Clustering is a fundamental task in data science, enabling the discovery of hidden structures in large datasets. In glycoinformatics, the classification and analysis of glycans present unique computational challenges due to their highly branched structures and compositional ambiguity, making conventional sequence-based approaches insufficient. This thesis explores an unsupervised clustering approach tailored for glycan data, integrating computational techniques to improve glycan classification and visualization.

We propose a multi-step clustering methodology implemented in Python, leveraging DBSCAN (Density-Based Spatial Clustering of Applications with Noise) with iterative refinements to enhance cluster accuracy. By incorporating domain-specific reassignment rules and substructure-based adjustments, the approach ensures that structurally similar glycans are grouped meaningfully. The results are visualized through an interactive D3.js Sunburst chart, providing an accessible and intuitive representation of glycan clusters.

The proposed framework was evaluated using data from GlyConnect, demonstrating its effectiveness in uncovering meaningful glycan relationships. However, challenges such as computational efficiency, automated validation, and scalability remain open for further research. Future work will explore deep learning-based clustering techniques, including Graph Neural Networks (GNNs), to enhance the clustering of glycans with complex structural variations.

By bridging computational clustering with glycoinformatics, this work contributes to the broader field of data science, emphasizing the role of unsupervised learning in biological data analysis