How It Works
Athnium Computational Methodology & Architecture
1. Bridging the Screen to the Wet Lab
Athnium Explorer is a specialized workspace built for synthetic chemists. By transforming abstract molecular notations (SMILES) into interactive physical models in seconds, we eliminate the guesswork of reaction planning. What typically requires intensive modeling or multiple rounds of failure can now be mapped instantly.
2. 3D Electrostatic Potential Profiles
Using RDKit, the engine builds 3D conformers for any SMILES input, adds implicit hydrogens, and runs MMFF force-field optimizations. It then computes **Gasteiger partial charges** across the molecule. Atoms are instantly color-coded to map potential: electron-poor sites (red) seek reactions, while electron-rich sites (blue) drive attacks.
3. Steric & Electronic Ligand Recommender
Choosing the right supporting ligand is the difference between a successful cross-coupling reaction and a flask of dead reagents. Athnium calculates ligand compatibility indices based on substrate steric volume (Molecular Weight proxy) and lipophilicity (logP proxy), providing an instant ranked leaderboard of phosphine helpers (such as XPhos, PtBu₃, PPh₃) for critical Suzuki-type reactions.
4. Decoupled, Swappable API Brain
We design software around strict data contracts. Right now, Athnium Explorer utilizes heuristic calculators to verify the concept. When our machine learning team deploys the deep neural AI models (predicting exact isotropic shielding values $\sigma$ and Electric Field Gradients), we only have to redirect a single environment variable.
5. Models & Data Credits
Athnium’s Stage 1 workspace integrates open-source scientific tools, machine learning pipelines, and datasets. We gratefully acknowledge the following works that power our computations:
AIMNet2 models provide near-DFT accuracy for atomic partial charges and molecular descriptors.
Provides sub-ppm carbon-13 shifts and proton shift predictions across broad chemical spaces.
Aizynthfinder powers complex retrosynthetic search trees, while Chemformer drives forward reaction prediction.
Powers substructure search, 2D coordinates generation, drawing generation, descriptors, and filters.