Command-Line Usage#
BATTER ships with a batter command-line tool powered by click. This page
summarises the key commands and options. Run batter --help or append --help to
any subcommand for the full syntax.
Run Workflows#
batter run examples/mabfe.yaml
Options:
--on-failure {prune,raise,retry}Control how ligand failures are handled (default:
raise).retryclears FAILED sentinels and reruns that phase once.--output-folder PATHOverride the system output directory.
--run-id TEXTOverride the execution run id (use
autoto reuse the latest).--allow-run-id-mismatch / --no-allow-run-id-mismatchAllow reuse of a run id even when the stored configuration hash differs.
--dry-run / --no-dry-runToggle
run.dry_runfrom the YAML.--clean-failures / --no-clean-failuresClear
FAILEDsentinels,job_attempt.txtretry counters, and progress caches before rerunning an execution.--only-equil / --fullRun only equilibration preparation steps. FE preparation is still performed (up to
prepare_fe_windows), but the FE equilibration/production/analyse phases are skipped.--slurm-submitEmit an
sbatchscript and submit the job instead of running locally.--slurm-manager-path PATHOptional path to a custom SLURM header/body pair for the manager submission.
--partition / -pOverride
run.slurm.partitionfrom the YAML.
Resume an existing execution (no YAML needed once seeded):
batter run-exec work/adrb2/executions/rep1
Notes:
The first
batter runstores a copy of the YAML plus any external restraint files (e.g.,extra_conformation_restraints) underartifacts/config/.run-execreuses that copy.
Generate Batch Scripts#
Use batter batch to emit an sbatch script that runs run-local-batch.bash (non-REMD)
across one or more execution folders:
batter batch -e work/adrb2/executions/rep1 -e work/adrb2/executions/rep2
Use --remd to switch to REMD mode (runs run-local-remd.bash):
batter batch --remd -e work/adrb2/executions/rep1 -e work/adrb2/executions/rep2
The command writes run-local-batch.bash into each component folder using the packaged
template and skips components that already contain FINISHED (or where all windows
are marked FINISHED).
Key options:
--gpusTotal GPUs to request (defaults to the total window count detected).
--gpus-per-nodeGPUs available per node (default: 8). Used to size per-task node allocations when
MPI_EXECissrun.--nodesOverride the total node count in the header.
--remdUse REMD execution mode (
run-local-remd.bash) instead of standard batch mode.--auto-resubmit/--no-auto-resubmitWhen enabled (default), the generated sbatch traps a pre-timeout signal, regenerates the batch script, and resubmits it until all components finish or the max resubmission count is reached.
--signal-minsMinutes before the time limit to trigger auto-resubmit (default: 90).
--max-resubmit-countMaximum total submissions for the script (including the first run; default: 4).
--current-submission-timeInternal counter for auto-resubmit; increments on each resubmission (default: 0).
Inspect Free-Energy Results#
List results stored in a portable work directory:
batter fe list work/adrb2
Use --format to output JSON/CSV/TSV instead of a table.
Display a specific FE record:
batter fe show work/adrb2 run-20240101 --ligand LIG1
If the run contains multiple ligands, pass --ligand to disambiguate.
Re-run FE analysis for a saved execution:
batter fe analyze work/adrb2 run-20240101 --ligand LIG1 --workers 4
To analyze every run under work/adrb2/executions (instead of one run), omit
run_id:
batter fe analyze work/adrb2 --workers 4
Use --workers to control parallelism and --analysis-start-step to skip early
production steps in each window. By default existing analysis outputs are preserved;
pass --overwrite to regenerate them. Pass --n-bootstrap to request MBAR
bootstrap resamples and --no-raise-on-error to continue if one ligand fails.
Re-run FE analysis for exactly one ligand folder:
batter fe ligand-analyze work/adrb2/executions/run-20240101/simulations/LIG1 --overwrite
ligand-analyze also accepts directories outside executions/<run_id> as long
as they contain an fe/ folder.
Clone Executions#
Duplicate the directory structure of an execution (useful for what-if experiments):
batter clone-exec work/adrb2 run-20240101 run-20240101-test --mode symlink --only-equil
Only metadata, inputs/params, equilibration artifacts, and a minimal fe/ scaffold are copied,
so the clone can start new simulations without recreating large FE dumps.
SLURM Utilities#
batter seed-headers– Seed packaged SLURM headers into~/.batter(or a custom directory via--dest). Use--forceto overwrite existing headers.batter diff-headers– Show a unified diff between your headers and the packaged defaults (use--destto point at a custom header directory).batter report-jobs– Summarise SLURM jobs launched by BATTER. Use--detailedto show per-job information and--partitionto filter by queue.batter cancel-jobs --contains TEXT– Cancel SLURM jobs whose name contains the supplied substring (matches thefep_...job names produced by BATTER).
FE Toolkit Schedules#
BATTER wraps the fetkutils-tischedule.py script from AMBERTOOLS via batter fek-schedule so
you can optimise or analyse lambda schedules without leaving the main CLI.
batter fek-schedule \
--opt 96 \
--ar --ssc --sym \
--start 4 --stop 8 \
-T 310 \
--out sched.ar.z.dat \
--plot sched.ar.z.png \
ADRB2_I/rep1/fe/pose0/sdr/z
Key options:
--opt N/--read FILEOptimise a fresh schedule with
Nlambda values or analyse an existing schedule.--pso|--ar|--klChoose the optimisation metric (phase-space overlap, replica-exchange ratio, or exp(-KL)).
--ssc/--sym/--alpha0/--alpha1Restrict optimisation to SSC(alpha) families, optionally symmetric around 0.5.
--plot PATH/--out PATHSave the interpolated heatmap and the digitised schedule to disk.
All options mirror the original script, so existing workflows can switch to the new command
without rewriting job scripts. Use --verbose to surface the diagnostic output emitted by
the underlying fetkutils routines.
Environment Notes#
The CLI uses
batter.apiand is safe to call from activated conda/virtualenv installations.batter runvalidates the YAML before invoking the orchestration layer.When submitting to SLURM, the generated manager script is rendered from
job_manager.header/job_manager.bodyinto<hash>_job_manager.sbatchusing a hash of the YAML contents and overrides.