One Descriptor to Fold Them All: Harnessing Intuition and Machine Learning to Identify Transferable Lasso Peptide...

Area of Science:

• Biophysics
• Computational Chemistry
• Protein Science

Background:

• Lasso peptides possess unique knot-like structures conferring exceptional stability.
• Their inherent stability makes them promising for therapeutic applications.
• Spontaneous folding into the lasso conformation is rare, hindering direct synthesis and therapeutic design.

Purpose of the Study:

• To develop a computational method for identifying reaction coordinates of lasso peptide folding.
• To characterize the folding free energy landscape of lasso peptides.
• To enable strategies for stabilizing the pre-lasso ensemble for efficient synthesis.

Main Methods:

• Combined collective variable (CV) discovery using chemical intuition and machine learning.
• Employed enhanced sampling techniques, specifically metadynamics.
• Utilized harmonic linear discriminant analysis (HLDA) for CV identification.

Main Results:

• Successfully identified collective variables (CVs) that distinguish the pre-lasso fold.
• Converged the folding free energy landscape of lasso peptides.
• Demonstrated the transferability of identified CVs across different lasso peptides.

Conclusions:

• The developed protocol effectively identifies folding reaction coordinates for lasso peptides.
• This method provides a pathway to understand and potentially engineer lasso peptide folding.
• Facilitates future efforts in the synthesis and therapeutic design of lasso peptides.