Label-Free Quantification (LFQ)

Label-free quantification estimates protein abundance directly from MS1 ion intensities, without chemical labeling. quantms implements LFQ through OpenMS ProteomicsLFQ, which handles chromatographic feature detection, cross-run alignment, match-between-runs (MBR), and protein-level aggregation in a single step.

---

How LFQ Works in quantms

1. Database search — Raw spectra are searched against a protein FASTA database (Comet, MS-GF+, or Sage)
2. PSM rescoring — MS2PIP + DeepLC + Percolator improve identification rates by 10–30%
3. Feature detection — Chromatographic features (MS1 isotope patterns) are detected per run using the OpenMS FeatureFinderIdentificationBased algorithm
4. Feature alignment — Features are aligned across runs using retention time correction to account for LC variation
5. Match-between-runs — Peptides identified in one run are transferred to runs where the same feature is detected but not identified, reducing missing values
6. Protein quantification — Peptide intensities are aggregated to the protein level (MaxLFQ algorithm by default)

---

Feature Detection and Alignment

quantms uses identification-guided feature detection: chromatographic features are seeded by confident PSMs rather than relying on de novo peak picking across the full mass range. This improves sensitivity in complex samples.

Feature alignment uses a map-aligner to correct systematic RT shifts. With --targeted_only false (default), all high-confidence features are used for alignment. For targeted workflows where you want to restrict quantification to peptides with direct identifications, set --targeted_only true.

---

When to Use LFQ vs TMT

Aspect	LFQ	TMT/iTRAQ
Multiplexing	One sample per raw file	6–18 samples per raw file
Dynamic range	Higher (no ratio compression)	Moderate (ratio compression at MS2)
Missing values	More frequent across runs	Fewer (within-plex MS2 always present)
Throughput	Lower (more instrument time)	Higher (many samples per run)
Best for	Discovery, large clinical cohorts, plasma	Precise quantification, time-course, paired designs
Cost	Lower reagent cost	Reagent cost + complexity

Use LFQ when you have many samples to profile at discovery scale and can tolerate more missing values. Use TMT/iTRAQ when you need precise ratios, have many replicates to pack into a single run, or are designing phosphoproteomics studies.

---

Key Parameters

Parameter	Default	Description
--protein_quant	true	Run protein quantification (ProteomicsLFQ)
--targeted_only	false	Restrict quantification to directly identified features
--mass_recalibration	false	Perform m/z recalibration before feature detection
--add_triqler_output	false	Generate Triqler-format output for Bayesian protein quantification
--quantification_method	feature_intensity	feature_intensity or spectral_counting
--protein_inference_method	aggregation	Protein grouping: aggregation or bayesian
--protein_level_fdr_cutoff	0.05	Protein-level FDR threshold
--psm_pep_fdr_cutoff	0.05	PSM/peptide FDR before quantification

Example Run

nextflow run bigbio/quantms \
    -profile docker \
    --input experiment.sdrf.tsv \
    --database uniprot_human.fasta \
    --search_engines "comet msgf" \
    --use_ms2pip true \
    --use_deeplc true \
    --protein_quant true \
    --targeted_only false \
    --protein_level_fdr_cutoff 0.01 \
    --outdir results_lfq/

---

Expected Output Files

File	Location	Description
out.mzTab	quant_tables/	Primary result: PSMs, peptides, and protein quantities
msstats_in.csv	quant_tables/	Long-format table for MSstats differential expression
out_msstats.mzTab	msstats/	mzTab with MSstats input format
peptide_out.tsv	quant_tables/	Peptide-level quantities from ProteinQuantifier
protein_out.tsv	quant_tables/	Protein-level quantities from ProteinQuantifier
out.consensusXML	quant_tables/	OpenMS ConsensusXML (for debugging or OpenMS tools)
multiqc_report.html	pmultiqc/	Interactive QC report

See Output for a full description of each file.

---

Tips for Large Cohorts

• Use SDRF input — SDRF encodes all search parameters in a single file, preventing parameter mismatches across large studies.
• Split into batches — For datasets with hundreds of raw files, consider batching by fractionation or plate to keep alignment tractable.
• Set --targeted_only true for strictly identified-only results if you need to minimize false features in very large cohorts.
• Enable rescoring — ML rescoring (--use_ms2pip true --use_deeplc true) is especially valuable when sample quality varies across a cohort, as it adapts to each run's RT and fragmentation characteristics.
• Use Singularity on HPC — Replace -profile docker with -profile singularity for cluster execution. Use -resume to restart failed runs without recomputing completed tasks.
• Increase resources — For runs with >100 files, increase memory limits: --max_memory 128.GB --max_cpus 32.
• Reference retention time alignment — If RT alignment is poor (check the pmultiqc report), verify that the same LC gradient was used across all runs.