The protein sequences designed by AI must ultimately be verified in the physical world. Molecular dynamics simulations are the "gold standard" for verification, but traditional force fields have a fundamental contradiction: force fields that can accurately describe folded proteins often cannot handle natural random proteins (IDPs); force fields that can describe IDPs are prone to disrupting the stability of structural proteins.
Over the course of twenty years, the team has iteratively developed a molecular force field system with independent intellectual property rights, achieving unified high-precision simulation of structural proteins and IDPs.
The ff03CMAP introduces an energy-based correction term on top of the classic ff03 force field. It is highly consistent with NMR experiments for chemical shifts, J-coupling, and order parameters of short peptides, IDPs, structural proteins, and rapidly folding proteins, and is superior to all force fields of the same period.
BSFF2 is designed for dynamic RNA simulation. Through reweighted nonbonded parameters and dihedral lattice energy correction, it accurately simulates RNA motifs such as tetranucleotides, short single strands, double strands, and protrusions. It successfully captures the r(G4C2) repeat element chain slippage mechanism, providing a new tool for research on neurodegenerative diseases such as ALS.
FB18CMAP was developed for phosphorylated modified proteins, correcting the helical unwinding and dihedral potential trap bias at phosphorylation sites, and significantly improving conformational sampling of phosphorylated structural proteins and random proteins.
These three force fields constitute the underlying physical engine of the AI design results. The backbone generated by GPDL, the sequence designed by GPD, and the peptide output by PharmaPepGen all need to undergo energy assessment and dynamic verification through molecular force field-driven MD simulations. Without high-precision force fields, AI design is merely theoretical.
WeChat Customer Service
