Machine learning-guided engineering of genetically encoded fluorescent calcium indicators.

Area of Science:

• Biotechnology
• Molecular Biology
• Biophysics

Background:

• Genetically encoded fluorescent indicators are vital for real-time biological activity monitoring.
• Protein engineering relies on understanding sequence-function relationships, often through extensive mutagenesis and screening.
• Calcium indicators like GCaMP are crucial tools in neuroscience and cell biology.

Purpose of the Study:

• To apply machine learning for predicting outcomes of protein sensor mutagenesis.
• To engineer novel genetically encoded fluorescent indicators with enhanced performance characteristics.
• To develop faster and more sensitive calcium indicators using computational approaches.

Main Methods:

• Utilized machine learning to predict functional outcomes of protein mutagenesis.
• Developed an ensemble of three regression models trained on GCaMP mutation libraries.
• Performed in silico functional screening of 1,423 novel GCaMP variants.

Main Results:

• Identified ensemble-derived GCaMP (eGCaMP) variants with faster kinetics and larger responses.
• Discovered eGCaMP and eGCaMP+ variants surpassing previously published fast variants.
• Developed eGCaMP2+ with an extraordinary dynamic range, outperforming multiple GCaMP generations.

Conclusions:

• Machine learning effectively facilitates the efficient engineering of proteins with desired biophysical properties.
• The developed eGCaMP variants offer significant improvements for biological sensing applications.
• This study highlights the potential of computational methods in advancing protein engineering for biosensors.