Comparing LigandMPNN and Directed Evolution for Altering the Effector-Binding Site in the RamR Transcription Factor

Area of Science:

• Protein engineering
• Computational biology
• Biochemistry

Background:

• Machine learning (ML) tools for protein design are rapidly advancing, with successes in stability and ligand binding.
• Designing proteins with intrinsic allosteric mechanisms remains a challenge.
• Allosteric transcription factors (aTFs) regulate gene expression via allosteric communication and are potential targets for biosensor development.

Purpose of the Study:

• To evaluate the efficacy of the LigandMPNN computational tool for redesigning the ligand binding pocket of an allosteric transcription factor.
• To compare computational redesign results with previous directed evolution efforts for altering effector specificity.

Main Methods:

• Computational redesign of the RamR transcriptional repressor's ligand binding pocket using LigandMPNN.
• Prediction of wild-type RamR structure in complex with five benzylisoquinoline alkaloids (BIAs).
• Experimental testing of nine redesigned protein variants for functionality in *Escherichia coli*.

Main Results:

• Little overlap was observed between computational redesign outcomes and previous directed evolution results.
• None of the nine computationally redesigned RamR variants were functional in *E. coli*.
• The study highlights limitations of current ML tools for designing allosteric proteins.

Conclusions:

• Current computational protein design methods, including LigandMPNN, may not be sufficient for engineering allosteric or conformationally flexible proteins.
• Further development of protein design strategies is needed to effectively incorporate allosteric mechanisms.
• The findings suggest a need for alternative or complementary approaches for designing proteins with complex regulatory functions.