Welcome to unZipro

unZipro (unsupervised Zero-shot inverse folding framework for protein evolution) is a lightweight GNN-based framework designed for AI-guided protein engineering.

How it works

• Zero-shot transfer learning captures a universal protein fitness landscape.
• Meta-learning adapts to family-specific fitness landscapes.
• Prioritizes the most promising high-fitness variants for experimental validation.

Key Features

• Zero-shot transfer – predict functional variants without large experimental datasets.
• Highly efficient – reduce experimental screening and computational costs.
• High accuracy – average 61% success for high-fitness mutations, up to 100% in some cases.
• Broad applicability – enzymes, nucleases, polymerases, transcription factors, virus-resistance proteins, etc.
• Structure-flexible – supports both experimental and AlphaFold-predicted structures.

Applications

• Enzyme engineering
• Genome editing tool optimization (SpCas9, Cas12, base/prime editors)
• Plant protein engineering (virus resistance, transcription factor modulation)
• Protein therapeutics
• General protein design in biotechnology & agriculture