Thread by @karmecon

#MDSI2019 Day2 Session 2: "Standardization of file format I/O" by @jppiquem.

Straight into problems -- input files are a tougher cookie to crack than output 🍪

Actually, Day 2 Session 3 #hasphdcantcount

#Input: Many problems -- many possible solutions! How to find the right one? #goldilocks

#Output: probably easier to handle and we can learn from other fields, like climate research.

One emerging consensus in the room -- #killpdb. However, what to use to replace it?

cc: @agrossfield

#HPC performance should be taken into account to prevent #MD being kicked out from supercomputers due to inefficiencies. We also need to take streaming and new tech, such as AR/VR, into account. But, how many #fileformats?

@TINKERtoolsMD chose to use #NetCDF for practical purposes. @daniel_r_roe cheers remotely for the wider adoption of NetCDF 🙌

Multiple solutions exist for output files too, but coming up with a clear #semantic definitions has been mentioned a lot today

The second speaker in the session is Joao Ribeiro of #NAMD -- only pointing fingers at themselves -- a lot of space for improvement.

This talk was fast!

Next is @eriklindahl will share some thoughts not on file formats, but on information exchange as a more important thinking category in this context.

"Interoperability is first #semantics, not #formats".

What coffee means in different countries, cities and cultures? According to @eriklindahl, the middle cup is sold as coffee, but it's actually sweet brown water...

So, what do we mean when we say things? #semantics strikes again

#Provenance is also an important part of this discussion.

Some suggested solutions below, in essence -- find a core set of common options that we all agree on. Which is a large amount of work.

If something was good 20-30 years ago, doesn't mean we should still use it in 2020 -- we need to move on and allow depreciation of (poorly) aged features

One potentially low hanging fruit for #output might be a seed for a common trajectory format. #tng was developed for efficient compression, small library and efficient I/O.

Current state of #tng -- great work with C API by Magnus Lundborg, but then #gromacs switched to C++. So 90% there, but the last 10% has been a challenge to overcome.

Suggestion -- make #tng a community effort!

Speaking now -- Peter Eastman from #OpenMM. #PDB is again mentioned as a villain in this file format saga, but even "a bad standard is better than no standard." When it comes to trajectories, we as a field at not doing too bad.

Where we do really bad in our field? Force fields and parameterized systems. How do we even start here? So many different ways to assign parameters and every package is doing it differently -- each of this ways is associated with a different workflow.

Any format must be #modular and #extensible!

Peter ends with an optimistic outlook: Sorting this out will be a lot of work that will keep occupied for a long time to come 🔧⚙️

@Bradley_Dice speaking now about #GSD #HOOMD schema which allows defining your own shapes

They are also working on #freud library for analysis of coarse grained models, nanoparticles and colloidal crystals.

glotzerlab.engin.umich.edu/hoomd-blue/

freud.readthedocs.io/en/stable/

The last talk of the session today is on #ParmEd by Jason Swails, starting with its brief history from motivation to use #charmm and #amber input files in #openmm and later to #gromacs and interconvert between these packages

Challenges with parameters ordered from easy to difficult.

Impropers are terrible. Terrible.