Protein Sequence Analysis using Generative Language Models

Centre for Smart Systems, AI and Cybersecurity (SSAICS)

This research will develop new embedding techniques for language models such as Bert to apply them in protein sequence analysis and classification of viral proteins.

This project aims to develop new techniques for protein sequence analysis using language models such as Bert, which have shown significant promise in improving accuracy and reducing computation time in biological applications. The limited availability of high-quality data is a challenge for training machine learning models, and to address this, new data augmentation techniques will be developed to generate synthetic data. The research will focus on developing new embedding techniques for language models that convert amino-acid sequences into vector representations, capturing both structural and functional traits of the sequences while preserving the integrity of the original data. This will enable the generation of more accurate and expressive sentence representations, important in protein sequence analysis. The developed techniques will be applied to classify viral proteins, focusing on influenza type A as a case study, which is essential for understanding the structure and function of the virus and developing effective treatments. The project aims to incorporate robust biological assumptions into these models and make them accessible to the wider scientific community.