Protein Design with Deep Learning.

Area of Science:

• Biochemistry
• Computational Biology
• Artificial Intelligence

Background:

• Computational Protein Design (CPD) has enabled significant advancements in engineering novel proteins for diverse applications.
• Recent efforts focus on integrating Deep Learning (DL) to enhance existing CPD methods by utilizing vast public protein datasets.

Purpose of the Study:

• To review and analyze various data representations used for protein sequences (1D) and structures (3D) in Deep Learning.
• To discuss the strengths and weaknesses of different representations for protein data.
• To detail associated Deep Learning architectures applied to protein design and related tasks.

Main Methods:

• Literature review of existing research on data representations for protein sequences and structures.
• Analysis of Deep Learning architectures employed in Computational Protein Design.
• Comparative discussion of the efficacy of different representation and architecture combinations.

Main Results:

• No single consensus exists on the optimal data representation for protein sequences and structures in DL.
• Different representations capture distinct aspects of protein information (1D sequence vs. 3D structure).
• Specific DL architectures are better suited for particular protein data representations and design objectives.

Conclusions:

• The choice of data representation is critical for successful Deep Learning applications in protein design.
• Further research is needed to establish standardized and optimal representations for diverse protein engineering tasks.
• Advancements in DL-based protein design are highly dependent on effective data representation and tailored architectures.