Skip to content

Commit

Permalink
up
Browse files Browse the repository at this point in the history
  • Loading branch information
@TuomasBorman
TuomasBorman committed Nov 13, 2024
1 parent 02b6dc8 commit 413fee3
Showing 1 changed file with 66 additions and 81 deletions.
147 changes: 66 additions & 81 deletions vignettes/case_study.Rmd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,8 @@ Our main study questions are:

* How does treatment influence the gut microbiota of salmon?
* Do gut flora and fatty acids composition evolve over time?
* Is there a relationship between gut microbiota and the fatty acid composition in muscle tissue?
* Is there a relationship between gut microbiota and the fatty acid composition
in muscle tissue?

```{r}
#| label: start
Expand Down Expand Up @@ -127,7 +128,8 @@ head(salmon_sample_ids)
The data returned above is a list of all sample accession numbers that are
associated with all salmons. For example, metagenomic amplicon samples, such as
[SAMEA112750580](https://www.holofooddata.org/sample/SAMEA112750580)
or fatty acid samples, [SAMEA112950027](https://www.holofooddata.org/sample/SAMEA112950027).
or fatty acid samples,
[SAMEA112950027](https://www.holofooddata.org/sample/SAMEA112950027).

We can use these accession numbers to fetch the data associated with each
sample type and store them as `experiments` in a `MultiAssayExperiment` (`MAE`)
Expand All @@ -151,7 +153,7 @@ mae <- HoloFoodR::getResult(
# Save salmon MAE This must be named differently
path <- system.file(
"extdata", "salmon_mae_without_mgnify.RDS", package = "HoloFoodR")
"extdata", "salmon_mae_without_mgnify.RDS", package = "HoloFoodR")
saveRDS(object = mae, file = path)
```

Expand All @@ -161,16 +163,18 @@ saveRDS(object = mae, file = path)
# Read MAE to avoid retrieval
path <- system.file(
"extdata", "salmon_mae_without_mgnify.RDS", package = "HoloFoodR")
"extdata", "salmon_mae_without_mgnify.RDS", package = "HoloFoodR")
mae <- readRDS(file = path)
```

### Fetch metagenomic data from MGnify

HoloFood database does not include the data for metagenomic data. This data
can be retrieved from the [MGnify portal](https://www.ebi.ac.uk/metagenomics).
For this purpose, we will use [MGnifyR package](https://bioconductor.org/packages/release/bioc/html/MGnifyR.html),
which in a similar fashion to HoloFoodR, allows simple interaction with MGnify API.
For this purpose, we will use
[MGnifyR package](https://bioconductor.org/packages/release/bioc/html/MGnifyR.html),
which in a similar fashion to HoloFoodR, allows simple interaction with MGnify
API.

```{r}
#| label: get_metagenomic_samples
Expand Down Expand Up @@ -217,7 +221,7 @@ tse <- MGnifyR::getResult(
# Save salmon metagenomic TreeSE
path <- system.file(
"extdata", "salmon_metagenomic_tse.RDS", package = "HoloFoodR")
"extdata", "salmon_metagenomic_tse.RDS", package = "HoloFoodR")
saveRDS(object = mae, file = path)
```

Expand All @@ -228,7 +232,7 @@ saveRDS(object = mae, file = path)
# Read in salmon metagenomic TreeSE object
path <- system.file(
"extdata", "salmon_metagenomic_tse.RDS", package = "HoloFoodR")
"extdata", "salmon_metagenomic_tse.RDS", package = "HoloFoodR")
tse <- readRDS(file = path)
```

Expand Down Expand Up @@ -293,8 +297,8 @@ mae <- mae[, colData(mae)[["Trial code"]] == "SA", ]

Some values of fatty acids are under detection thresholds. We assume them to be
zeroes. Moreover, the data includes a feature that just states from where the
fatty acids were collected. We remove this feature to ensure that the assay contains
only numeric values.
fatty acids were collected. We remove this feature to ensure that the assay
contains only numeric values.

```{r}
#| label: preprocess2
Expand Down Expand Up @@ -323,15 +327,17 @@ information.
# Add time points
timepoints <- colData(mae[[2]])
timepoints <- timepoints[match(timepoints[["animal"]], rownames(colData(mae))), ]
timepoints <- timepoints[
match(timepoints[["animal"]], rownames(colData(mae))), ]
timepoints <- ifelse(timepoints[["trial.timepoint"]] == 0, "start", "end")
timepoints <- factor(timepoints, levels = c("start", "end"))
colData(mae)[["timepoint"]] <- timepoints
# Add treatment groups
colData(mae)[["study_group"]] <- ifelse(
colData(mae)[["Treatment concentration"]]>0, "treatment", "control")
colData(mae)[colData(mae)[["timepoint"]] == "start" , "study_group"] <- "control"
colData(mae)[["Treatment concentration"]]>0, "treatment", "control")
colData(mae)[colData(mae)[["timepoint"]] == "start" , "study_group"] <-
"control"
# Add animal metadata to separate experiments
mae[[1]] <- getWithColData(mae, 1)
Expand All @@ -344,7 +350,7 @@ Next, we can agglomerate features by prevalence to reduce the number of
low-abundant taxa and contaminants.

First, we visualize prevalence distribution of taxa with a histogram to decide
the prevalence threshold to use. We use 0.1% detection level to filter out
the prevalence threshold to use. We use 0.2% detection level to filter out
extremely low-abundant genera.

```{r}
Expand All @@ -362,13 +368,13 @@ prevalence <- getPrevalence(
assay.type = "relabundance",
na.rm = TRUE,
sort = TRUE,
detection = 0.1 / 100
detection = 0.2 / 100
)
# Exclude microbes with 0 prevalence
prevalence <- prevalence[prevalence != 0]
hist(prevalence, xlab = "Prevalence")
hist(prevalence, main = "", xlab = "Prevalence")
```

We can also look at the raw prevalence numbers.
Expand All @@ -381,7 +387,8 @@ sort(prevalence, decreasing = TRUE) |> head(10)
```

_Mycoplasma_ is present in all samples, which is not surprising as this genus
was found to be one of the most common in salmon intestine (see @zarkasiPyrosequencingbasedCharacterizationGastrointestinal2014).
was found to be one of the most common in salmon intestine
(see @zarkasiPyrosequencingbasedCharacterizationGastrointestinal2014).

We then agglomerate our data by prevalence and by taxonomic rank to obtain group
all genera which are below the specified thresholds to the "Other" group.
Expand All @@ -402,22 +409,6 @@ altExp(mae[[2]], "prev_genus") <- agglomerateByPrevalence(
)
```

Moreover, we agglomerate the data to include only 19 most abundant genera.

```{r}
#| label: select_core
tse <- agglomerateByRank(mae[[2]], rank = "Genus")
core <- getTop(tse, top = 19)
# Get vector that tells whether taxa belongs to core taxa or not
core_names <- rownames(tse)
core_names[ !core_names %in% core ] <- "other"
# Agglomerate based on core taxa
rowData(tse)[["core"]] <- core_names
altExp(mae[[2]], "core_genus") <- agglomerateByVariable(
tse, by = "rows", group = "core")
```

Due to the limited number of samples, we also filter the fatty acid data to
include only those fatty acids that show variation within the dataset. The
rationale is that if a fatty acid does not vary, it cannot exhibit differences
Expand Down Expand Up @@ -459,14 +450,14 @@ well-established biological relevance. These include:
#| label: filter_fatty3
relevant_fatty_acids <- c(
"Docosahexaenoic acid 22:6n-3 (DHA)",
"Eicosapentaenoic acid 20:5n-3 (EPA)",
"Alpha-Linolenic acid 18:3n-3",
"Arachidonic acid 20:4n-6 (ARA)",
"Linoleic acid 18:2n-6",
"Oleic acid 18:1n-9",
"Palmitic acid 16:0",
"Stearic acid 18:0"
"Docosahexaenoic acid 22:6n-3 (DHA)",
"Eicosapentaenoic acid 20:5n-3 (EPA)",
"Alpha-Linolenic acid 18:3n-3",
"Arachidonic acid 20:4n-6 (ARA)",
"Linoleic acid 18:2n-6",
"Oleic acid 18:1n-9",
"Palmitic acid 16:0",
"Stearic acid 18:0"
)
altExp(mae[[1]], "relevant") <- mae[[1]][
rownames(mae[[1]]) %in% relevant_fatty_acids, ]
Expand Down Expand Up @@ -579,21 +570,13 @@ composition with relative abundance barplot.
#| label: abundance_plot
#| fig-cap: Relative abundance of core microbial genera across samples.
p <- plotAbundance(
altExp(mae[[2]], "core_genus"),
assay.type = "relabundance",
col.var = c("study_group", "timepoint"),
facet.cols = TRUE, scales = "free_x"
)
p <- plotAbundance(
altExp(mae[[2]], "prev_genus"),
assay.type = "relabundance",
col.var = c("study_group", "timepoint"),
facet.cols = TRUE, scales = "free_x"
)
p
p
```

Salmon gut seems to be dominated by either genus _Mycoplasma_ or
Expand Down Expand Up @@ -627,7 +610,7 @@ res

Based on the results, we conclude that older salmon exhibit distinct microbial
diversity compared to younger ones. Additionally, there appears to be a
slight—though not statistically significant—effect of treatment on microbial
slight — though not statistically significant — effect of treatment on microbial
diversity.

```{r}
Expand Down Expand Up @@ -735,9 +718,9 @@ p <- p + geom_shadowtext(aes(label = df[["percentage"]]))
p
```

Factor 1 primarily captures variance within the metagenomics data, while Factor
2 mostly represents variance in fatty acids, including some shared variability.
Factors 3 and 4 capture shared variability between the metagenomic data and
Factor 1 captures only variance within the metagenomics data, while over 2/3 of
variance captured by Factor 2 represents variance in fatty acids.
Factor 3 captures shared variability between the metagenomic data and
fatty acids, reflecting interconnected patterns between the two datasets.

Before exploring the shared variability, we first examine which metagenomic
Expand All @@ -746,13 +729,13 @@ variability is captured by Factor 1.
```{r}
#| label: plot_factor1
#| fig-width: 10
#| fig-height: 8
#| fig-height: 6
#| fig-cap: Features with the highest loadings for Factor 1.
p1 <- plot_top_weights(model, view = 1, factors = 1, nfeatures = 25) +
labs(title = "Fatty acids")
labs(title = "Fatty acids")
p2 <- plot_top_weights(model, view = 2, factors = 1, nfeatures = 25) +
labs(title = "Microbiota")
labs(title = "Microbiota")
p1 + p2
```
Expand All @@ -765,46 +748,48 @@ Let us then focus on loadings of factor 2.
```{r}
#| label: plot_factor2
#| fig-width: 10
#| fig-height: 8
#| fig-height: 6
#| fig-cap: Features with the highest loadings for Factor 2.
p1 <- plot_top_weights(model, view = 1, factors = 2, nfeatures = 25) +
labs(title = "Fatty acids")
labs(title = "Fatty acids")
p2 <- plot_top_weights(model, view = 2, factors = 2, nfeatures = 25) +
labs(title = "Microbiota")
labs(title = "Microbiota")
p1 + p2
```

In the microbial data, particularly _Cetobacterium_ and _Photobacterium_ show a
positive association with Factor 2. Additionally, many fatty acids display
significant weights in this factor, though no single fatty acid can be
specifically tied to these taxa. This suggests that as the abundances of these
taxa rise, there is a corresponding increase in overall fatty acid levels.
In the microbial data, particularly _Cetobacterium_, _Vibrio_, and _Aliivibrio_
show a negative association with Factor 2. Additionally, many fatty acids
display significant negative weights in this factor, though no single fatty
acid can be specifically tied to these taxa. This suggests that as the
abundances of these taxa is rise (or decrease), there is a corresponding
increase (or decrease) in overall fatty acid levels.

Next, we visualize Factor 3 that captured variance more evenly between microbes
and fatty acids.

```{r}
#| label: plot_factor3
#| fig-width: 10
#| fig-height: 15
#| fig-cap: Features with the highest loadings for Factors 3 and 4.
p <- lapply(c(3, 4), function(x){
p1 <- plot_top_weights(model, view = 1, factors = x, nfeatures = 25) +
labs(title = paste0("Fatty acids vs factor ", x))
p2 <- plot_top_weights(model, view = 2, factors = x, nfeatures = 25) +
labs(title = paste0("Microbiota vs factor ", x))
res <- list(p1, p2)
return(res)
})
p <- unlist(p, recursive = FALSE)
#| fig-height: 6
#| fig-cap: Features with the highest loadings for Factors 3.
p1 <- plot_top_weights(model, view = 1, factors = 3, nfeatures = 25) +
labs(title = "Fatty acids")
p2 <- plot_top_weights(model, view = 2, factors = 3, nfeatures = 25) +
labs(title = "Microbiota")
wrap_plots(p, ncol = 2)
p1 + p2
```

From the shared Factors 3 and 4, _Aliivibrio_, _Cetobacterium_, _Vibrio_, and
_Photobacterium_ emerge prominently. Their association with the 22:1n-11 fatty
acid and omega-3 fatty acids suggests a potential biological link between these
microbes and specific fatty acid profiles, hinting at a functional relationship.
From the shared Factor 3, _Photobacterium_ emerge prominently. Similarly to
Factor 2, no single fatty acid can be directly associated with
_Photobacterium_.

Worth noting is that, out of these 5 taxa, only _Mycoplasma_ does not appear to
share any variability with fatty acids as all its variability was captured by
the first factor which did not associate with fatty acids.

## Conclusions

Expand Down

0 comments on commit 413fee3

Please sign in to comment.