Skip to content

implementation.md

Latest commit

 

History

History
34 lines (21 loc) · 4.93 KB

implementation.md

File metadata and controls

34 lines (21 loc) · 4.93 KB
Raw

GDSC Anchor processing decisions

Public original and validation raw wells are normalized by scan as (intensity - mean B positive control) / (mean NC-1 negative control - mean B positive control), clipped to [0,1]. Repeated tag records become one physical well. Anchor-only controls provide observed anchor viability. Mixed-treatment component doses remain in the response keys; curve fitting uses the first component dose as the reference script does. Day-1 and fitted/portal tables are comparison inputs, not substitutes for raw wells. A 600-record peer-normalized sample had 596 public matches, agreeing within 5.6e-16; the four unmatched records relate to known public-release coverage differences.

Expression uses the original public CMP rnaseq_all_20220624.zip processed FPKM, TPM and read-count matrices. The workflow downloads and imports this archive directly. Microarray and CEL processing are excluded; no local processed microarray file or archived PSet is required.

Package and measurement conventions

  • packages.PharmacoGx selects PR #178, branch feat/mcp. Validation installed commit 28c58707eece80086bee75d8ef0e8afae96389ca (3.14.1); CoreGx main resolved to bf6d814abd3dc0fc98f9f419f430cc6473046cb2 (2.7.3).
  • renv.lock pins the tested R dependencies, Bioconductor release and exact CoreGx/PharmacoGx commits. pixi.lock pins the system and R environment; setup restores into the project’s Pixi library and verifies the resolved package versions.
  • PR178 Hill fits retain E0, E_inf, slope, EC50, IC50, fit status and Rsquare. Fits use responses clipped to [0,1]; source measurements remain available. Doses retain their source unit in the TRE.
  • AUC means normalized viability area under PR178; AAC means normalized response area. Both are exported under distinct names.
  • HSA/Bliss observed and recomputed fitted references are separate. Fitted predictions use the fitted E0. Positive excess is reference viability minus combination viability. Missing monotherapy references remain missing.
  • The package's Loewe and ZIP APIs assume E0=1. Comparison columns explicitly use the Loewe_E0_1 / ZIP_E0_1 prefix and must not be interpreted as predictions from the complete four-parameter fit. They are computed for eligible Matrix/ALMANAC pairs, following the peer API path. Negative/nonfinite slopes or invalid doses/EC50 are excluded.
  • Source IDs remain primary keys. AnnotationDB enriches cell lines and compounds; ambiguous matches remain unmapped. AnnotationGx fields are NA because that method was not run. Gene mapping uses the configured AnnotationDB-hosted GENCODE v50 reference; gene In.AnnotationDB denotes membership in that reference, not a separate API lookup.
  • Unique current Ensembl IDs are used for gene-level assays. Multiple features mapping to a gene use lexical-first source feature, with an exclusion audit. Fusions retain paired gene IDs. Heterogeneous DTP targets retain source feature IDs and separate method/unit assays.
  • Molecular batchid identifies the source processing cohort, not an inferred experimental batch. Variant zeros mean no variant reported in the source, not confirmed wild type.

Implementation and validation

The scripts implement download, extraction, response processing, annotation, curve fitting, synergy, molecular import, PSet construction and QC. A script being implemented does not imply that its full dataset run passed. See validation.md for the latest run state. Generated logs, fit checkpoints, receipts and mapping audits live under data/, outside Git.

Downloads are atomic, checksum-verified and resumable. Original archives are retained. Extraction rejects traversal/symlink entries and streams only gene records from GTF to avoid a multi-gigabyte uncompressed annotation. Expensive fits checkpoint complete chunks. A failed fit is retained with its error status; all-failed runs stop.

Four repositories are retained because response schemas and normalization differ substantially. Small utilities remain repo-local so each repository can run independently. No peer-created RDS or prebuilt PSet is a production input. PSet data and rendered reports are distributed separately from this source repository; no ORCESTRA upload has been performed.

Executable source cleanup

Reference-only Anchor portal/fitted/day-1 tables and ALMANAC field documentation were removed from the production sources list. Their existing downloads and source links remain in the historical provenance notes. Matrix portal tables remain required for compound/model metadata.

Four-parameter dose support

Three distinct doses preserve the peer fitting threshold, but cannot independently identify four Hill parameters. QC reports four_parameter_dose_support and dose counts; successful optimization is not evidence that the fit is identifiable. ALMANAC observed monotherapy means are floored at zero without an upper cap; fitting clips to [0,1], while raw percent growth remains intact.