# Atacama Soil Microbiome

These chapters demonstrate the QIIME2 plugins for high-dimensional statistics on the
Atacama soil microbiome dataset {cite}`neilson2017significant`. q2-gglasso solves a
range of graphical lasso problems to identify microbial associations, and q2-classo
assesses those associations by fitting sparse log-contrast models. The sequence data
were processed under QIIME 2 version 2026.7 {cite}`bolyen2019reproducible`: the
q2-demux plugin demultiplexed and quality-filtered the raw reads, DADA2
{cite}`callahan2016dada2` denoised them through the q2-dada2 plugin, and the
q2-feature-classifier {cite}`bokulich2018q2` assigned taxonomy to the amplicon
sequence variants (ASVs) with a naive Bayes classifier trained on the Silva database
{cite}`quast2012silva`.

## Data preprocessing pipeline

Preprocessing follows the published Atacama
[tutorial](https://amplicon-docs.qiime2.org/en/latest/tutorials/atacama-soils.html).
To give these ASVs meaningful names, this book renames the ASV keys as follows:

| ASV label | Real ASV ID                      |
| --------- | -------------------------------- |
| ASV-1     | 89cb1ddf89dcf11d86f725dbcaa9a5ce |
| ASV-2     | cdb9c0ee3bba4c3d8b9292eb575bd9e3 |
| ASV-3     | 4c8ff0ea98d2c0ebb486e33bd96c61f9 |
| ASV-4     | a56b903521b8ab33a7878b35582803a8 |
| ASV-5     | 5c78314ff92e6fec9aa07acc1fa0dc24 |
| ASV-6     | dc8a2f47b3d1dc2e1f5f805891976b29 |
| ASV-7     | 6b780e361cfc5f06def718518324bdcb |
| ASV-8     | f6c10a04d57159c0d64d6bc30c677471 |
| ASV-9     | ffd60d684f32e6fd5b47fe90095f9d34 |
| ASV-10    | ef3fdbe1dcde754d91130cde6a4b4d61 |
| ASV-11    | a36b38f754f6abd278aeb9dbc7696343 |
| ASV-12    | a7b877ae6d2f079a15b6b192a4425620 |
| ASV-13    | 409faa5f5353e543bf6d99125c7c0e83 |

## Data for downstream analysis

The Atacama soil microbiome dataset {cite}`neilson2017significant` supplies:

- 50 samples from Atacama Desert soil
- 13 microbial taxa (ASVs)
- Environmental covariates: pH, elevation, temperature, humidity and vegetation

The original data is available through the European Nucleotide Archive under accession [ERP019482](https://www.ebi.ac.uk/ena/browser/view/PRJEB17617).

| Sample ID | ASV-1 | ASV-2 | ASV-3 | ... | ASV-11 | ASV-12 | ASV-13 | Elevation | pH | Avg Soil RH | Avg Soil Temp | Vegetation |
|-----------|-------|-------|-------|-----|--------|--------|--------|-----------|----|-----------|--------------|-----------| 
| BAQ2420.1.1 | 0.0 | 11.0 | 0.0 | ... | 115.0 | 0.0 | 0.0 | 2420 | 9.33 | 82.54 | 22.45 | no |
| BAQ2420.1.2 | 0.0 | 0.0 | 0.0 | ... | 0.0 | 0.0 | 0.0 | 2420 | 9.36 | 82.54 | 22.45 | no |
| BAQ2420.1.3 | 0.0 | 0.0 | 0.0 | ... | 0.0 | 0.0 | 0.0 | 2420 | 8.90 | 82.54 | 22.45 | no |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| YUN3856.2 | 6.0 | 13.0 | 26.0 | ... | 0.0 | 0.0 | 104.0 | 3856 | 7.43 | 99.44 | 9.51 | yes |
| YUN3856.3 | 21.0 | 36.0 | 23.0 | ... | 33.0 | 0.0 | 0.0 | 3856 | 7.43 | 99.44 | 9.51 | yes |

The table above is a snapshot of the counts. Download the QIIME2
[artifact](https://github.com/Vlasovets/q2-gglasso/blob/main/data/atacama-counts.qza) and
the matching
[metadata](https://data.qiime2.org/2026.7/tutorials/atacama-soils/sample_metadata.tsv).

## Covariates and missing values

Four numeric covariates accompany these samples — pH, elevation, average soil
relative humidity and average soil temperature. They are the outcomes and the
adjustment variables in the [log-contrast regression](../03_lowdim_classo/03_regression/01_logcontrast.md)
chapters, and the latent components in
[Latent Components & Covariates](../04_highdim_atacama/04_latent_pca.md) are tested
against them.

```{figure} ../../images/png/ph.png
:name: fig-covariate-ph
:width: 100%

Soil pH across the 75 samples, before and after scaling. The bulk of the
distribution sits between 6 and 9 — alkaline, as expected for this desert. The
bar at zero holds the samples whose pH was never recorded.
```

```{important}
**Some covariate values are missing and coded as `0`, not as blanks.** Counted
directly from `atacama-selected-covariates-veg.tsv` (75 samples):

| covariate | zeros | plausible as a real value? |
|---|---|---|
| `ph` | 8 | No — soil pH of 0 is not physically possible |
| `average-soil-relative-humidity` | 3 | Implausible |
| `average-soil-temperature` | 3 | Possible at altitude, but suspicious |
| `elevation` | 0 | — (range 895–4700 m) |

The zeros do not co-occur: none of the humidity-zero or temperature-zero samples
is also a pH-zero sample, so this is per-measurement missingness rather than
eight incomplete records. The eight pH-zero samples are `BAQ1370.3`, `BAQ1552.2`,
`BAQ895.2`, `BAQ895.3`, `YUN1005.2`, `YUN3008.2`, `YUN3008.3` and `YUN3184.2`.

A zero is not neutral. Scaled, those eight samples land at $-2.80$ — the minimum
of the distribution — so a regression that takes them at face value is told that
eight sites are far more acidic than any other, when their pH was never
recorded. Decide whether to drop those samples, impute them, or exclude pH as a
covariate, and state which you did. The tutorials that follow pass the file
through unchanged, which demonstrates the commands rather than a treatment of
missing data.
```