can generate have all limited their widespread usage in structural biology. This work demonstrates and evaluates universal approach to nucleic acid structure prediction that focuses on modelling trends based on the chemical and physical features of both DNA and RNA from a much more streamlined attention mechanism that uses lower computational power.
RkJQdWJsaXNoZXIy NDk5Njg=