Alexander Rose

PDB101: Register for the Aug 11 Virtual Office Hour on Pairwise Alignment PDB-101: Training, Outreach, and Education portal of RCSB PDB

Register for the Aug 11 Virtual Office Hour on Pairwise Alignment
Learn how to align one or more protein chains to a reference structure in a pairwise manner
pdb101.rcsb.org/news...

New release for #ipymolstar: I've added a widget for the amazing and very powerful MolViewSpec. This allows you to control the camera📷from python🐍, as well as all other molviewspec features 🎉

ipymolstar repo: github.com/jhsmit/ipymo...
Molviewspec: molstar.org/mol-view-spec/

sorry, trying better res

Isosurface of X-ray density only lit by emissive slice of same density. Illuminated in #molstar.

Ipymolstar Annotate Colors - a Hugging Face Space by Jhsmit Annotate protein structural models with colors

You can try it on Huggingface: huggingface.co/spaces/Jhsmi...

The data is only encoded in the link and not stored so you can use it to share unpublished data*

(at your own risk, no warranty, links might stop working)

Here's a neat way to share you protein-annotated data 🚀🧪

Upload your .csv with residue/value data, select data column, colormap and norm to interactively color your protein model.

Then, add title/description and click the button to open a shareable view with url-encoded data and settings.

A quick molecule of the month render for November's Malaria Parasite PTEX (pdb101.rcsb.org/motm/299) using #b3d and #MolecularNodes

#GeometryNodes #SciArt

#ipymolstar v0.0.9: Added click interactions

try it on pycafé: py.cafe/jhsmit/ipymo...

Splendid, thank you

Try Mol* illumination mode: molstar.org/viewer/?illu...

Example: molstar.org/viewer/?snap...

Integration Progressive rendering Keep viewport interactive Use requestAnimationFrame Also check `gl.SYNC_GPU_COMMANDS_COMPLETE` if available Because requestAnimationFrame may fire even if GPU is still busy Adjust quality parameters based on previous iterations’ render time Adjust denoise threshold to hide graininess So instead of grainy to non-grainy it goes from blurry to non-blurry Screenshots Browsers may lose the WebGL context if render time of a single frame is too large or too many frames are submitted to the GPU Check `gl.SYNC_GPU_COMMANDS_COMPLETE` if available, otherwise use `gl.readPixels` to wait Anti-aliasing Since normals and direct-light are pre-calculated, jittering rays is not enough Recalculate normals and direct light with jittered camera, trace and blend

A few notes on integrating SSGI into Mol*.

Thickness For tracing in screen-space we need to estimate the thickness Automatic thickness (estimate) Base thickness as max(backface depth) - min(frontface depth) Per object density factor to adjust base thickness Fixed thickness Want to avoid as we have representations with different “densities” Still support in-case user wants to do manual tweaks

For tracing in screen-space we need to recover thickness.
Want it automatic, not fiddle with parameters.

Insight: Molecules are shown with visuals of very different density.

So we do 2 things
1) Base thickness from front- and back-face depth
2) Density factor to adjust for each type of visual

Tracing Inputs Shaded color (rgba, standard pass) Normal (xyz) + emissive strength (w) Diffuse color (rgb) + “density” (a) Front depth Back depth Sampling Backwards Monte Carlo Only diffuse (“specular” from shaded input) Russian Roulette to boost non-terminated rays Tracing Hit background Return background color Hit surface, then nothing Return shaded color Hit surface, bounce x times, then nothing Return shaded color plus accumulated diffuse of bounces Hit surface, bounce more than x times Return no color

Inputs are rendered by the same shader code as the standard rendering. There are just two extra output targets: normal + emissive strength and diffuse + density. Additionally we do a pass to get back face depth.

Passes On change (camera, geometry, lighting, …) Transparent geometry pass Post-processing input passes (outline, marking, transparent ssao) Tracing input pass Backface depth pass Sampling Tracing sample pass Compose Compose pass (denoise, blend transparent geometry, apply outlines, transparent ssao) Background pass Antialiasing pass Bloom pass DoF pass

Main idea is to treat the SSGI as a progressive enhancement of sorts. Try to reuse as many of the passes used for normal rendering. And keep the overall look as similar as possible.

High quality lighting from "path-traced" screen-space global illumination. Even works with only emissive lights.

About Mol* Suite of tools for 3D molecular data on the web Custom WebGL rendering engine Scales from single atoms to cells with billions Wanted nicer lighting that still scales “Path-traced” SSGI Screen-space, so OK with billions of atoms Sample-based, so good to adjust quality

Gave a short presentation on Illuminating Molecules in #molstar at the November 2024 #WebGL & #WebGPU Meetup

www.khronos.org/events/webgl...

How long does the denoiser take compared to the rest?

Hard to judge detail quality but looks so smooth, very nice.

Significant new work on LOD of aggregates! mangosister.github.io/scene_agn_si...

🚀 Excited about our pre-print:
AlphaBridge: A User-Friendly Tool for Interpreting Protein Predictions

Whether you're an expert or just new to the field of Protein Prediction, AlphaBridge is the way to go!

🔁Try it and share your thoughts!
📜 Article: www.biorxiv.org/content/10.1...

Very cool to see the cell floating in midair

InterProt Interpreting Proteins through Language Models

InterProt (interprot.com#/sae-viz/SAE...) is a tool that applies the same interpretability analysis used to make Golden Gate Claude to protein language models. For example feature 4000 recognizes residues lining inside of de novo designed & natural transmembrane beta-barrels (PDBs 6X1K, 6X9Z, 2MLH)

Two-way tertiary/primary structure hover and highlight with #ipymolstar and #altair

Try it on pycafe: py.cafe/jhsmit/ipymo...

Ionic lock in inactive rhodopsin - illuminated in #molstar