Update README.md

README.md

@@ -18,17 +18,17 @@ pretty_name: scMPRA stratified negative binomial fits
 Fitted parameters for every stratified negative binomial (NB) and
 zero-inflated negative binomial (ZINB) model reported in *Modeling,
 calibration, and power analysis of single-cell massively parallel reporter
-assays*. Fifteen fits over three published scMPRA datasets: three canonical
+assays*. There are fifteen fits over three published scMPRA datasets: three canonical
 (one per dataset, the model the paper's analyses use) and twelve
 counterfactuals kept so the model-selection comparisons can be reproduced.
 
-Two layers. The parameter tables are plain parquet and need nothing but
+The parameter tables are plain parquet and need nothing but
 pandas. The orthos are the complete fitted objects, loadable with
-`scMPRAforge`, each carrying the count table it was fit on.
+`scMPRAforge`. Bounds objects derived from these fits are included with `scMPRAforge`.
 
 ## Files
 
-| file | rows | what it is |
+| file | rows | description |
 |---|---|---|
 | `fitted_means.parquet` | 45,296 | fitted mean and dispersion, one row per level |
 | `zero_inflation.parquet` | 6,016 | fitted zero inflation, ZINB fits only |
@@ -36,17 +36,6 @@ pandas. The orthos are the complete fitted objects, loadable with
 | `export_fits.py` | | the script that produced all three |
 | `orthos/<fit>/` | 15 | the complete fitted objects, one directory per fit |
 
-### Why two tables
-
-A fit is stratified along one axis and estimates a mean for every level of the
-other: a by-cell-type fit gives one `mu` per CRE within each cell type, and a
-by-CRE fit one `mu` per cell type within each CRE. Both directions are
-released; they converge on the same estimates, which is itself a result in the
-paper.
-
-Zero inflation is **not** on that axis. It is fit per sequencing replicate, so
-`zero_inflation.parquet` is indexed by `rep_id` and cannot be joined onto a
-mean row. Dispersion (`theta`) is fit once per stratum and so repeats down it.
 
 ### Schema
 
@@ -70,17 +59,7 @@ and `level_axis` always `rep_id`.
 | `cohen` | Zhao et al. 2023, *Nat Genet* 55:346, [10.1038/s41588-022-01278-7](https://doi.org/10.1038/s41588-022-01278-7) | scMPRA, promoter variants in live mouse retinas |
 | `seelig` | Yin et al. 2025, *Cell Systems* 16:101302, [10.1016/j.cels.2025.101302](https://doi.org/10.1016/j.cels.2025.101302) | scMPRA of designed enhancers in HepG2/K562 |
 
-The lab code names are kept as identifiers so these join to the analysis code;
-cite the papers above.
-
-## Which fit to use
-
-`fits.csv` carries `canonical`. The three canonical fits are the ones the paper
-uses; the rest vary one modelling decision at a time (NB against ZINB, how
-unobserved combinations are treated, whether a coarse transfection reporter
-expands to one zero or many) and exist so those choices can be re-examined.
-A counterfactual is not a worse fit of the same thing — it is a fit of a
-different assumption.
+The lab names are kept as identifiers so these join to the analysis code. Make sure you cite the papers above in downstream analysis.
 
 ## Loading
 
@@ -94,12 +73,10 @@ canonical = set(fits.loc[fits.canonical, "fit"])
 mu = means[means.fit.isin(canonical) & (means.stratified_by == "cell_type")]
 ```
 
-No pickles, and nothing here needs `scMPRAforge` installed to read.
-
 ### Orthos
 
 Each ortho holds its training data in `training_data.scmpra/`: the count table
-as the fit saw it, after filtering and negative-control pooling, as parquet.
+post after filtering and negative-control pooling.
 `ortho.load` reads it from there, so an ortho loads and computes from any
 working directory.
 
@@ -112,32 +89,17 @@ o = ortho.load(client, "orthos", "seelig_cm_moib_nb_phantom")
 ```
 
 This needs `scMPRAforge` 0.1.0.dev1 or later
-(`pip install "scMPRAforge>=0.1.0.dev1"`). An older version loads the models
-and parameters but warns and leaves `training_data` as `None`.
-
-Orthos are pickles. Unpickling runs code, so load them only from a copy of
-this repository you trust.
+(`pip install "scMPRAforge>=0.1.0.dev1"`).
 
-## How they were fit
-
-Maximum likelihood via a TensorFlow solver, each on four CPU cores of an Intel
-Xeon 8562Y+ (Emerald Rapids) node with no GPU. Per-fit wall time and peak
-memory are reported in the paper. `fits.csv` records each fit's zero handling
-and the script that produced it.
 
 ## Licence and source data
 
-Two licences apply, by file:
 
-| files | licence | why |
+| files | licence | reason |
 |---|---|---|
-| `orthos/seelig_*/` | [CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/) | carry Yin et al.'s count table, published under CC-BY-NC-4.0 |
+| `orthos/seelig_*/` | [CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/) | contains Yin et al.'s count table, published under CC-BY-NC-4.0 |
 | everything else | [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) | our fits, and the `shendure` and `cohen` count tables, whose sources are CC-BY-4.0 |
 
-The parameter tables include the `seelig` fits; those are our estimates, not
-Yin et al.'s measurements, and are CC-BY-4.0 with the rest. Commercial use of
-the `seelig` orthos needs Yin et al.'s terms, not ours.
-
 The count tables are processed from the depositing studies' public data:
 
 | dataset | deposit | source terms |
@@ -146,7 +108,6 @@ The count tables are processed from the depositing studies' public data:
 | `cohen` | GEO [GSE212842](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE212842), Zenodo [14907846](https://doi.org/10.5281/zenodo.14907846) | Zenodo deposit is CC-BY-4.0 |
 | `seelig` | GEO [GSE269037](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE269037) (scMPRA), [GSE269036](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE269036) (bulk) | paper is CC-BY-NC-4.0 |
 
-Cite the three papers alongside this release.
 
 ## Citation