Abstract:
Chemical compound data, often represented as graph-structured molecular graphs, plays a vital role in fields such as drug discovery, chemical informatics, and bioinformatics. Outsourcing such compound databases to public cloud platforms offers scalable computational resources for handling complex analytical tasks. Among these, full graph similarity search is an essential function for identifying structurally related compounds, which is crucial for compound retrieval, activity prediction, and molecular property analysis. However, outsourcing proprietary chemical data to the public cloud introduces serious privacy risks, as sensitive compounds must be encrypted before outsourcing. This encryption not only hinders the effectiveness of traditional similarity search algorithms but also complicates the enforcement of secure access control over outsourced data. In this paper, we tackle the problem of proprietary-assured secure chemical compound search in a privacy-preserving cloud computing framework. We propose a structure-aware and privacy preserving similarity search scheme with integrated access control. The compound owner (CO) utilizes the neural Graph2vec model to generate feature indices for encrypted molecular graphs. A secure transfer learning mechanism enables compound searchers (CS) to construct query feature indices using the same pre-trained model without access to the raw data. Furthermore, the CO is able to define and enforce flexible access-control policies to safeguard proprietary chemical knowledge. A rigorous security analysis shows that the proposed framework achieves strong privacy guarantees against both known cipher text and known-background attacks, while maintaining efficient search performance. This makes our scheme a practical and secure solution for chemical compound retrieval over the public cloud.
Page(s):
91-91
DOI:
DOI not available
Published:
Journal: 1st International Conference on "Recent Advances in Green Biotechnology and Climate Resilience", September 15-16, 2025, Volume: 1, Issue: 1, Year: 2025