brettsp/de-limp-proteomics

DE-LIMP: Differential Expression & Limpa Proteomics 🧬

An interactive R Shiny application for differential expression analysis of DIA-NN proteomics data. Built on Limpa (a Bioconductor package for DIA proteomics normalization and quantification) and Limma (a widely-used statistical framework for differential expression), with optional AI assistance (Google Gemini or an OpenAI-compatible endpoint).

✨ What's New in v4.0.0

• —🧪 De novo sequencing + DDA search: Beyond DIA differential expression — de novo (Cascadia / Casanovo) and DDA (Sage) workflows with per-spectrum sequencing, Sage-vs-de-novo agreement views, and homology-based species ID (DIAMOND vs nr + LCA) calibrated with a decoy-spectra FDR. Alignment views show only real BLAST alignments.
• —🧬 Proteogenomics — Build sample-specific search databases from RNA-seq: Upload fastq files or SRA/SLIMS accessions. DE-LIMP runs a complete HPC pipeline (fastp → bowtie2 rRNA filter → STAR → stringtie → merge → gffcompare → gffread → TransDecoder) to predict novel ORFs and alternative splice variants. Final FASTA auto-assembles with UniProt or NCBI RefSeq and registers in the FASTA library as "Proteogenomics DBs" on the main search page. Supports Human, Mouse, Pig, Rat, Arabidopsis. Multi-user catalog discovery via "Restore from Hive" and "Discover from Hive" buttons.

Previous highlights (v3.10): Two analysis pipelines (DPC-Quant or MaxLFQ), QuantUMS quality filters, On/Off proteins, Coverage filter, FASTA picker, provenance block. (v3.7): Contaminant Analysis (keratin flagging + heatmap), Data Explorer (quartile profiles + sample scatter), NCBI Proteome Download with gene mapping, SSH File Browser, No-Replicates Mode, Environment Badge.

Earlier (v3.5): Run Comparator, Search & Analysis History, Chromatography QC. (v3.1): UI overhaul, accordion sidebar, DE Dashboard sub-tabs. (v3.0): Multi-Omics MOFA2 | Phosphoproteomics (site-level DE, KSEA, motifs) | GSEA 4-database | AI Summary | XIC Viewer (local/HPC only)

🚀 Features

📊 Interactive Analysis

• —Volcano Plots - Fully interactive (Plotly). Click or box-select to highlight
• —Heatmaps - Auto-scaled Z-score heatmaps of significant proteins
• —Contaminant Analysis - Summary cards, bar charts, keratin flagging, and contaminant heatmap
• —Data Explorer - Quartile abundance profiles and sample-sample scatter plots
• —QC Trends - Monitor run quality with group averages
• —Multi-Protein Violin Plots - Compare expression distributions

🤖 AI-Powered Analysis (Google Gemini or an OpenAI-compatible endpoint)

API Key Required: Choose a provider in the AI Chat sidebar and bring your own key — a free Gemini key from Google AI Studio, or the key for an OpenAI-compatible service. On this public Space the endpoint must be a public https:// address. No per-sample expression values are sent by either feature. AI Summary sends, per comparison, the top significant proteins and the largest increases/decreases by fold-change (accession, gene name, log2FC, adjusted p-value, precursor count and observed proportion when the quantification pipeline provides them, and a contaminant flag), plus cross-comparison and CV-stability summaries. Data Chat sends the QC table (run names, groups, precursor/protein counts, MS1 signal), the top 100–800 DE proteins for the selected comparison with the same per-protein statistics plus AveExpr, t, P.Value and B, your selected proteins, your messages and the last few turns of the conversation. The model is told to name proteins only from the supplied gene names — it is not asked to recall biology from memory.

• —AI Summary - Analyzes all contrasts at once: top DE proteins, cross-comparison biomarkers, and CV stability metrics. Export as standalone HTML report
• —Interactive Data Chat - Conversational AI with full dataset context (QC stats, top DE proteins, phospho sites when available). Auto-Analyze button for one-click reports
• —Interactive AI + Plot Connection - Select proteins in volcano/table to set AI context; AI responses highlight proteins in your plots automatically
• —Export for Claude - Download your complete analysis as a .zip for deep analysis with Claude, ChatGPT, or other AI assistants (includes DE results, expression matrix, QC metrics, GSEA, methods text, and more)
• —Save Chat History - Download conversations as plain text

💾 Session Management

• —Save/Load Sessions - Preserve analysis state (.rds files)
• —Reproducibility Logging - Export complete R code
• —Example Data - One-click demo dataset (Affinisep vs Evosep)

🎓 Education & Resources

• —Embedded proteomics training materials
• —UC Davis Proteomics video tutorials
• —Methodology citations (limpa, limma, DIA-NN)

📖 Quick Start

1. 1.Load Data: Upload DIA-NN .parquet file or use "Load Example Data"
2. 2.Assign Groups: Use auto-guess or manual assignment
3. 3.Run Pipeline: Click "▶ Run Pipeline" -- the app normalizes your data, quantifies proteins, and runs statistical tests to identify which proteins differ significantly between your groups
4. 4.Explore Results: Interactive plots, tables, GSEA, AI chat

🔬 Methodology

• —Normalization: Data Point Correspondence (DPC-CN)
• —Quantification: DPC-Quant (Detection Probability Curve Quantification)
• —Statistics: limma empirical Bayes moderation
• —FDR Correction: Benjamini-Hochberg

📚 Resources

• —GitHub: github.com/bsphinney/DE-LIMP
• —Discussions: github.com/bsphinney/DE-LIMP/discussions
• —Website: bsphinney.github.io/DE-LIMP
• —YouTube: UC Davis Proteomics
• —Core Facility: proteomics.ucdavis.edu

⚠️ Hosted Version Limitations

This Hugging Face deployment has some limitations compared to a local installation:

• —No DIA-NN search or XIC viewer -- these require local file access and are only available on local/HPC installations
• —Upload size limits -- very large files may fail to upload or cause timeouts
• —Sessions don't persist -- your analysis is lost when you close the browser tab; use "Save Session" to download an .rds file you can reload later
• —Shared resources -- large datasets or complex analyses may be slower than on a dedicated machine

For the full experience, install DE-LIMP locally.

🛠 System Requirements (Local Installation Only)

The following requirements apply only if you are installing DE-LIMP on your own machine. The hosted version above requires nothing but a web browser.

• —R 4.5+
• —Bioconductor 3.22+
• —All dependencies auto-install on first run

👨‍🔬 Developer

Brett Phinney - UC Davis Proteomics Core Facility

Built with ❤️ for the proteomics community