Artificial intelligence in fusion protein three-dimensional structure prediction: Review and perspective

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Artificial intelligence (AI) has advanced protein structure prediction, but fusion proteins pose unique challenges.
• Fusion proteins arise from gene fusions in cancer, and their 3D structures are crucial for understanding function.
• Template-based modeling is difficult due to the lack of known structures for novel fusion proteins.

Purpose of the Study:

• To review the latest advancements in predicting 3D structures of fusion proteins using deep learning (DL) models.
• To explore the advantages and challenges of specific DL tools (AlphaFold2, RoseTTAFold, tr-Rosetta, D-I-TASSER) for fusion protein modeling.
• To outline the pipeline and factors for AI-driven fusion protein structure prediction.

Main Methods:

• Review of current literature on AI and DL applications in protein structure prediction.
• Analysis of deep learning models including AlphaFold2, RoseTTAFold, tr-Rosetta, and D-I-TASSER.
• Discussion of multi-level protein information utilization in DL models.

Main Results:

• Deep learning models show significant promise for predicting fusion protein 3D structures.
• Specific DL tools offer varying degrees of success and face distinct challenges.
• Key factors influencing prediction accuracy across different AI approaches are identified.

Conclusions:

• Accurate 3D structure prediction of fusion proteins using DL is an evolving field with ongoing challenges.
• Understanding these challenges is vital for advancing the functional analysis of chimeric proteins in cancer and other areas.
• Further development of AI models tailored for fusion proteins is necessary.