pySaySuite

python asSay Suite for quick, robust and reproducible assay analyses in python.

Installation (general)

The package can be obtained from PyPI:

pip install pySaySuite

qPCR module

Standardized, reproducible qPCR differential-expression analysis for the lab. One canonical, tested implementation of the ANCOVA / multivariable-linear-model method of

Hampton et al., Analyzing qPCR data: better practices to facilitate rigor and reproducibility, Biochem. Biophys. Rep. 44 (2025) 102356.

The statistical method is theirs; this package ports it to Python for Applied Biosystems ViiA 7 / QuantStudio exports.

Details

The module is qPCR-focused and reproduces the published better practices paper 1:1 (Tables 3 & 4 ANCOVA to 3 dp, geNorm conclusion, GAPDH–MMP10 correlation, and the amplification efficiencies 1.53/1.57 to ~1e-6 against the qpcR R oracle). The whole analysis is a short, curated set of functions in two modules:

• pysaysuite.qpcr.core — the headline pipeline:
  • read_plates({name: path}, kind="results"|"amplification", batch_map=None) — ingest QuantStudio/ViiA 7 exports into a tidy long table, retaining plate identity (D9).
  • assign_group(long, group) — configurable sample→group assignment (prefix-dict / callable / layout table), flagging unmatched samples instead of silently grouping them.
  • build_wide(long, ...) — collapse technical replicates to mean CT; flag noisy replicate groups (wide.attrs["noisy_replicates"], D11).
  • run_ancova(wide, targets, refs, ...) — per-target ANCOVA on raw CT, returning a tidy coefficient DataFrame (one row per model term). Supports a formula= override and reference="gmean" (default) / a gene name / "separate" (D3); BH-FDR across the panel (D5) and an observations-per-parameter guard (D8).
  • check_identifiability(...) — Group⊥plate separability (D6); run_ancova also warns on confounded designs automatically.
• pysaysuite.qpcr.qc — quality screens and efficiency-based cross-checks:
  • genorm2 — geNorm M reference selection, a 1:1 port of ctrlGene::geNorm2 (D14).
  • reference_stability, target_reference_correlation — reference QC.
  • estimate_eff(amp, design=..., method="spline") — per-gene efficiency; the optional design= metadata join restricts the screen to real sample wells (drops NTC/water), which is what makes the published 1.53/1.57 reproduce exactly. Two estimators are kept and never conflated (D15): well_efficiency_spline (qpcR-faithful) and well_efficiency (a robust, baseline-subtracted LinRegPCR screen).
  • run_ancova_efficiency — ANCOVA on efficiency-weighted CT (robustness variant).
  • pfaffl_compare — Pfaffl / 2^-ΔΔCt efficiency-corrected cross-check (D7).
• pysaysuite.qpcr.plot / pysaysuite.qpcr.report — figures (forest, efficiency, reference stability, diagnostics, covariate, Pfaffl) and the methods-stamp / output writers (D10).
• pysaysuite.Config — typed analysis config, loadable from TOML.
• pysaysuite.core — intentionally empty for now; a shared cross-assay layer is premature with only the qPCR module present (each assay's input differs), and is deferred until a second assay lands (D12).

The headline functions are also re-exported at the top level: from pysaysuite import read_plates, build_wide, run_ancova, genorm2, estimate_eff, ....

Quickstart

import pysaysuite as ps

# 1. Ingest QuantStudio exports (one entry per plate) and assign groups.
long = ps.read_plates({"Plate1": "plate1.xls", "Plate2": "plate2.xls"})
long = ps.assign_group(long, {"N": "Control", "T": "Case"})   # by sample-name prefix

# 2. Reference-gene QC and selection.
wide = ps.build_wide(long)
stability = ps.genorm2(wide, ["RNU48", "U6", "GAPDH"])         # geNorm M ranking

# 3. Headline differential expression (ANCOVA on raw CT, geometric-mean reference).
result = ps.run_ancova(wide, targets=["MIR-21", "MIR-331"], refs=["RNU48", "U6"])
groups = result[result["is_group_effect"]]                    # one row per target
print(groups[["target", "estimate", "p_value", "p_fdr", "direction"]])

# 4. (Optional) efficiency screen + Pfaffl cross-check.
amp  = ps.read_plates({"Plate1": "plate1.xls"}, kind="amplification")
eff  = ps.estimate_eff(amp, design=long[["well"]].drop_duplicates())
xchk = ps.pfaffl_compare(wide, ["MIR-21"], ["RNU48", "U6"], eff, result)

run_ancova returns a tidy table with columns target, term, estimate, se, t, p_value, is_group_effect, p_fdr, direction, n, n_params, obs_per_param, r2_adj; the analysis parameters for the methods stamp live in result.attrs["params"].

Reproduce the paper, 1:1

docs/qpcr_examples/qpcr_paper_replication_pysaysuite.ipynb runs the package end-to-end on Hampton et al.'s own data (docs/qpcr_examples/data/) and asserts, in-notebook, that every result matches the published values: Tables 3 & 4 (ANCOVA), the geNorm conclusion, the GAPDH–MMP10 correlation, and the GAPDH/MMP10 amplification efficiencies. Execute it with:

conda run -n pysaysuite jupyter nbconvert --to notebook --execute --inplace \
  docs/testfiles/qpcr_paper_replication_pysaysuite.ipynb

Install (for development)

The package is developed in its own conda environment:

conda create -n pysaysuite python=3.12 -y
conda activate pysaysuite
pip install -e ".[dev]"