NanoMelt

NanoMelt predicts nanobody thermostability using an ensemble model that stacks multiple regression approaches with diverse sequence embeddings.

Key Features:

• Ensemble Learning: An ensemble model stacks multiple regression using diverse sequence embeddings to achieve state-of-the-art accuracy in predicting nanobody thermostability.
• Pre-trained Embeddings: The framework leverages pre-trained embeddings for protein fitness prediction from limited data.
• Repeated Stratified Nested Cross-validation: This approach ensures an unbiased assessment of model performances during training and evaluation.
• Curated Dataset: The tool is trained on a dataset of 640 measurements of apparent melting temperature, integrating data from the literature with 129 new measurements.

Scientific Applications:

• Nanobody Development: NanoMelt guides the selection of highly stable nanobodies to streamline nanobody development.
• Protein Fitness Prediction from Limited Data: The framework addresses in-silico prediction of protein biophysical traits when experimental data availability is limited and heterogeneous.

Methodology:

NanoMelt employs an ensemble model stacking multiple regression using diverse sequence embeddings, with training based on repeated stratified nested cross-validation.