Esperanto/Protein-Phi-3-mini
Esperanto/Protein-Phi-3-mini is a 4 billion parameter language model, fine-tuned from Phi-3-mini-4k-instruct, specifically for protein language modeling. Developed by Aayush Shah and Shankar Jayaratnam, it excels at generating novel protein sequences based on natural language prompts. This model supports both uncontrollable and controllable protein generation, making it a valuable tool for advancements in protein engineering, drug development, and biotechnological applications.
Loading preview...
Protein-Phi-3-mini: Specialized Protein Language Model
Protein-Phi-3-mini is a specialized version of the Phi-3-mini-4k-instruct large language model, continually pre-trained using LoRA on extensive protein sequence datasets. This 4 billion parameter model is designed to understand and generate protein sequences, significantly accelerating the protein engineering process compared to traditional, labor-intensive methods.
Key Capabilities
- Novel Protein Generation: Generates new protein sequences based on natural language prompts.
- Controllable Generation: Allows users to specify desired characteristics for the proteins, supporting generation within 10 protein family classes (e.g., "[Generate Ligase enzyme protein] Seq=<").
- Uncontrollable Generation: Capable of generating diverse protein sequences without specific constraints using a simple prompt like "Seq=<".
- Accelerated Discovery: Leverages LLM capabilities to generate and evaluate protein sequences rapidly, enhancing diversity and enabling the creation of novel proteins with unprecedented functions.
Use Cases
- Drug Development: Facilitates the design of new proteins for therapeutic applications.
- Chemical Synthesis: Aids in creating proteins with specific functions for industrial and research purposes.
- Biotechnological Applications: Supports various advancements in biotechnology by streamlining protein engineering workflows.
For more in-depth information, refer to the associated paper.