Tutorials#

Interactive marimo notebooks under tutorials/ in the repository. They are not rendered here; run one with

pip install "chem-tactics[tutorials]"
marimo edit tutorials/<name>.py      # or: marimo run ... for app mode

Working with the package#

Thompson Sampling benchmark

thompson_sampling_tutorial.py — pick strategies and warmups, run them on the bundled thrombin library, and compare recovery with TS_Benchmarks. Blocks 1–3, 6.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/thompson_sampling_tutorial.py
Library enumeration

library_enumeration_tutorial.py — single-step, multi-step and alternative-SMARTS pipelines, full enumeration, then a search. Block 1, with a taste of 3.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/library_enumeration_tutorial.py
Reaction config builder

reaction_config_builder.py — build a ReactionConfig in a form, validate it against your reagent files, see the failures. Block 1.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/reaction_config_builder.py
Custom evaluator tester

custom_evaluator_tester.py — paste a scoring function, check it compiles and behaves, run a short search with it. Block 2.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/custom_evaluator_tester.py
Diagnostic benchmark plots

diagnostic_benchmark_plots.py — the mechanism plots (plot_rws_diagnostic, plot_ttts_diagnostic, the layered panels) over the published diagnostic benchmark. Blocks 5–6. Needs the benchmark parquet under outputs/.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/diagnostic_benchmark_plots.py
Interactive SAR explorer

interactive_sar_explorer.py — hover a point, see the structure; per-component oracle GMIC over the ground-truth scores. Needs data/scores.

https://github.com/aakankschit/TACTICS/blob/main/tutorials/interactive_sar_explorer.py

Manuscript figures#

Reproduce the paper’s figures from the benchmark output (not the package API). Each expects the benchmark parquet files under outputs/.

  • manuscript_plots_ROCS.py — ROCS libraries: aggregate top-N recovery, per-library breakdowns, Tukey HSD, budget sensitivity.

  • manuscript_plots_docking.py — the eight docking libraries: method × library heatmap, significance, budget sensitivity.

  • manuscript_sar_plots.py — reagent score landscapes from the brute-force ground truth.