CodeFP: A New Model for De Novo Functional Protein Design
A study published on arXiv (2605.00948) presents CodeFP, a co-generative model designed for creating functional proteins from scratch. This model effectively decodes both sequence and structure tokens, enabling it to achieve desired functionality and foldability, thus addressing the shortcomings of current direct mapping and decoupled methods. By utilizing functional local structures, CodeFP enhances semantic encodings and incorporates auxiliary functional supervision to mitigate training ambiguities associated with one-to-many structure mappings. The implications of this research are particularly relevant for the fields of biotechnology and medicine.
Key facts
- CodeFP is a co-generative protein language model for de novo functional protein design.
- It simultaneously decodes sequence and structure tokens.
- It aims to achieve both functionality and foldability.
- Existing approaches often fail to achieve both simultaneously.
- CodeFP uses functional local structures to enrich semantic encodings.
- It introduces auxiliary functional supervision to reduce training ambiguity.
- The paper is available on arXiv with ID 2605.00948.
- Applications include biotechnology and medicine.
Entities
Institutions
- arXiv