Machine Learning Models for Local Optimization of Red Fluorescent Protein Variants in a Low-Data Setting.

Area of Science:

• Biochemistry
• Molecular Biology
• Biotechnology

Background:

• Fluorescent proteins (FPs) are crucial for visualizing cellular processes.
• Traditional FP engineering methods (rational design, directed evolution) are expert-intensive and labor-intensive.
• AI applications for FP engineering are emerging, but primarily for green FPs, with limited use for red fluorescent proteins (RFPs).

Purpose of the Study:

• To develop and apply machine learning (ML) for the local optimization of red fluorescent protein (RFP) variants.
• To address the scarcity of AI-driven engineering approaches for RFPs.
• To create a data-efficient ML framework for targeted protein engineering.

Main Methods:

• Trained lightweight, descriptor-based ML models using a dataset of over 150 RFP sequences.
• Focused on local sequence space optimization around the mScarlet-I3 RFP.
• Utilized model predictions to guide the selection of promising RFP variants.

Main Results:

• Identified RFP variants with red-shifted emission peaks.
• Discovered variants exhibiting large Stokes shifts.
• Found variants with brightness comparable to the parental mScarlet-I3 RFP.
• Demonstrated the effectiveness of interpretable, data-efficient ML models for guiding protein engineering.

Conclusions:

• Machine learning provides an effective auxiliary tool for targeted local engineering of RFPs.
• The developed ML framework offers a practical approach for optimizing fluorescent proteins.
• This study expands AI applications to RFP engineering, facilitating advanced cellular imaging.