nfcore-laya-decisions

nf-core & Nextflow Laya Decision Dataset (nfcore-laya-decisions-v3)

A standardized, multi-task decision-layer dataset for training Laya (System-1 Decision Engines) and fast autonomous agents across Nextflow DSL2 and the complete nf-core bioinformatics ecosystem.

Dataset Summary

• Total Decision Records: 11,654 records across 8 core architectural pillars: 1. QC Read Adaptation (1,500 records): Illumina NovaSeq/NextSeq, PacBio HiFi CCS, Oxford Nanopore Direct RNA/DNA quality control adaptation and tool swapping. 2. Resource Profiling & Directives (1,500 records): Process memory, CPU, time, and dynamic retry/error-strategy directive rules across genome-scale tools. 3. Subworkflow Packaging (622 records): Determining when to bundle multi-process tasks into local subworkflows or reuse official nf-core subworkflows. 4. Samplesheet Schema & Header Inference (1,950 records): JSON Schema draft-07 type, format, pattern, and enum constraints for bioinformatics samplesheets. 5. Tool & Module Selection (4,426 records): Spans 2,153 nf-core modules and subworkflows with mined hard-negative distractors. 6. DSL2 Syntax & Rules Check (800 records): Balanced binary noul decisions evaluating idiomatic DSL2 patterns vs anti-patterns. 7. nf-core Pipeline Routing (656 records): Matching biological assay requirements to official nf-core workflows. 8. Intent Routing (200 records): Disambiguating compilation, remediation, parameter tuning, and execution commands.

Split Information

Data Schema

Each line in data/*.jsonl is formatted for direct consumption by Laya agents and Hugging Face datasets:

{
  "id": "qc_adapt_ont_nanoplot_1_34",
  "category": "qc_read_adaptation",
  "state": {
    "assay": "Direct RNA sequencing on Oxford Nanopore PromethION",
    "tool": "FastQC",
    "read_type": "long_reads_direct_rna"
  },
  "question": {
    "type": "choice",
    "instructions": "How should QC step FastQC be configured given sequencing characteristics: long_reads_direct_rna?",
    "criteria": {
      "Keep FastQC": null,
      "Drop FastQC": null,
      "Swap for NanoPlot": null
    }
  },
  "target": "Swap for NanoPlot",
  "target_idx": 2
}

Quickstart

Load via Hugging Face datasets

from datasets import load_dataset

dataset = load_dataset("Primeomicx/nfcore-laya-decisions")
print(dataset)

Trained Model

The fine-tuned model checkpoint trained on this dataset is available at: Primeomicx/laya-nextflow-nfcore