ucdavis-proteomics-core-pipeline

DE-LIMP: Differential Expression & Limpa Proteomics

Find which proteins are significantly different between your experimental conditions -- upload a DIA-NN output file and get interactive volcano plots, heatmaps, pathway enrichment, and AI-powered interpretation, all without writing code.

Built on R Shiny with the limpa pipeline for normalization and protein quantification, and limma for statistical testing with FDR correction. See USER_GUIDE.md for methodology details.

Web app input: DIA-NN report.parquet | Also supported (local/HPC install): DDA search via Sage + de novo sequencing (Cascadia/Casanovo) | Not for: TMT/iTRAQ, Spectronaut/MaxQuant output

Not sure if your data is DIA? If your core facility used DIA-NN to process your samples, you have DIA data. Look for a report.parquet file in your results folder. If your data was processed with MaxQuant, Spectronaut, or Proteome Discoverer, or if you used isobaric labels (TMT, iTRAQ), DE-LIMP is not the right tool.

Try it now: huggingface.co/spaces/brettsp/de-limp-proteomics -- no installation required

Project Website: bsphinney.github.io/DE-LIMP | Docs: USER_GUIDE.md | CLAUDE.md

What's New in v4.0.0

De novo sequencing + DDA database search -- DE-LIMP now goes beyond DIA differential expression. New de novo (Cascadia / Casanovo) and DDA (Sage) workflows add per-spectrum sequencing, Sage-vs-de-novo agreement views, and homology-based species identification (DIAMOND against nr with LCA assignment) — with a decoy-spectra-calibrated FDR so you can report confirmed peptides at a controlled error rate. Alignment views render only real BLAST alignments, never fabricated positions.

Proteogenomics — Build sample-specific FASTA databases from your matched RNA-seq -- Upload RNA-seq fastq files or provide SRA/SLIMS accessions, and DE-LIMP runs an HPC pipeline chain (fastp → bowtie2 rRNA filter → STAR → stringtie → merge → gffcompare → gffread → TransDecoder) to predict novel ORFs and alternative splice variants. The final FASTA auto-assembles with your canonical proteome (UniProt or NCBI RefSeq), registers in the FASTA library catalog, and shows up as a "Proteogenomics DBs" option in the main search page's FASTA Database dropdown. Reference genomes: Human, Mouse, Pig, Rat, Arabidopsis (Bovine and Maize building). Multi-user catalog discovery via "Restore from Hive" and "Discover from Hive" buttons. Includes an "Explain this workflow" modal for proteomics users new to RNA-seq.

Two analysis pipelines, one app -- Choose between DPC-Quant + limma (limpa's detection-probability model, default) and MaxLFQ + limma (paper-faithful Moschem et al. 2025 implementation). The pipeline you pick is recorded in the dataset itself; methods text, AI prompts, exports, and the Reproducibility log all describe whichever pipeline actually ran -- no hardcoded "DPC-Quant" strings anywhere.

QuantUMS quality filters -- Optional pre-filtering of DIA-NN precursors by eQ, qQ, and pgQ quality scores (Moschem 2025). Applied at pipeline run-time on the parquet, with a waterfall showing how many precursors and proteins survived each filter. Defaults to off; opt-in checkbox lets you feed filtered data through limpa's DPC-Quant if you want to test the combination.

On/Off Proteins panel -- New sub-tab in DE Dashboard surfaces proteins detected in ≥N samples of one condition AND zero samples of the other. limma assigns these NA logFC, so they're invisible in the volcano -- the new panel makes them findable.

Coverage filter -- Drop proteins with fewer than X non-NA samples before limma fits the model (UC Davis Bioinformatics Core convention). Live waterfall shows protein retention. Available in MaxLFQ + limma mode.

Run Comparator pipeline-aware -- Cross-tool DE comparison (DE-LIMP vs DE-LIMP / Spectronaut / FragPipe) now reads the pipeline descriptor on both sides. Hypothesis-engine rules (rollup, normalization, peptide rules) emit the correct contrast based on whether each side used MaxLFQ or DPC-Quant.