Water geometries are not fixed

[amcisaac]

Hi, I've been working with the SPICE dataset and noticed that in the Solvated Amino Acids and Water Clusters subsets, the water geometries aren't fixed to the TIP3P-FB values (bond length of 1.01 A, and an angle of 108.1 degrees). I didn't notice the issue in the Solvated PubChem subset.

I'm attaching examples showing the angle deviation from arg (conformer 0) in the Solvated Amino Acid subset and conformer 2 of the water cluster subset, but I've seen it in every structure I've examined from these two subsets. I've seen bond lengths ranging from 0.93-1.07 A and angles from 95-120 degrees.

I'm wondering if it could be related to this collection of bugs in the barostat used to generate the structure? Or maybe I've misunderstood the details of the water model reported in the paper? Please let me know if I can provide more information.

[peastman]

That's intentional. SPICE is a dataset of quantum calculations, not energies from a classical force field. It samples internal motions, not just rigid displacements of whole molecules.

[amcisaac]

I see, that makes sense. Could you explain how the internal motions were sampled? In the paper, it says that the solvated amino acid conformers are generated by taking snapshots from TIP3P-FB run at 300K, so I was just a bit surprised since it's a rigid water model.

[amcisaac]

After reading carefully through the code for these two subsets, I see that the solvated amino acids use TIP3P-FB with rigid_water False, and the water clusters are started from TIP3P-FB but actually simulated with AMOEBA. That explains the non-rigid geometries. Thanks!