Feature Guide · Install & Overview¶
SynOmega exposes six capabilities. This chapter gives one section per feature with the how-to (command line and Python API); the model / algorithm behind each is in the matching research report chapter — the two chapters correspond one-to-one:
| Feature | How to use (this chapter) | How it works (research) |
|---|---|---|
| Single-step forward prediction | ↗ | ↗ |
| Multi-component evolution | ↗ | ↗ |
| Single-step retrosynthesis | ↗ | ↗ |
| Multi-step route planning | ↗ | ↗ |
| Reaction plausibility | ↗ | ↗ |
| Synthesizability score (SynScore) | ↗ | ↗ |
Install¶
pip install synomega # core: rdkit + numpy (the template-rule backend works as is)
pip install "synomega[gnn]" # + the D-MPNN neural single-step backend (torch), recommended
The neural backend is an optional extra: the template-rule backend runs
without torch; install [gnn] when you want the neural template classifier
(forward / retro / evolution / plausibility all build on it). The default model
weights and the ZINC in-stock building-block set are downloaded on first use
into ~/.cache/synomega (override with SYNOMEGA_CACHE; pick a mirror with
SYNOMEGA_MIRROR=ustc|github), not shipped in the wheel. Pre-fetch with
synomega download. Requires Python ≥ 3.10.