Geoff Hutchison
@geoffhutchison.net
2.7K followers 1.1K following 440 posts
Chemist, Scientist, Husband, Dad, Runner, Teacher. I develop Avogadro + OpenBabel to find new materials for energy applications @Pitt_Chemistry he/him
Posts Media Videos Starter Packs
Pinned
geoffhutchison.net
Welcome to all the newcomers to #ChemSky 🧪

I don't post much, but you'll get a dose of #compchem and materials discovery, a bit of #machinelearning (with skepticism) and some sprinkling of #Pittsburgh (including @pitt.bsky.social and @pittchem.bsky.social)

Enjoy some recent @avogadro.cc renders
Image of an endofullerene - a translucent sodium atom encased in a C60 buckminsterfullerene molecule, rendered by Avogadro Image of a zirconium metal complex with a translucent green sphere representing the Zr metal ion interacting, labeled by Zr characters, bonded to two cyclopentadienyl rings, a methyl group and a hydrogen
geoffhutchison.net
Yes, it already has distance, angle, torsion constraints. At the moment, we don't have "generate a set of geometries," (i.e., bond length or torsion scan) but it's a reasonable ask. Probably not until after this Oct. 23rd release though.
geoffhutchison.net
Avogadro v1 had this feature - it’s taken a while to build up the pieces in Avo2, and I’m sure there are still bugs. The optimize geometry is still there, but interactive optimization is useful for some tasks like teaching or steering towards a particular confirmation.
geoffhutchison.net
Pixi is great because it works with uv as well as conda. Since a bunch of chemistry-related packages are on conda-forge (e.g., xtb, crest, psi4…) it's convenient to have both. Really fast too.
geoffhutchison.net
You can already do that. That's how I wrote the adapter scripts. 😉

The problem is that a lot of end users don't have Python and don't know how to set things up. @olexandr.bsky.social is right - it's possible to compile models, but in the meantime, we'll use pixi to create environments if needed
geoffhutchison.net
Contributions always welcome. If you can get someone to get initial code working, I’m happy to help.
geoffhutchison.net
And to be complete, I have a whole pile of scripts for ML potentials, XTB .. calculating charges with antechamber (AM1BCC, etc.) as well as using pymatgen, ASE, etc.

Avogadro2 is really easy to extend with Python
geoffhutchison.net
I have scripts for ANI and AIMNET2 already. The main holdup is that they assume you have all the Python packages installed, setup, etc.

Coming soon is a feature that uses pixi to install anything you need when adding Avogadro2 plugins.

BTW - is there syntax for AIMNET with unit cells?
geoffhutchison.net
Well, the framework can use any method now. You can write Python adapters to ML, GFN-FF, DFTB or tblite, etc. Just needs energies and gradients.

For small molecules, MMFF94 or GAFF work fairly well.

Several MOF / materials folks did ask for UFF with periodic boundaries though. 🤷‍♂️
geoffhutchison.net
Sure. I could use some suggestions in terms of thermostats, since I've never implemented them. Berendsen is fairly easy, but seems like a bad idea?

I guess Bussi-Donadio-Parrinello?
geoffhutchison.net
Oh, it's wonderful for teaching, as well as pushing and pulling into a particular conformation.

Just required building up piece-by-piece since people want to use ML methods, GFN2, etc.
geoffhutchison.net
Tested out some water clusters during lunch.
- MMFF94 has point charges and hydrogen bonds
- UFF does not (so water molecules push apart)

At some point I'll add some simple dynamics…

#compchem #compchemsky
geoffhutchison.net
That's really useful - including the example with the chelating ligand.
geoffhutchison.net
I know this has been one of the most widely requested features. Should be in the release on 10/23

#chemsky #compchemsky #opensource
avogadro.cc
If you've been wondering about the status of the AutoOptimize interactive optimization in Avogadro2 - coming later this month 👀
Screenshot of Avogadro2 using the interactive AutoOptimize tool to minimize a molecule (n-pentane) using the UFF force field. Text reads UFF ΔE = -229.80 kcal/mol
geoffhutchison.net
That particular image is 1536 x 604, or roughly 5" x 2" at 300 dpi. (I have a high-res screen.)

PDF is something of a "container" format. You can store bitmap images, text, vector graphics.. a lot of different things. The best are resolution-independent (e.g, draw a line from x to y points)
geoffhutchison.net
Probably depends a bit on the resolution - the PDF is actually surprisingly small (65 Kb for this one) because it's all vectors.

But yes, GIF can get excellent compression when there's limited colors like line plots.
geoffhutchison.net
#compchemsky

Now offers tools to save computed spectra as PNG, PDF, SVG .. or the underlying data.

Found a really nice #opensource #openscience plotting library
avogadro.cc
👀

Coming soon...
Dialog titled "Vibrational Spectra" showing the computed infrared spectra of caffeine with two plots, one in black labeled "smoothed" and one in red labeled "raw"
geoffhutchison.net
BTW, lots of great collaborations between @pittchemistry.bsky.social and the School of Pharmacy
velvetjournigan.bsky.social
Come to our Graduate Program Open House at the University of Pittsburgh School of Pharmacy, October 22nd ! 🔬💊🌠
geoffhutchison.net
#chemsky
royalsociety.org
Our Career Development Fellowships are now for applications. This scheme provides talented early career scientists from underrepresented backgrounds in STEM with research funding and training opportunities to build a successful research career: #RSGrants royalsociety.org/grants/caree...
geoffhutchison.net
aka "my AI model has no idea the relationship between a mile and a kilometer" or "my AI model can't do math" - not sure which is worse!
geoffhutchison.net
I would *love* to turn this off by default and am considering switching search engines.

Same thing with images. Just tons of AI-generated stuff now.
geoffhutchison.net
Yes - I need to upload more - I have a little script that can generate molecular models for pretty much anything.

If anyone in #chemsky wants a STL / 3D model, just let me know what you want.
geoffhutchison.net
(sigh) - hopefully it's just another TACO
geoffhutchison.net
Contacting faculty on a mentee's behalf is great. In my experience it doesn't happen often.

This year, I'm telling everyone (mentees, etc.) that it will be a unpredictable cycle. Almost every program is targeting a smaller cohort. Have a backup plan or two.