# Comparison with Niaré et al. (2026)

**Status: published comparison partially extracted; exact Extended Data table cells still need direct verification.** This is a comparison of published results with Cathedral's unadjusted reanalysis, not a like-for-like model replication.

## What the paper reports

The paper compared PIN and LMD parasites meeting its read-depth (at least 25x) and major-allele (at least 90%) criteria. It reports 465 samples overall and two-sided Wilcoxon rank-sum tests. In the K13-wild-type stratum, the paper reports:

| Assay | PIN median (n) | LMD median (n) | Published result |
| --- | ---: | ---: | --- |
| DHA IC50 | 3.7 nM (110) | 1.8 nM (240) | PIN higher; paper reports p < 0.001 overall |
| Lumefantrine IC50 | 14.3 nM (110) | 6.2 nM (240) | PIN higher; paper reports p < 0.001 overall |
| Mefloquine IC50 | 17.1 nM (109) | 11.0 nM (236) | PIN higher; paper reports p < 0.001 overall |
| Ex-vivo RSA | -- | -- | No significant difference reported; n = 169 overall |

The paper also reports adjusted mixed-effects findings. Its model controls for site, parasitemia, complexity of infection (COI), collection year, and K13 genotype. For DHA, the adjusted estimated marginal means are 3.1 nM (PIN) and 2.3 nM (LMD), p = 3.0e-8, Cohen's d = 0.44; PIN accounts for 12.3% of phenotypic variance in K13-wild-type samples. For lumefantrine, adjusted marginal means are 14.2 nM (PIN) and 8.8 nM (LMD), p = 0.0025, and the PIN term explains 4.9% of variance. Mefloquine's PIN term explains 6.5% of variance. RSA is not associated in the adjusted model (p = 0.13). The published narrative does not give the adjusted confidence interval bounds or every coefficient; consult Extended Data Table 1 rather than inferring them here.

For the PX1-by-K13 interaction model, the paper reports no significant interaction for lumefantrine, mefloquine, or RSA (p > 0.65); the DHA interaction is near, but does not reach, conventional significance (p = 0.061). The exact table cells are in Extended Data Table 2 and should be checked before using any more precise interaction estimates.

## What Cathedral's reanalysis reports

Cathedral applies the available source-workbook labels and fields (PIN/LMD, WSAF > 0.9, eligible K13 categories), then performs an unadjusted two-sided Mann-Whitney/Wilcoxon rank-sum comparison. All eligible complete-case counts and estimates are in `phenotype-comparisons.csv`; the headline rows are:

| Assay | PIN median (n) | LMD median (n) | PIN minus LMD | Bootstrap 95% interval | BH q |
| --- | ---: | ---: | ---: | ---: | ---: |
| DHA IC50 | 3.71 nM (156) | 1.80 nM (253) | 1.91 nM | 1.590 to 2.4645 | 4.276412e-19 |
| Lumefantrine IC50 | 15.14 nM (157) | 6.25 nM (258) | 8.89 nM | 7.40 to 11.61 | 4.214171e-20 |
| Mefloquine IC50 | 16.95 nM (154) | 11.125 nM (254) | 5.825 nM | 3.50 to 8.97 | 1.308759e-10 |
| RSA survival | 4.07% (115) | 3.0965986395% (64) | 0.9734013605 percentage points | -1.079619456 to 2.717987165 | 0.2627997608 |

These are unadjusted differences between medians. The confidence intervals are deterministic bootstrap percentile intervals (4,000 resamples); p-values use a tie-corrected normal approximation with continuity correction. Benjamini-Hochberg q-values are adjusted within the four-assay family.

## Consistency and why the numbers differ

The directions agree for DHA, lumefantrine, and mefloquine: the PIN group has higher IC50 medians. Both analyses report no clear RSA difference. The magnitudes and sample counts must not be compared as if they came from the same cohort or estimator. In particular, the current workbook filter retains 772 records before assay-specific missingness, but has no numeric per-sample read depth; it cannot apply the article's >=25x requirement or reproduce its depth-qualified 465-sample cohort. The source workbook also does not provide all sample-linked COI and model inputs. Cathedral therefore has not recreated the original adjusted model or interaction model.

The K13-wild-type DHA count illustrates the difference: the paper reports PIN n=110 and LMD n=240 (medians 3.7 and 1.8 nM); Cathedral reports PIN n=115 and LMD n=236 (medians 3.70 and 1.77 nM) from its available workbook filter. The count differences are not explained away by similar medians.

## What remains to complete a like-for-like reproduction

1. The exact sample-matched depth/allele-support data and documented 465-sample inclusion list.
2. The sample-matched COI, parasitemia, site, year, K13, and phenotype table used in the models, with data-use conditions and stable join keys.
3. The authors' model formula, coding/contrasts, diagnostics, and execution environment from the archived analysis code.
4. Direct extraction and independent verification of the exact confidence intervals and estimates in Extended Data Tables 1 and 2.

## Sources

- Niaré K et al. *Emergence and spread of Plasmodium falciparum PX1 polymorphisms associated with decreased susceptibility to antimalarials in Uganda*. Nature Medicine (2026). [Article and Extended Data Tables 1-2](https://www.nature.com/articles/s41591-026-04590-5), [DOI 10.1038/s41591-026-04590-5](https://doi.org/10.1038/s41591-026-04590-5).
- Article source data and code are linked from the article's Data availability and Code availability sections. The code repository is archived at [Zenodo 10.5281/zenodo.21283869](https://doi.org/10.5281/zenodo.21283869).
- Cathedral source checksums, dates, filters, and reuse notes: `data_manifest.json`.