Discovery and design of potent cell surface display elements

Area of Science:

• Bioengineering and Synthetic Biology
• Artificial Intelligence in Molecular Biology
• Protein Engineering and Display Technologies

Background:

• Cell surface display (CSD) elements are crucial bioengineering tools, but their sequence-function relationships remain poorly understood.
• A lack of systematic rules hinders the reliable design and optimization of potent CSD modules for applications like mRNA antigen display.

Purpose of the Study:

• To develop a deep learning and artificial intelligence framework, DeepSCan, for systematically mapping CSD sequence-function relationships.
• To capture conserved features of potent CSD elements and iteratively design novel CSD modules for enhanced mRNA antigen display.

Main Methods:

• Development and application of DeepSCan, a suite of AI models, to analyze CSD sequence-function relationships.
• Experimental quantification of surface expression for over 570 chimeric antigens and derivation of cell surface translocation strength labels for approximately 310 CSD elements.
• Compilation of ~45 independent training datasets and training of three generations of DeepSCan models.

Main Results:

• DeepSCan models successfully captured conserved features of potent CSD elements.
• Computational design and experimental validation yielded 7 novel generative CSDs matching or exceeding the performance of natural CSDs.
• Enhanced surface displays were validated across multiple cell types and demonstrated functionality in an antigen-specific CAR-T cytotoxicity assay.

Conclusions:

• DeepSCan provides a systematic approach to understanding and engineering CSD elements.
• The identified generative CSDs offer improved tools for cell surface display applications, particularly in mRNA antigen display and immunotherapy.
• This work advances the design principles for potent bioengineering modules and their functional integration into cellular systems.