Science30 June 2026 · 6 min read

MOF screening in 6 hours,
not 10 days.

We ran a real Grand Canonical Monte Carlo simulation on PeakOps Eleven — CO₂ adsorption in IRMOF-1 — and filmed the whole thing. Here is the number, the pipeline, and why it took one click instead of a week of setup.

The full MOF science → platform film (~95s). Rendered with Remotion; every number is real.

The number

CO₂ adsorption in IRMOF-1 at 298 K, 1 bar 4.274 molecules / unit cell. Not a mock-up — it came straight out of a RASPA GCMC run launched from the platform, reading the bundled IRMOF-1.cif with the example MOF force field.

The pipeline, pre-built

High-throughput MOF screening is a chain, and every link is already installed and validated in the RASPA / MOF Screening golden environment:

Structure
pymatgen · ASE · CIF handling
Zeo++
pore geometry, PLD/LCD, surface area
PACMAN
ML partial charges for GCMC electrostatics
RASPA 2 + 3
GCMC uptake, Henry coefficients, isotherms
scikit-learn · XGBoost
rank thousands of candidates
MOFid · RDKit · OpenBabel
identity, dedup, format conversion

Feed it the CoRE MOF 2024 database, filter geometrically with Zeo++, assign charges with PACMAN, run RASPA in a Slurm job array, and rank the winners with ML — before you spend a single GPU-hour on the wrong candidates.

Why it was one click

No modules to load, no conda conflicts, no ParallelCluster YAML. You pick the environment, launch a machine on your own AWS account in minutes, connect with one-click SSH, and run. Per-project budget guardrails keep the bill honest; RASPA2 stays for backward compatibility while RASPA3 brings the speed.