A reanalysis of public single-cell RNA-seq from fifteen carotid endarterectomy plaques, asking which macrophage state carries the cholesterol efflux program I measured by qPCR for my thesis. The answer is the lipid-loaded state. The site walks through the pipeline one step at a time: quality control on data that turned out to have been filtered before deposition, correction for the fact that each patient ran on their own plate, lineage assignment, and gene program scoring.

The last step is live rather than a figure. It recomputes the comparison with the unit of replication set to either the patient or the cell, which moves the same result from p = 0.001 to p = 1.4e-62 on the same 2,454 cells. A closing section sets out two errors in the analysis, both of which flattered the result.

A compact transcript panel separated four recorded laboratory stimulation conditions in human macrophages (GEO GSE46903), on donors held out of training and tuning. A 10–8–4 PyTorch network and a regularized linear model both scored 1.00 balanced accuracy and chose the same ten genes, which places the signal in the data rather than the architecture. Because a perfect score cannot show whether a model knows when it is wrong, calibration was measured where errors exist: on a five-gene panel the margin between the top two classes falls from 0.42 when it is right to 0.09 when it is wrong, and those six errors fall in three of seven donors. The site serves the stored predictions, the per-fold training loss and the recomputed metrics, and states the limits: seven donors cannot establish panel stability, training loss near 0.005 on twenty-one profiles is close to memorisation, and the deposited matrix was quantile-normalised upstream across all arrays.

A single workbench for a clinical systematic review: literature harvesting and dedup, dual-blind screening with an auto-generated PRISMA diagram, and a statistics engine that checks its own assumptions before running. Produces publication figures, risk-of-bias assessment, and a reproducibility bundle with the exact re-run code.

CURRENTLY UNDER DEVELOPMENT; But you can load a sample data file (SPSS/csv/xlsx and similar) and run tests and get figures, or run a meta analysis.

A recurring query against the ClinicalTrials.gov registry, scoped to stem-cell and cell-based interventions in neurodegenerative and neurotraumatic conditions. Phase and recruitment-status breakdowns, registration timeline, CSV export. Snapshots are cached per query to stay within the registry’s rate limits. Built as a monitoring companion for systematic review.