Peptide benchmarks | @Chapin Cavender | Slides will be uploaded DLM – do you think we could be overfitting to GB3 specifically? MG – what’s the time frame of results from AF? CC – OPC outperforms TIP3P on the small peptide benchmarks – other investigated water models are somewhere in the middle MG – although on this benchmark, we already outperform ff14sb CC – shows results broken down by residue. AMBER was trained to QM on Ala and Gly (esp Ala5) and does well on them, but much worse on bulkier residues
CC – do we want to do protein simulations with one of the 3-site water models? Or do we want to do further experiments first? MS – let’s get timings from AF, let’s just run both if we can CC – yep, I can branch out on the protein dimension with OPC water, she can look at different water models on GB3 DLM – loop me in if you need more computing time
MS – we also wanted to run some of the mixture calculations locally, especially with Tim. I’ll coordinate with Simon on running Evaluator on these things over the next week or so. CC – currently writing scripts to analyse other observables, e.g. order parameters, residual dipolar couplings, will try to show these results at the next meeting DLM – Want to avoid benchmarking everything on everything. Let’s watch for scope creep and avoid trying to solve unsolved problems. Let’s prioritise tests we can do with less Chapin time and more computer time. MS – agree CC – still considering the “tier 2 benchmarks” of ~a dozen proteins, and some disordered proteins, for handing off a candidate to LW and make that the release candidate. CC – My only concern now is if we need to benchmark this with more than one water model. It’s looking like that now, since TIP3P doesn’t do so well for GB3, but we should see if that’s true for other proteins
PB – if you start with unfolded GB3, does it fold? CC – I have not tested this. Generally you need to set up a simulation for this specifically PB – just wondering since TIP3P-FB looks better than TIP3P, is it possible to test it quickly on GB3? CC – Not quickly. These would be hard to do without something like Anton CC – so we can fold small motifs like helices over ~10 us or so.
|
