#36 · Structural bio & protein design

ML and simulations reveal GPCR activation pathways and lipid allosteric binding

The common thread is computational analysis of GPCR structure and dynamics, where simulation and learned models explain how ligand identity and allosteric factors shift receptor conformation. Molecular dynamics and enhanced sampling show that 18:0 and 18:1 lysophosphatidic acids adopt distinct orthosteric poses and tail orientations at LPA1, altering CWxP and PIF motifs. DRUMBEAT applies interpretable machine learning to MD data to identify residue communities driving conformational transitions, validated on β2AR. Gcoupler combines de novo ligand design with graph neural networks to identify endogenous sterols as allosteric modulators of the GPCR-Gα interface in yeast. What is new is the pairing of interpretable learning with MD to extract residue-level mechanistic networks, and the use of generative design to surface endogenous allosteric ligands. The methods differ across items, so the shared element is mechanistic interpretation of receptor activation.