← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-metabolomics-lipidomics / references / usage-guide.md

Lipidomics Usage Guide

Overview

Lipidomics is a combinatorial-structure problem wearing a quantification problem's clothes. This skill keeps the agent honest about two things software routinely overstates: the structural-resolution level a lipid name actually carries (the shorthand separator space -> _ -> / -> (9Z) encodes what was measured, and sn-position is almost never measured), and quantification (one isotope-labeled internal standard per class is non-negotiable because ESI response is head-group-dependent). It guards against in-source-fragment phantom lyso-lipids, sn over-claims, ether/plasmalogen ambiguity, and invalid cross-class comparisons.

Prerequisites

# R (primary)
# BiocManager::install("lipidr")

# Nomenclature parsing
pip install pygoslin

Conceptual prerequisites: a quantified lipid table (Skyline, MS-DIAL, or LipidSearch export) with sample group annotations, and knowledge of which internal standards were spiked and at what stage. To trust any sn or double-bond claim, the agent needs to know whether EAD/OzID/PB/UVPD data were acquired - under routine CID they were not.

Quick Start

Tell your AI agent what you want to do: - "Load my lipidomics table, normalize within class, and find lipids changing between groups" - "Canonicalize these lipid names and downgrade any that over-claim sn-position" - "Check whether my elevated LPC signal is real or an in-source fragment of PC" - "Design an internal-standard strategy for class-based quantification"

Example Prompts

Annotation honesty

"Parse these lipid names through Goslin and report the structural-resolution level each one actually claims." "Re-emit my LipidSearch output at molecular-species level since we only ran CID - drop the sn slashes." "Flag any plasmalogen (P-) call that lacks vinyl-ether diagnostic evidence."

Quantification

"Normalize each lipid class to its matched internal standard, not a single global standard." "Is comparing PE to PC abundance valid in my dataset given the standards I used?" "Set up class-based quantification using my EquiSPLASH internal standards."

Differential and enrichment analysis

"Run differential lipid analysis between treatment and control and make a class-faceted volcano plot." "Test whether any lipid class, chain length, or unsaturation pattern is enriched among the changed lipids."

Artifact triage

"My LPC pool is unexpectedly high - check retention-time co-elution against the parent PCs." "This apparent odd-chain PC 33:1 - is it real or an isotope/in-source artifact?"

What the Agent Will Do

  1. Canonicalize names through Goslin and assign each an honest resolution level, defaulting down where evidence is absent.
  2. Set up class-based internal-standard normalization (one IS per class) or PQN, and state semi-quant vs absolute.
  3. Run de_analysis with an explicit contrast and class/chain-aware visualization.
  4. Run lipid set enrichment (class / chain length / unsaturation) with lsea.
  5. Triage in-source fragments, odd-chain sums, and ether/plasmalogen calls before reporting.
  6. Export results with class, chain, and resolution-level metadata.

Tips