curate-analysis-type

name	curate-analysis-type
description	Review datasets whose analysisType is missing or set to "not specified", recommend an existing analysisType when strongly supported, otherwise keep "not specified" and optionally suggest a new category for schema review.
allowed-tools	["Skill","mcp__bioenergy-datasets__search_datasets","mcp__bioenergy-datasets__get_dataset"]

Curate dataset analysisType

Use this skill to review datasets with missing analysisType or analysisType: "not specified".

This skill is read-only. Do not modify dataset records. Produce a running recommendation list for curator review.

Allowed current analysisType values

These values are currently valid in the schema:

• affinity_purification
• cross_linking
• Expression profiling
• Genomic - SNP calling
• image_analysis
• Ms_imaging
• shotgun_proteomics
• srm_mrm
• swath_ms
• Targeted Locus (Loci)
• not specified

not specified is a valid outcome. Do not force a classification when the evidence is weak or ambiguous.

Scope limit

Review a maximum of 5 datasets per run.

If more than 5 candidate datasets are returned, process only the first 5 results from the current search unless the user explicitly asks for another batch.

Goal

For each reviewed dataset, determine one of the following outcomes:

1. Assign existing value

   • The metadata strongly supports one existing allowed analysisType value.

2. Leave as not specified

   • The metadata is weak, ambiguous, incomplete, or does not justify a confident assignment.

3. Suggest new value for schema review

   • The metadata strongly supports a coherent analysis category, but none of the currently allowed values fits well.
   • In this case, still keep the recommended analysisType as not specified and record the suggested new value separately.

Required output format

Maintain a running list. For each dataset reviewed, record:

• uid
• identifier, if available
• title, if available
• current analysisType
• recommended analysisType
• confidence: high, medium, or low
• outcome:
  • assign existing value
  • leave as not specified
  • suggest new value for schema review
• suggested new value, only when outcome is "suggest new value for schema review"
• evidence summary: 1 to 3 short sentences based on the metadata

Use this exact template for each entry:

UID: Identifier: <identifier or "not available"> Title: <title or "not available"> Current analysisType: <current value or "not specified"> Recommended analysisType: <existing enum value or "not specified"> Confidence: <high|medium|low> Outcome: <assign existing value|leave as not specified|suggest new value for schema review> Suggested new value: <value or "none"> Evidence:

Workflow

1. Call search_datasets to find candidate datasets with:

   • analysisType = "not specified"
   • optional brc filter if the user requests a specific center
   • optional text query if the user wants to focus on a topic

2. Limit the review to at most 5 datasets from the current result set.

3. For each selected dataset, call get_dataset to inspect the full record.

4. Review the metadata carefully. Prefer the following evidence sources:

   • title
   • summary or description
   • methods
   • platform
   • instrument
   • keywords
   • assay details
   • source or repository metadata
   • any fields that indicate the experimental or computational analysis type

5. Decide the outcome conservatively:

   • If one existing enum value is strongly supported, recommend it.
   • If no existing enum fits and a new category is strongly evidenced, recommend not specified and separately suggest a new value.
   • If the evidence is weak or mixed, recommend not specified.

6. Continue building the running list for only those 5 datasets.

Classification guidance

Use conservative mapping. Prefer precision over coverage.

• image_analysis

  • Use for image-derived analysis workflows, microscopy image analysis, computer vision on experimental images, or other image-centric analyses.

• shotgun_proteomics

  • Use for untargeted proteomics datasets, especially LC-MS/MS proteomics where broad protein identification is the goal.

• Expression profiling

  • Use for transcriptomics, gene expression profiling, RNA expression measurements, or expression-focused omics datasets.

• Genomic - SNP calling

  • Use for datasets centered on variant calling, SNP detection, or genotype calling.

• Ms_imaging

  • Use for mass spectrometry imaging workflows.

• srm_mrm

  • Use for targeted SRM/MRM mass spectrometry.

• swath_ms

  • Use for SWATH or DIA proteomics.

• Targeted Locus (Loci)

  • Use for targeted locus-based sequencing or locus-specific assays.

• affinity_purification

  • Use only when affinity purification is clearly part of the experimental design or analysis.

• cross_linking

  • Use only when cross-linking is clearly described in the workflow.

Rules

• Always use MCP tools to access dataset data.
• When reading search results, always use the uid field for subsequent get_dataset calls.
• Never use the id field from search results as input to get_dataset.
• Prefer the smallest result set needed for triage.
• Do not call the API directly using curl or other shell commands.
• Do not construct HTTP requests manually.
• For Recommended analysisType, use only a currently allowed schema value or not specified.
• If no current schema value fits well, keep Recommended analysisType as not specified.
• If the metadata strongly supports a coherent category that is missing from the current schema, record it separately in Suggested new value.
• Do not treat Suggested new value as an assigned dataset value.
• Do not update any dataset records. This skill is read-only and produces recommendations only.
• Prefer leaving analysisType as not specified over making a weak or uncertain assignment.
• Never review more than 5 datasets in a single run unless the user explicitly requests another batch.
• Keep evidence summaries short, concrete, and grounded in the dataset metadata.