AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

relion

Validates and executes RELION single-particle cryo-EM refinement and half-map postprocessing. Supports STAR optics/acquisition checks, particle-stack consistency, gold-standard half sets, soft-mask validation, diagnostic Fourier shell correlation, and restart guidance.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--relion at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 3

references/acquisition-and-restarts.md

Acquisition, job selection, and restart decisions

Before extracted particles

Keep the project's movie/micrograph paths, optics groups, dose information, and gain/detector metadata. For movies, confirm physical and super-resolution pixel sizes separately; a factor-of-two pixel-size error propagates into defocus, extraction diameter, map scale and reported resolution. For dose weighting, confirm dose per frame and pre-exposure rather than copying another dataset's settings. Inspect drift trajectories, damaged frames and residual motion after correction.

RELION has an internal CPU motion-correction implementation and wrappers for external tools. For CTF estimation, RELION 5.0 documentation specifies CTFFIND 4.1.x, not CTFFIND 5.x. A successful wrapper launch does not validate a CTF fit: inspect Thon rings, astigmatism, fit resolution, and defocus against acquisition expectations. Rejecting all high-defocus particles can introduce a selection bias; use documented criteria appropriate to the specimen and target resolution.

Before 3D auto-refine, select a reasonably homogeneous particle population using actual 2D/3D classification results. Do not interpret class occupancy as quantitative state populations without considering selection and alignment bias. Avoid imposing high symmetry solely because it improves nominal resolution. Use C1 when symmetry is unknown and test a justified alternative separately.

Path and sampling invariants

A STAR file in Extract/job010/ often points to paths relative to the project root. Moving only the STAR file does not move its dependencies. The bundled validator takes --project to make this explicit. It checks stack indices/box dimensions; optics pixel size is the sampling authority for particle processing and still needs acquisition/re-extraction verification.

After resizing a particle box, adjust image pixel size to preserve physical field of view and resample the initial reference consistently. Do not merely edit an MRC header to make a mismatch disappear. Keep translations in their declared units (Origin*Angst versus legacy pixel fields). For RELION's centered coordinate conventions, follow the official conventions page rather than interpreting map array indices as laboratory coordinates.

_rlnCoordinateX/_rlnCoordinateY are pixels in the aligned, summed micrograph, not the movie's super-resolution grid. RELION's Euler angles rotate the reference into the observed particle; origin offsets shift the observation into the reference projection, before rotation. The actual 5.0.1 labels are _rlnOriginXAngst/_rlnOriginYAngst (the conventions page spells the suffix out as Angstrom; the label definitions are authoritative). Avoid sign changes or Å-to-pixel conversion by intuition.

When merging optics tables, reconcile group names and acquisition settings, then remap IDs; the same numeric ID in two projects need not denote the same group. The validator is read-only and does not merge tables. RELION distinguishes maps (.mrc) from particle stacks (.mrcs), and the helper supports real-valued MRC modes including float16 mode 12. Native RELION can read packed mode 101, but mrcfile cannot; convert that input with a suitable native utility first.

RELION uses right-handed coordinates, but reconstruction handedness still requires independent assessment (for example, recognizable helix chirality at sufficient resolution or tilt data). FSC alone cannot choose the hand. If a justified hand inversion is needed, transform both half maps, mask and downstream coordinates consistently, and preserve the original maps.

Restart a stopped refinement

A completed iteration's _optimiser.star points to the sampling/model/data state required for continuation. Keep those referenced files and their relative paths together. Preserve the original half-set split. An example continuation, from the same project root with the original MPI runtime:

mpirun -np 3 relion_refine_mpi \
  --continue RefinePilot/run_it005_optimiser.star \
  --o RefineContinue/run --j 2

The iteration filename is illustrative: use the last intact checkpoint actually present and a new output prefix. Inspect the failing job log before deciding whether the cause was transient I/O, insufficient memory, invalid CTF metadata, an absent external tool, or scientific nonconvergence. Change one relevant condition, preserve the old job, and resume that checkpoint. Do not repeatedly restart a deterministic invalid-input failure.

The GUI's Continue operation uses the existing job directory; refinement continuations get prefixes such as run_ct23. The illustrative direct command above instead requests a separate output prefix. It does not register a job in default_pipeline.star, so retain the command, version/build, output prefix and source checkpoint alongside the results. This helper neither submits scheduler jobs nor implements RELION's External-job marker/output-node contract. Use the native GUI or Schemes when pipeline tracking is required; no remote service, HTTP API, authentication or pagination is involved here.

Reporting resolution

Report the FSC threshold and mask correction, pixel size, particle count, imposed symmetry and processing version. Keep both original half maps, the postprocessing STAR/PDF and mask. A scalar FSC estimate can hide preferred orientation and regional flexibility; inspect local and directional resolution and map-model validation before making structural claims. A high unmasked correlation from duplicated or coupled half maps is a failure of independence, not a high-resolution result. Independent half-set histories must also survive re-extraction and restart: renumbering particles or copying stacks does not create independent observations. The helper detects repeated references, not duplicate image content stored under distinct filenames.