Ab-initio amino acid sequence design from protein text description with ProtDAT

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Protein design is crucial for drug development and enzyme engineering.
• Current large language models struggle with multi-modal protein data (sequences and text).
• Existing methods lack the ability to effectively integrate diverse protein data types.

Purpose of the Study:

• To introduce ProtDAT, a de novo framework for fine-grained, multi-modal protein data interaction.
• To enable de novo protein design directly from descriptive text input.
• To unify protein sequences and textual information for enhanced design capabilities.

Main Methods:

• ProtDAT framework unifies protein sequences and text data.
• Utilizes a novel Multi-modal Cross-attention mechanism for integrated analysis.
• Evaluates protein design using metrics like pLDDT, TM-score, and RMSD.

Main Results:

• Experiments on 20,000 Swiss-Prot text-sequence pairs demonstrate ProtDAT's effectiveness.
• Achieved a 23.34% increase in pLDDT, indicating improved structural plausibility.
• Showcased a 76.45% increase in TM-score and a 24.41% reduction in RMSD, signifying enhanced accuracy and validity.

Conclusions:

• ProtDAT offers a significant advancement in protein design by effectively integrating multi-modal data.
• The framework demonstrates superior performance in generating accurate and functional protein sequences.
• This approach holds promise for accelerating applications in drug development and enzyme engineering.