In progress
01 · Import & QC
Identification depth, Q-value checks, missingness, CVs, retention-time and ion-mobility stability, injection-order drift on QC pools, correlation and PCA.
Analysis portfolio
Proteomics data analysis in Python
A reproducible, end-to-end analysis of public DIA proteomics data — built to show how I evaluate data quality, choose normalization, and set statistical thresholds before trusting a result. Every step is scored against datasets where the right answer is known.
Dataset A
Benchmarking a DIA pipeline against known truth
Hybrid-proteome LFQ benchmark (Van Puyvelde et al., Sci Data 2022; PRIDE PXD028735), timsTOF Pro diaPASEF, processed with DIA-NN. Human, yeast, and E. coli digests are mixed at defined ratios, so the true log2 fold changes are known (human 0, yeast +1, E. coli −2) — every QC, normalization, and statistics choice can be graded, not just asserted.
Planned
02 · Normalization & roll-up
None vs. median vs. quantile vs. DIA-NN's own normalization; MaxLFQ vs. custom protein roll-up — each scored against the expected human log2FC of 0.
Planned
03 · Differential abundance calibration
Welch's t vs. moderated t; technical vs. biological replicate structure; empirical FDR from species truth; ROC curves and fold-change compression.
Dataset B
Applying the calibrated pipeline to a degrader screen
The thresholds validated on Dataset A will be carried into a public molecular-glue degrader proteomics dataset on the same platform (timsTOF diaPASEF + DIA-NN). Coming soon.