Sequence representation approaches for sequence-based protein prediction tasks that use deep learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Deep learning is increasingly vital in bioinformatics for sequence-based protein prediction.
• Availability of large biological datasets and rapid advancements in deep learning drive this trend.

Purpose of the Study:

• To summarize main approaches for protein sequence data representation (encoding/embedding).
• To review architectures and development of various sequence embedding models.
• To aid researchers in selecting suitable models for their requirements.

Main Methods:

• Categorization of protein sequence embedding methods: end-to-end, non-contextual, and transfer learning-based.
• Inclusion of task-specific embeddings (e.g., for structure prediction, drug discovery).
• Theoretical review of embedding model architectures and their evolution.

Main Results:

• Comprehensive overview of diverse protein sequence embedding strategies.
• Detailed explanation of different embedding methodologies and their applications.
• Identification of key factors influencing model performance in sequence-based prediction.

Conclusions:

• Effective protein sequence representation is crucial for deep learning model performance.
• A wide array of embedding techniques are available, catering to various bioinformatics tasks.
• This review provides a valuable resource for selecting appropriate sequence embedding models.