Biological Sequence Representation Methods and Recent Advances: A Review

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in biology

Background:

• Biological sequence representation is crucial for machine learning in computational biology.
• Methods transform nucleotide and protein sequences for enhanced predictive modeling.

Purpose of the Study:

• To review and categorize biological-sequence representation methods.
• To detail principles, applications, and limitations of different methods.
• To outline future directions in the field.

Main Methods:

• Categorization into computational-based, word embedding-based, and LLM-based methods.
• Analysis of k-mer counting, PSSM, Word2Vec, GloVe, and Transformer architectures (ESM3, RNAErnie).

Main Results:

• Computational methods capture statistical/evolutionary patterns.
• Word embedding methods capture contextual relationships.
• LLM-based methods model long-range dependencies for superior accuracy in tasks like RNA structure prediction.

Conclusions:

• Challenges include computational complexity and interpretability.
• Future work focuses on multimodal data integration and explainable AI.
• Advancements promise transformative applications in drug discovery, disease prediction, and genomics.