Pro-ker

Data Already Loaded

An analysis is already in progress. What would you like to do?

Export Canvas

Format

Options

Export graphs separately (stacked vertically)

Uncheck to export the canvas exactly as displayed, preserving graph positions, text boxes, lines, and shapes.

Width (px)

Height (px)

SVG — fully vectorized, editable in Illustrator/Inkscape/Figma. All text remains as text elements.
PNG — high-resolution raster at 2× scale.

Session

Save

Session is also auto-saved to browser storage.

Restore

Pro-ker Proteomics Analysis

Version: 4.14.0

Publisher: Billy M Ngo

Published: February 2026

Updated: May 2026

Ngo, B.M. (2026). Pro-ker Proteomics Analysis [Software].

References & Acknowledgments

Statistical methods

Benjamini & Hochberg (1995) — J. R. Stat. Soc. B 57:289 — FDR correction

Tusher, Tibshirani & Chu (2001) — PNAS 98:5116 — SAM / S0 fudge factor

Smyth (2004) — Stat. Appl. Genet. Mol. Biol. 3:3 — limma moderated t-statistic (empirical Bayes)

Zhu et al. (2020) — Mol. Cell. Proteomics 19:1047 — DEqMS peptide-count covariate

Tyanova et al. (2016) — Nature Protocols 11:2301 — Perseus platform & default imputation

Lazar et al. (2016) — J. Proteome Res. 15:1116 — MinProb imputation

Wei et al. (2018) — Sci. Rep. 8:663 — QRILC imputation

Troyanskaya et al. (2001) — Bioinformatics 17:520 — kNN imputation

Kim et al. (2005) — Bioinformatics 21:187 — LLS imputation

Stekhoven & Bühlmann (2012) — Bioinformatics 28:112 — MissForest imputation

Wang et al. (2020) — Nucleic Acids Res. 48:e83 — NAguideR benchmark of 23 methods

Data & tools

Cox & Mann (2008) — Nature Biotechnology 26:1367 — MaxQuant proteomics processing

D3.js — Scale math (axis tick computation only)

SheetJS — XLSX export

Statistical Methods

Volcano Plot — Differential Expression

Statistical test — Three options, selectable in the volcano configuration:
• Welch’s t-test (default): two-sample unequal-variance t-test comparing group means, with Welch–Satterthwaite degrees of freedom. P-values from the regularized incomplete beta function. Standard textbook test; appropriate baseline.
• Moderated t-test (limma): empirical-Bayes shrinkage of per-protein variance toward a fitted scaled-inverse-χ² prior (Smyth 2004). Posterior variance s²_post = (d₀·s²₀ + d_g·s²_g) / (d₀ + d_g); posterior df = d_g + d₀. Hyperparameters d₀ (prior df) and s²₀ (prior variance) are estimated from the empirical distribution of per-protein log-variances via method of moments using the digamma/trigamma identities. Recommended for studies with few replicates (≤ 5/group). Ref: Smyth (2004) Stat. Appl. Genet. Mol. Biol. 3:3.
• Peptide-count moderated (DEqMS): extends limma by replacing the single global prior with a per-protein prior that depends on peptide count. Pro-ker’s implementation bins proteins by log₂(peptide count) and uses each bin’s median log-variance as the per-bin prior. Estimates d₀ from the residual log-variance after the binned smoother (so d₀ reflects within-bin variance, not total variance). Falls back to limma if the file has no Peptides column. Ref: Zhu et al. (2020) MCP 19:1047.

Benjamini–Hochberg FDR — Multiple-testing correction controlling the false discovery rate. Adjusted p-values (q-values) are computed by ranking raw p-values and scaling by n/rank. Applied to whichever test was selected above.

S0 correction — A fudge factor added to the t-test denominator (SE + S0) to penalize fold changes driven by low-abundance noise (Tusher et al., 2001). At S0 = 0 this is the standard test. Typical values range from 0.1–2.0. Layered on top of any of the three tests above.

Missing value imputation — Seven methods available, covering the main families established in the proteomics literature (NAguideR, Wang et al. 2020):
• Perseus-style Gaussian (default): draws from a down-shifted normal distribution in log2 space (shift = 1.8 SD, width = 0.3 SD), using the comparison-wide non-zero values to estimate μ/σ. MNAR-aware. Ref: Tyanova et al. (2016) Nature Protocols 11:2301.
• MinProb (canonical): per-sample. Computes MinDet = the smallest observed log2 value in each sample, then for each missing cell draws from N(MinDet_sample, σ_sample) where σ is 1% of the sample's range. MNAR-aware, sample-specific. Ref: Lazar et al. (2016) J. Proteome Res. 15:1116; imputeLCMD::impute.MinProb.
• QRILC: draws from a Gaussian centered at the 1st percentile of the comparison's global log2 distribution. MNAR-aware, single global tail. Ref: Wei et al. (2018) Sci. Rep. 8:663.
• kNN: inverse-distance-weighted mean of the k = 10 nearest proteins (Euclidean distance on shared observed positions). MCAR-aware. Ref: Troyanskaya et al. (2001) Bioinformatics 17:520.
• LLS (Local Least Squares): fits a linear regression of the target protein on its k = 10 nearest similar proteins (in log2 space, with a small ridge for stability), then predicts the missing values. MCAR-aware, often outperforms kNN at high dimensionality. Ref: Kim et al. (2005) Bioinformatics 21:187.
• Random Forest (MissForest): iterative non-linear regression. For each sample column with missing values, trains a random forest on the other samples to predict it (50 trees, depth 6, m_try = √p); iterates with relative-change convergence (tol = 0.01, max 5 iterations). Top-ranked in multiple proteomics benchmarks. Ref: Stekhoven & Bühlmann (2012) Bioinformatics 28:112.
• Min/2: replaces missing values with half the per-protein minimum non-zero value. Simple deterministic baseline; produces "bifurcate" volcano patterns at high missingness, included only as a sanity check.
Reproducibility: All four stochastic methods (Perseus, MinProb, QRILC, MissForest) consume a seed for reproducibility. Per-protein methods (Perseus, MinProb, QRILC) derive a per-protein PRNG seed from the user-supplied Random seed (default 42) and the protein's filtered index: (seed·1000003 + 2·fi + 1or2) mod 2³². Matrix-wide MissForest uses the global seed once for bootstrap and feature subsampling across the whole forest. Same seed + same data → identical imputed values and an identical volcano plot. Change the seed to verify a conclusion is robust to the specific draws — a result that depends on a single seed is not robust. The seed is recorded on the Volcano Settings sheet of the XLSX export. kNN, LLS, and Min/2 are deterministic and ignore the seed.

Log2 Fold Change — Calculated as log2(mean_X / mean_Y). Positive values indicate higher abundance in Group X; negative values indicate higher abundance in Group Y.

PCA — Principal Component Analysis

Algorithm — Dual-space PCA (kernel method) optimized for datasets where the number of samples is much smaller than the number of features. Computes the n×n kernel matrix K = X·X^T from the mean-centered data matrix, then performs Jacobi eigendecomposition.

Scores — PC scores are eigenvectors scaled by the square root of their eigenvalues. Variance explained is reported as a percentage of total variance for each component.

Version History

4.14.0 (May 2026)
Three statistical tests for the volcano plot: Welch’s t-test (default, unchanged), moderated t-test (limma; empirical-Bayes shrinkage of per-protein variance), and peptide-count moderated (DEqMS; per-bin prior tied to peptide count). New "Statistical test" dropdown in the volcano configuration. Pure-JS digamma/trigamma special functions verified against R’s psigamma to 1e-6.

4.13.0 (May 2026)
Three new imputation methods: Random Forest (MissForest), LLS (Local Least Squares), canonical MinProb (per-sample MinDet). Old "MinProb" honestly renamed to QRILC. Seven imputation methods total spanning every NAguideR family. Seed traceability across all stochastic methods.

4.12.x (May 2026)
JS error containment (per-plot render isolation, step-isolated renderAll, visible error toast, async rejection handler, emergency reset). Desktop icon redesigned with transparent background, multi-resolution PNG-in-ICO. README rewritten to publication quality, LICENSE created.

4.11.0 (May 2026)
Parser robustness: NaN/Inf collapse to 0, OverflowError caught for Inf in int fields, UTF-16 LE/BE and UTF-8 BOM auto-detection. Zero-protein uploads blocked with clear error. Format-info banner for parser adaptations. Disabled-with-explanation for column-dependent filters. Plot palette + group/analysis tooltips.

— prior versions in CHANGELOG.md —

4.2.0–4.10.1 (April 2026, consolidated)
Multi-file workflow with cross-file guards, self-contained sessions (embedded RAW), XLSX export with cell-clipping, per-plot Prism-ready CSV export, comprehensive PCA customization, deterministic per-protein imputation seed, volcano presence-only side bands, single-number GOI input, embedded user-designed favicon, parser support for orthofinder-merged files. See CHANGELOG.md for full details.

4.1.0 (April 2026)
+Shape annotations (rectangle, circle, diamond, triangle with color/fill/border), two-row toolbar with editing controls below, separate volcano FC/FDR threshold line toggles, canvas panning by click-drag on empty space

4.0.0 (April 2026)
Published imputation methods (Perseus, MinProb, kNN), robust session save/load with graph positions and annotations, searchable protein dropdown with FASTA drag-drop for coverage map, amino acid-level peptide display with boundary markers, context menu color swatch, UI tooltips, upload performance fix

3.9.0 (April 2026)
Protein Abundance ranked plot with error bars (SD, SEM, range), volcano tooltip guidance for FC threshold and S0

3.8.0 (April 2026)
Unified color swatch palette across all pickers, swatch popup toolbar fix, dynamic macOS plist version

3.7.0 (April 2026)
Fix canvas text/line annotation glitches, unified default marker size, auto-shutdown lock path fix, merge all pending features

3.6.0 (April 2026)
Groups of Interest, species highlight, 2-of-diamonds logo, hover shows protein ID, min click target, toolbar enhancements, color swatch on hover, remove Bokeh

3.5.0 (April 2026)
Peptide coverage map (FASTA sequence input, residue-level visualization), desktop icon fix

3.4.0 (April 2026)
Volcano threshold lines & group side labels, S0 low-abundance correction, volcano/PCA data in CSV export, statistical methods documentation, README

3.3.0 (April 2026)
BiNgo color swatches, undo/redo, hollow markers, resizable panels, volcano fixes

3.2.0 (March 2026)
Intra-group dot plots, toolbar annotations, PCA fix, dot plot axis breaks

3.1.0 (March 2026)
Number inputs for graph settings, centered plots, scroll zoom

3.0.0 (March 2026)
Custom ProkerChart SVG engine (replaces Plotly), context toolbar, canvas zoom

2.9.0 (March 2026)
Beeswarm unique plot, radio aggregation, freeze graphs

2.8.0 (March 2026)
Dynamic version, rebuilt tutorial, transparent backgrounds

2.7.0 (March 2026)
Fix color selection, clean labels, draggable plots

2.6.0 (February 2026)
Comprehensive CSV export

2.5.0 (February 2026)
Export Canvas (SVG/PNG), text boxes, lines, grid toggle

2.4.0 (February 2026)
Arial font, improved visual quality, ProkerChart SVG engine

2.3.0 (February 2026)
Fix axis/title editing

2.2.0 (February 2026)
Editable titles/axes, right-click selection

2.1.0 (February 2026)
Graph settings, click-to-label, intensity toggle, axis editing

2.0.0 (February 2026)
BiNgo-style header, info panel, themes, session save, auto-update, desktop shortcut