b_case_studies.Rmd

---
title: "Case studies"
author:
- name: Martin Morgan
  affiliation: Roswell Park Comprehensive Cancer Center
  email: Martin.Morgan@RoswellPark.org
package: cellxgenedp
output:
  BiocStyle::html_document
abstract: |  
  This article summarizes short case studies and solutions arising
  from user queries.
vignette: >
  %\VignetteIndexEntry{Case studies}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

# Setup

For each case study, ensure that cellxgenedp (see the
[Bioconductor][cellxgenedp-bioc] package landing page, or
[GitHub.io][cellxgenedp] site) is installed (additional installation
options are at <https://mtmorgan.github.io/cellxgenedp/>).

[cellxgenedp-bioc]: https://bioconductor.org/packages/cellxgenedp
[cellxgenedp]: https://mtmorgan.github.io/cellxgenedp

```{r install, eval = FALSE}
if (!"BiocManager" %in% rownames(installed.packages()))
    install.packages("BiocManager", repos = "https://CRAN.R-project.org")
BiocManager::install("cellxgenedp")
```

Load the package.

```{r setup, message = FALSE}
library(cellxgenedp)
```

# Case study: authors & datasets

## Challenge and solution

This case study arose from a question on the CZI Science Community
Slack. A user asked

> Hi! Is it possible to search CELLxGENE and identify all datasets by
> a specific author or set of authors?

Unfortunately, this is not possible from the [CELLxGENE][] web site --
authors are only associated with collections, and collections can only
be sorted or filtered by title (or publication / tissue / disease /
organism).

[CELLxGENE]: https://cellxgene.cziscience.com/

A [cellxgenedp][] solution uses `authors()` to discover authors and
their collections, and joins this information to `datasets()`.

```{r}
author_datasets <- left_join(
    authors(),
    datasets(),
    by = "collection_id",
    relationship = "many-to-many"
)
author_datasets
```

`author_datasets` provides a convenient point from which to make basic
queries, e.g., finding the authors contributing the most datasets.

```{r}
author_datasets |>
    count(family, given, sort = TRUE)
```

Perhaps one is interested in the most prolific authors based on
'collections', rather than 'datasets'. The five most prolific authors
by collection are

```{r prolific authors}
prolific_authors <-
    authors() |>
    count(family, given, sort = TRUE) |>
    slice(1:5)
prolific_authors
```

The datasets associated with authors are

```{r prolific-author-datasets}
right_join(
    author_datasets,
    prolific_authors,
    by = c("family", "given")
)
```

Alternatively, one might be interested in specific authors.  This is
most easily accomplished with a simple filter on `author_datasets`, e.g.,

```{r specific-authors}
author_datasets |>
    filter(
        family %in% c("Teichmann", "Regev", "Haniffa")
    )
```

or more carefully by constructing at `data.frame` of family and given
names, and performing a join with `author_datasets`

```{r authors-of-interest}
authors_of_interest <-
    tibble(
        family = c("Teichmann", "Regev", "Haniffa"),
        given = c("Sarah A.", "Aviv", "Muzlifah")
    )
right_join(
    author_datasets,
    authors_of_interest,
    by = c("family", "given")
)
```

## Areas of interest

There are several interesting questions that suggest themselves, and
several areas where some additional work is required.

It might be interesting to identify authors working on similar
disease, or other areas of interest. The `disease` column in the
`author_datasets` table is a list.

```{r disease}
author_datasets |>
    select(family, given, dataset_id, disease)
```

This is because a single dataset may involve more than one
disease. Furthermore, each entry in the list contains two elements,
the `label` and `ontology_term_id` of the disease. There are two
approaches to working with this data.

One approach to working with this data uses facilities in
[cellxgenedp][] as outlined in an accompanying article. Discover
possible diseases.

```{r disease-facets}
facets(db(), "disease")
```

Focus on `COVID-19`, and use `facets_filter()` to select relevant
author-dataset combinations.

```{r disease-facet-filter}
author_datasets |>
    filter(facets_filter(disease, "label", "COVID-19"))
```

Authors contributing to these datasets are

```{r disease-facet-fitler-authors}
author_datasets |>
    filter(facets_filter(disease, "label", "COVID-19")) |>
    count(family, given, sort = TRUE)
```

A second approach is to follow the practices in [R for Data
Science][r4ds], the `disease` column can be 'unnested' twice, the
first time to expand the `author_datasets` table for each disease, and
the second time to separate the two columns of each disease.

```{r disease-unnest}
author_dataset_diseases <-
    author_datasets |>
    select(family, given, dataset_id, disease) |>
    tidyr::unnest_longer(disease) |>
    tidyr::unnest_wider(disease)
author_dataset_diseases
```

Author-dataset combinations associated with COVID-19, and contributors
to these datasets, are

```{r covid-19, eval = FALSE}
author_dataset_diseases |>
    filter(label == "COVID-19")

author_dataset_diseases |>
    filter(label == "COVID-19") |>
    count(family, given, sort = TRUE)
```

These computations are the same as the earlier iteration using
functionality in [cellxgenedp][].

A further resource that might be of interest is the [OSLr][] package
article illustrating how the ontologies used by CELLxGENE can be
manipulated to, e.g., identify studies with terms that derive from a
common term (e.g., all disease terms related to 'carcinoma').

[r4ds]: https://r4ds.hadley.nz/rectangling
[OLSr]: https://mtmorgan.github.io/OLSr/articles/

## Collaboration

TODO.

It might be interesting to know which authors have collaborated with
one another. This can be computed from the `author_datasets` table,
following approaches developed in the [grantpubcite][] package to
identify collaborations between projects in the NIH-funded ITCR
program. See the graph visualization in the [ITCR collaboration][]
section for inspiration.

[grantpubcite]: https://mtmorgan.github.io/grant
[ITCR collaboration]: https://mtmorgan.github.io/grantpubcite/articles/case_study_itcr.html#itcr-collaboration

## Duplicate collection-author combinations

Here are the authors

```{r}
authors <- authors()
authors
```

There are `r nrow(authors)` collection-author combinations. We expect
these to be distinct (each row identifying a unique collection-author
combination). But this is not true

```{r}
nrow(authors) == nrow(distinct(authors))
```

Duplicated data are

```{r}
authors |> 
    count(collection_id, family, given, consortium, sort = TRUE) |>
    filter(n > 1)
```

Discover details of the first duplicated collection,
`e5f58829-1a66-40b5-a624-9046778e74f5`

```{r}
duplicate_authors <-
    collections() |>
    filter(collection_id == "e5f58829-1a66-40b5-a624-9046778e74f5")
duplicate_authors
```
 The author information comes from the `publisher_metadata` column
 
```{r}
publisher_metadata <-
    duplicate_authors |>
    pull(publisher_metadata)
```

This is a 'list-of-lists', with relevant information as elements in
the first list

```{r}
names(publisher_metadata[[1]])
```

and relevant information in the `authors` field, of which there are 221

```{r}
length(publisher_metadata[[1]][["authors"]])
```

Inspection shows that there are four authors with family name `Pisco`
and given name `Angela Oliveira`: it appears that the data provided by
CZI indeed includes duplicate author names.

From a pragmatic perspective, it might make sense to remove duplicate
entries from `authors` before down-stream analysis.

```{r}
deduplicated_authors <- distinct(authors)
```

Tools that I have found useful when working with list-of-lists style
data rare [listviewer::jsonedit()][listviewer] for visualization, and
[rjsoncons][] for filtering and querying these data using JSONpointer,
JSONpath, or JMESpath expression (a more R-centric tool is the
[purrr][] package).

[listviewer]: https://CRAN.r-project.org/package=listviewer
[rjsoncons]: https://CRAN.r-project.org/package=rjsoncons
[purrr]: https://CRAN.r-project.org/package=purrr

### What is an 'author'?

The combination of family and given name may refer to two (or more)
different individuals (e.g., two individuals named 'Martin Morgan'),
or a single individual may be recorded under two different names
(e.g., given name sometimes 'Martin' and sometimes 'Martin T.'). It is
not clear how this could be resolved; recording ORCID identifiers
migth help with disambiguation.

# Case study: using ontology to identify datasets

This case study was developed in response to the following Slack
question:

> CELLxGENE's webpage is using different ontologies and displaying
> them in an easy to interogate manner (choosing amongst 3 possible
> coarseness for cell types, tissues and age) I was wondering if this
> simplified tree of the 3 subgroups for cell type, tissue and age
> categories was available somewhere?

As indicated in the question, CELLxGENE provides some access to
ontologies through a hand-curated three-tiered classification of
specific facets; the tiers can be retrieved from publicly available
code, but one might want to develop a more flexible or principled
approach.

CELLxGENE dataset facets like 'disease' and 'cell type' use terms from
ontologies. Ontologies arrange terms in directed acyclic graphs, and
use of ontologies can be useful to identify related datasets. For
instance, one might be interesed in cancer-related datasets (derived
from the 'carcinoma' term in the corresponding ontology) in general,
rather than, e.g., 'B-cell non-Hodgkins lymphoma'. 

In exploring this question in *R*, I found myself developing the
[OLSr][] package to query and process ontologies from the EMBL-EBI
[Ontology Lookup Service][OLS]. See the '[Case Study: CELLxGENE
Ontologies][OLSr-case-study]' article in the OLSr package for full
details.

[OLSr]:https://mtmorgan.github.io/OLSr
[OLS]: https://www.ebi.ac.uk/ols4/
[OLSr-case-study]: https://mtmorgan.github.io/OLSr/articles/b_case_study_cxg.html

# Session information {.unnumbered}

```{r sessionInfo, echo = FALSE}
sessionInfo()
```