Worked Examples (Glide)

One command screens a make-on-demand chemical space against a target with your own Glide, running the resumable propose → dock → ingest loop.

One command screens a make-on-demand chemical space against a target with your own Glide, running the resumable propose → dock → ingest loop for a fixed number of iterations. Everything here is customer-facing: the optimizer runs in the licensed container; docking runs on your host in your Schrödinger environment (Schrödinger is never inside the image).

Three worked targets are wired end-to-end — KIF11, PYRD, TGFR1 — each with a real example Glide grid and the same drug-like property filter and per-target hit threshold we use internally.

TL;DR

# from the repo root, after: install → login → pull → update-license → data install
export SCHRODINGER=/opt/schrodinger2026-1              # your install
examples/run_navigator.sh TGFR1                        # gamma, 100k budget, 10 rounds, Glide

Prerequisites

navigator installed and the image pulled (./install_navigator.sh, navigator login, navigator pull).

A valid license activated (navigator update-license …, navigator verify-license).

The database installed once:

navigator data install freedom-space-5    # ~1.8M synthons, encrypted; decrypts locally

For real docking, SCHRODINGER pointing at your install (Glide + LigPrep). No Schrödinger? Use --scorer mock to watch the loop turn (a stand-in score, not chemistry).

The one command

examples/run_navigator.sh <TARGET> [options]

Option	Values	Default	Meaning
`<TARGET>`	`KIF11` `PYRD` `TGFR1`	—	which `examples/configs/<TARGET>.json`
`--method`	`gamma` `alpha` `beta` `analog` `all`	`gamma`	strategy (or all four)
`--budget`	`10k` `100k` `1m` or an integer	`100k`	total molecules docked
`--iters`	integer	`10`	propose/dock/ingest rounds (batch = budget ÷ iters)
`--database`	`db@release`	`freedom-space-5@2026-03-296b.2`	installed release to screen
`--scorer`	`glide` `mock`	`glide`	real Glide, or the no-Schrödinger stand-in
`--pool`	integer	`20000`	surrogate candidate-pool per round
`--gpu`	flag	off	XGBoost surrogate on CUDA (falls back to CPU if absent)
`--precision`	`HTVS` `SP` `XP`	`HTVS`	Glide precision (overrides docking_settings)
`--status`	flag	—	print status of this target's runs and exit

Examples

examples/run_navigator.sh TGFR1                          # default: gamma, 100k, Glide HTVS
examples/run_navigator.sh KIF11 --budget 10k --gpu       # quick 10k run, GPU surrogate
examples/run_navigator.sh PYRD  --method all             # gamma+alpha+beta+analog, separate runs
examples/run_navigator.sh TGFR1 --scorer mock --budget 200 --iters 2   # smoke, no Schrödinger
examples/run_navigator.sh KIF11 --database enamine-real-v5a@2026-07-02.1  # a different space

What it does

Each round: the container proposes a batch (surrogate-guided, diversity-aware), you dock it with Glide on the host, the container ingests the scores and refits — for --iters rounds. State lives in runs/<target>_<method>_<budget>_<scorer>/.

Resumable. Re-run the exact same command to continue after a Ctrl-C, reboot, or spot reclaim — nothing already docked is re-docked, and a batch that was proposed but not yet scored is recovered automatically.
Compiled logs. Progress prints one clean line per phase and is appended to runs/<run>/pipeline.log; the optimizer prints single-line errors by design. Set DMC_NAV_DEBUG=1 in .env only when you need full tracebacks.
Property filter. The config's space.property_constraints is the additive (generation-time) drug-like prefilter — approximate by design. Reactive-group / exact-structure exclusions are left to your scoring step.

Bring your own target

Copy a config, point space.database at your installed release (or your own space.reactions_path/synthons_path under ./inputs), set objective, and drop your Glide grid into examples/docking/ with a matching docking_settings.json entry. glide_batch.py is yours to adapt to your site's Glide/scheduler — the optimizer only consumes the returned batch_id,product_id,status,score CSV.

Files

Path	Role
`run_navigator.sh`	the one-command pipeline (host orchestrator)
`configs/{KIF11,PYRD,TGFR1}.json`	per-target run configs (Freedom + property filter + gamma)
`docking/docking_settings.json`	per-target Glide grid / precision / pH
`docking/glide-grid_.zip`, `glide-dock_.in`	example grids (replace with your own)
`scoring/glide_batch.py`	your Glide adapter (LigPrep+Epik → Glide → scores); resume-safe
`scoring/mock_score.py`	dependency-free stand-in scorer (smoke only)