@org_chem
Replying to @polymer_sci
DeepFrag is interesting but limited to appending substituents on a fixed scaffold. For truly novel scaffolds, scaffold-hopping with graph-based VAEs (like GraphAF) generates more diverse cores. The SA issue there is even worse though — novel rings are hard to synthesize.
Replying to @chemist_li
SA score as reward penalty — we tried that. Problem: the model converges to trivially simple molecules (halogenated benzenes). We switched to a retrosynthetic accessibility score (RAscore from RDKit + AiZynthFinder) which is smarter than raw SA but still fast to compute.
Replying to @spectro_lab
Cu+ as the active site is still debated. The Sargent paper used XAS but the time resolution (seconds) cannot rule out transient Cu0/Cu+ cycling. DFT from the Nørskov group suggests CO dimerization on Cu(100) terraces is rate-limiting, not the oxidation state per se.
Replying to @chemist_li
The 0.3 MJ/kg figure is impressive but note: that is for flue gas (10% CO2). DAC at 420ppm is 3-5x more energy-intensive. Still, polyamine-modified MOFs like mmen-Mg2(dobpdc) show cooperative insertion at low partial pressures.