Use Our Optimization and Prediction Models (OP Models) to Analyze and Optimize Sequences Based On Your Data

Our OP Models tool allows you to use our protein large language models to learn then predict sequence-to-function relationships from your experimental data. Use your functional measurements to train machine learning models, then use these custom models to design and visualize sequences within your specified constraints.

Our models provide a quantitative score that a given sequence will achieve your design objective. They also output a mean expected value of the property and a standard deviation value to indicate the confidence in that prediction. This lets you make data-driven decisions about your library selection and optimize for performance despite cost and workflow constraints.

You can optimize for multiple properties in a single design cycle, as well as train models iteratively across multiple experimental cycles. Add more data at any time to further optimize your designs.

After you have designed libraries that meet your criteria, refine your variants by predicting the effect of each mutating site in your experimental context.

Learn more and get started with our tutorials

General tutorials

• Uploading your data

• OP Models scoring and log-likelihood
• Visualizing your data

Using the web application

• Model training and evaluation
• Designing sequences

• Navigating your projects
• Substitution analysis with OP Models

Using the Python client API

• Sequence-based learning
• Protein language models and embeddings

• Quantitative decision making for library design