MELD by Alberto Perez AMAP slide: JW – Does MELD approach assume that all native regions are known, and FF priors can help model transitions between them? Or can FF identify other native states even if those aren’t part of the known set? JW – Are CMAPs necessary? We generally think torsions are largely useful for cleaning up places where electrostatics and vdW forces aren’t accurate enough. What alternatives might exist? AP – No strong opinion. I dind’t think I’d like implicit solvent either. CMAPs helped us a lot in our projects. Cheatham used to say phase should either be 0 or 180, and that’s a purist view. But CMAPs are a clean way to add corrections in a modular way. Some of our existing work is looking more at this with like AI. First time CMAPs were introduced at AMBER dev meeting, some folks argued we were throwing out transferability.
LW – Thoughts on implicit vs. explicit solvent? Might implicit solv lead to different minima? AP – Some cases will need explicit, others can use implicit. Will affect sampling needs. We tried using explicit, but temperature ladder was problematic. For some targets water mediated hbonds were important. Often found that water-water interactions were really strong, so it was hard to get protein-water interactions (big energy penalty). So had to add extra restraints to get ligand out of binding site. So now thinking like DE Shaw about explicit solvent, namely that it should include temperature replica exchange.
JW – We may be able to do noncanonical parameterization for your unusual-crosslink case. Could I schedule a time with a grad student? … LW – We’re also trying to develop a protein FF. LW – Surprised that 99SB doesn’t overrepresent helices.
|