Sam Tyner

The free online book [*Open Forensic Science in
R*](https://sctyner.github.io/OpenForSciR/) was created to foster open
science practices in the forensic science community. It is comprised of
eight chapters: an introduction and seven chapters covering different
areas of forensic science: the validation of DNA interpretation systems,
firearms analysis of bullets and casings, latent fingerprints, shoe
outsole impressions, trace glass evidence, and decision-making in
forensic identification tasks. The chapters of *Open Forensic Science in
R* have the same five sections: Introduction, Data, R Package(s),
Drawing Conclusions, and Case Study. There is R code throughout the
chapter to guide the reader along in an analysis, and the case study
walks the reader through solving a forensic science problem in R, from
reading the data to answering a specific question such as, "Were these
two bullets fired by the same gun?"

### The Center for Statistics and Applications in Forensic Evidence (CSAFE)

To help more scholars access forensic science research, *Open Forensic
Science in R* brings together many open resources created and/or used by
the Center for Statistics and Applications in Forensic Evidence
([CSAFE](https://forensicstats.org/)) and the National Institute for
Standards and Technology ([NIST](https://www.nist.gov/)). CSAFE was
founded in 2015 with the mission of building up the statistical
foundations in forensic science. CSAFE is an interdisciplinary NIST
[Center of Excellence](https://www.nist.gov/coe) comprised of four
institutions: Iowa State University (ISU), Carnegie Mellon University
(CMU), University of California Irvine (UCI), and University of Virginia
(UVA). The statistics faculty members, postdoctoral researchers, and
graduate students working with CSAFE have written many R packages to
complete a variety of forensic science tasks, from analyzing shoeprint
impressions to comparing marking on bullets. *Open Forensic Science in
R* brings many of these forensic science R packages together in one
place.

CSAFE is
[committed](https://forensicstats.org/blog/2017/11/29/open-source-software-applications-forensics-courtroom/)
to open science, and many CSAFE researchers contributed to *Open
Forensic Science in R*. As most CSAFE researchers are statisticians,
much of CSAFE's research uses R in some capacity. Dedication to open
source is vital for advancing the field of forensic science because the
current barrier to entry into forensic science research is very high:
equipment is very expensive and most software in the field is
proprietary, so no one outside of the company selling the software knows
exactly how it works. These barriers limit the number of researchers who
have access to the field. There are also limited data available to
researchers. In many cases, data are limited for practical reasons
because they are from real, identifiable people and investigations.
CSAFE is committed to releasing as much data as possible to help advance
forensic science research, and hosts a large [Forensic Science Data
Portal](https://forensicstats.org/data/) to make the data widely
available. *Open Forensic Science in R* brings many of the open
resources available to current and future forensic science researchers
together in one place to encourage openness in the field.

### Example: Comparing Bullets

::: {.gallery .caption-position-bottom .caption-effect-slide .hover-effect-zoom .hover-transition itemscope="" itemtype="http://schema.org/ImageGallery"}
::: {.box}
<figure itemprop="associatedMedia" itemscope=""
itemtype="http://schema.org/ImageObject">
<div class="img"
style="background-image: url(&#39;/img/blog-images/2019-08-20-forensic-science/bullet1.png&#39;);">
<img src="/img/blog-images/2019-08-20-forensic-science/bullet1.png"
itemprop="thumbnail" alt="bullet 1" />
</div>
<a href="/img/blog-images/2019-08-20-forensic-science/bullet1.png"
itemprop="contentUrl"></a>
<figcaption><p>bullet 1</p></figcaption>
</figure>
:::

::: {.box}
<figure itemprop="associatedMedia" itemscope=""
itemtype="http://schema.org/ImageObject">
<div class="img"
style="background-image: url(&#39;/img/blog-images/2019-08-20-forensic-science/bullet2.png&#39;);">
<img src="/img/blog-images/2019-08-20-forensic-science/bullet2.png"
itemprop="thumbnail" alt="bullet 2" />
</div>
<a href="/img/blog-images/2019-08-20-forensic-science/bullet2.png"
itemprop="contentUrl"></a>
<figcaption><p>bullet 2</p></figcaption>
</figure>
:::
:::

The chapter ["Firearms:
bullets"](https://sctyner.github.io/OpenForSciR/bullets.html) begins by
introducing the reader to the terminology of firearms and bullets, and
describes the methods used in forensic science to compare bullets. Then,
the chapter discusses the open source work by current and former CSAFE
researchers. Three dimensional bullet scans, the data of interest, are
stored in the [x3p](https://sourceforge.net/p/open-gps/mwiki/X3p/)
standard format, and the R package
[`x3ptools`](https://heike.github.io/x3ptools/) is used to read the
bullet scans in R ^[1](#fn:1){.footnote-ref role="doc-noteref"}^. Then,
the bullet data are analyzed with the
[`bulletxtrctr`](https://heike.github.io/bulletxtrctr/) package
^[2](#fn:2){.footnote-ref role="doc-noteref"}^. Each bullet is comprised
of many surface scans corresponding to the number of
[lands](https://sctyner.github.io/OpenForSciR/glossary.html#def:gunbarrel)
on the bullet which came in contact with the gun barrel when fired. The
number of surface scans varies by type and manufacturer of the gun. For
a comparison of two bullets, a representative cross-section of each 3D
surface scan is selected, the curve is removed, some noise is removed,
and only the smoothed *bullet signature* remains. (See Figure 2.) This
procedure is repeated for all lands of the bullets of interest, and the
signatures are compared to each other using a trained random forest
available in the package, resulting in scores from zero to one
indicating how similar the two signatures are ^[3](#fn:3){.footnote-ref
role="doc-noteref"}^. In this case, bullets 1 and 2 were fired from the
same gun, which can be seen by comparing the bullet 1 lands 2, 3, 4, 5,
6, 1 to the bullet 2 lands 3, 4, 5, 6, 1, 2, respectively. For complete
details, see the [Case
Study](https://sctyner.github.io/OpenForSciR/bullets.html#case-study-1)
section of the chapter.

::: {.gallery .caption-position-bottom .caption-effect-slide .hover-effect-zoom .hover-transition itemscope="" itemtype="http://schema.org/ImageGallery"}
::: {.box style="max-width:50%"}
<figure itemprop="associatedMedia" itemscope=""
itemtype="http://schema.org/ImageObject">
<div class="img"
style="background-image: url(&#39;https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cscrosscut-1.png&#39;);">
<img
src="https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cscrosscut-1.png"
itemprop="thumbnail" />
</div>
<a
href="https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cscrosscut-1.png"
itemprop="contentUrl"></a>
</figure>
:::

::: {.box style="max-width:50%"}
<figure itemprop="associatedMedia" itemscope=""
itemtype="http://schema.org/ImageObject">
<div class="img"
style="background-image: url(&#39;https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cssigs-1.png&#39;);">
<img
src="https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cssigs-1.png"
itemprop="thumbnail" />
</div>
<a
href="https://sctyner.github.io/OpenForSciR/open-forensic-science-in-r_files/figure-html/bullets-cssigs-1.png"
itemprop="contentUrl"></a>
</figure>
:::
:::

### How to Contribute & Acknowledgements

The book will continue to grow, and contributions are welcome via issues
or pull requests on the [Github
repo](https://github.com/sctyner/OpenForSciR). If you would like to
contribute, please follow our [Contributor Code of
Conduct](https://github.com/sctyner/OpenForSciR/blob/master/CODE_OF_CONDUCT.md).
Thank you to the many CSAFE and NIST researchers who contributed to this
project: [Dr. Heike Hofmann](https://hofmann.public.iastate.edu/) (ISU),
Dr. Soyoung Park (ISU), [Xiao Hui Tai](http://www.stat.cmu.edu/~xtai/)
(CMU), [Dr. Eric Hare](https://twitter.com/ericrhare) (Omni Analytics,
formerly ISU), [Dr. Karen
Kafadar](https://statistics.as.virginia.edu/faculty-staff/profile/kk3ab)
(UVA), Karen Pan (UVA), [Dr. Amanda
Luby](https://twitter.com/amandaluby) (CMU), and Dr. Sarah Riman (NIST).
Finally, a big thank you goes to the [rOpenSci Fellowship
program](/blog/2018/02/14/announcing-2018-ropensci-fellows/) for funding
this project.

::: {.footnotes role="doc-endnotes"}

------------------------------------------------------------------------

1.  ::: {#fn:1}
    Heike Hofmann, Susan Vanderplas, Ganesh Krishnan and Eric Hare
    (2019). x3ptools: Tools for Working with 3D Surface Measurements. R
    package version 0.0.2.9000.
    <https://github.com/heike/x3ptools> [↩︎](#fnref:1){.footnote-backref
    role="doc-backlink"}
    :::

2.  ::: {#fn:2}
    Heike Hofmann, Susan Vanderplas and Ganesh Krishnan (2018).
    bulletxtrctr: Automatic Matching of Bullet Striae. R package version
    0.2.0.
    <https://heike.github.io/bulletxtrctr/> [↩︎](#fnref:2){.footnote-backref
    role="doc-backlink"}
    :::

3.  ::: {#fn:3}
    Hare, Eric, Heike Hofmann, and Alicia Carriquiry. 2017. "Automatic
    Matching of Bullet Land Impressions." *The Annals of Applied
    Statistics* 11 (4): 2332--56. [↩︎](#fnref:3){.footnote-backref
    role="doc-backlink"}
    :::
:::


The free online book
<em>
 Open Forensic Science in R
</em>
was created to foster open science practices in the forensic science community.

Introducing Open Forensic Science in R

Karthik Ram

rOpenSci's mission is to enable and support a thriving community of
researchers who embrace open and reproducible research practices as part
of their work. Since our inception, one of the mechanisms through which
we have supported the community is by developing high-quality open
source tools that lower barriers to working with scientific data.
Equally important to our mission is to build capacity and promote
researchers who are engaged in such practices within their disciplinary
communities. This fellowship program is a unique opportunity for us to
enable such individuals to have a bigger voice in their communities.

This year, a diverse committee (comprised of [Di
Cook,](https://research.monash.edu/en/persons/dianne-cook) [Mine
Cetinkaya-Rundel](http://www2.stat.duke.edu/~mc301/), [Matt
Jones](https://matt.magisa.org/), [Ken Benoit](https://kenbenoit.net/)
and myself) reviewed 64 applications from researchers in various
disciplines to select four winners. Emerging from this impressive pool
of applicants are four outstanding researchers who constitute the 2018
Fellows. They are **Dr. Samantha Tyner** (Iowa State University), **Mr.
Zachary Foster** (Oregon State University), **Ms. Kelly Hondula**
(SeSync, University of Maryland) and **Dr. Jon Zelner** (University of
Michigan).

### Meet our 2018 Fellows

#### Sam Tyner

When you hear the term "forensic science" what comes to mind? If you're
like most Americans, you probably think of TV shows like "CSI" or
"Criminal Minds", where forensic science is used to solve crimes in
incredibly fast, effective, and technologically sophisticated ways. In
reality, however, most forensic science moves slowly, is very expensive,
and relies on human judgement, not sophisticated technologies. Where
advanced machines and algorithms exist, such as in DNA analyses, the
current best and most widely used tools are proprietary, siloed, and
extremely expensive. At the [Center for Statistics and Applications in
Forensic Evidence (CSAFE)](https://forensicstats.org/), our primary goal
is to improve the foundation of forensic feature comparison techniques
in a transparent way. It has been well-established that many of these
forensic science disciplines lack the scientific foundation of
single-source DNA comparison, which is considered the "gold standard" of
forensic science. CSAFE is committed to establishing the scientific and
statistical foundations of these evidence types using open source tools
to increase the overall transparency of forensic science disciplines.

In the R community, where open source rules, users have developed some
packages with forensic science applications, such as glass transfer and
DNA comparisons, while additional R packages for other feature
comparison methods, including bullet and cartridge case identification,
shoe sole impressions, glass element analysis, and handwriting source
analysis, are currently in development at CSAFE institutions. In order
to further establish the open-source foundation in forensic science, and
with the award of the rOpenSci fellowship, Sam Tyner, a postdoc at
CSAFE, and Heike Hofmann, professor of statistics at Iowa State
University, will create an online book, "Open Forensic Science in R,"
which will be made freely available on CSAFE's website. Each chapter of
the book will walk the reader through a forensic analysis step-by-step
using an R package uniquely designed for that forensic discipline. We
also hope to contribute to many of these packages by authoring package
vignettes and making significant additions to their documentation.
Through writing this book, we want to set an example for current and
future forensic statisticians by participating in cross-disciplinary and
cross-institution collaboration, both within and outside of CSAFE. Our
ultimate goal in authoring this book is to lower the financial barriers
to entry when performing forensic science analyses, both for researchers
in academia and for small-scale forensic laboratories. By improving
access to some of the tools needed by practitioners and researchers, the
field of forensic science can start to reap the benefits of open source
technologies, just like data science has in the last decade.

#### About Sam

<figure class="pull-left">
<img src="/img/team/sam-tyner.jpg" width="150" alt="Sam Tyner" />
</figure>

Sam Tyner holds a PhD in Statistics from Iowa State University, where
she is currently a postdoctoral research associate with CSAFE, focusing
on education, outreach, and computing. Her other research interests
outside of forensics include statistical graphics and social network
analysis & visualization. She is also the co-founder and co-organizer of
the Ames chapter of R-Ladies, and is a proud alumna of Augustana
College.

[Twitter](https://twitter.com/sctyner),
[GitHub](https://github.com/sctyner), [Web
page](https://sctyner.github.io/)

------------------------------------------------------------------------

#### Zachary Foster

Affordable high-throughput DNA sequencing has led to a massive increase
in ecological research that uses DNA sequences to record and classify
organisms. DNA from a sample of water can reveal the species of fish
living in a river. Every insect species can be identified from samples
collected in a trap without any expertise in entomology. DNA from an
animal's droppings can tell us about its diet. This technique has
resulted in the recent explosion of microbiome research. It has lead to
the realization that every organism (plants, animals, and even us
humans) is an ecosystem or "superorganism" reliant on a diverse
community of microbes for essential functions; a sterile homo sapiens is
a dead homo sapiens.

The R community has responded to this abundance of data by making
numerous packages that work with the taxonomic makeup of communities
(e.g. `phyloseq`) or query the ever-larger reference databases
containing taxonomic information associated with sequences or occurrence
data (e.g.` taxize`, `gbif`). However, taxonomic data is hierarchical,
and hierarchical data is hard to deal with compared to tabular data, so
each package typically has its own solution to the problems of parsing,
encoding, and manipulating taxonomic data. This is also true for the
reference databases and other tools outside of R. In many cases, the
hierarchical nature of the data is simply ignored and only specific
ranks in the taxonomy (e.g. genus) are considered.With this fellowship,
we are attempting to address these issues by introducing a standard in R
for parsing, encoding, manipulating, and visualizing taxonomic data and
any application-specific information associated with it. We are
developing the `taxa` package to provide an all-purpose set of taxonomic
classes and functions to manipulate them. We are also developing the
`metacoder` package, which builds on taxa and adds graphing of taxonomic
information and tools intended for metagenomics research.

The impact of this work will ultimately rely on adoption by the research
community as a standard, or at least a valid option, for working with
taxonomic data in R. We are working to adapt parts of taxize, the
primary package for querying online databases with taxonomic
information, to use the classes in taxa. We hope that the generally
useful functionality provided by the taxize and metacoder packages will
jump-start the adoption of the taxa framework.

#### About Zachary

<figure class="pull-left">
<img src="/img/team/zachary-foster.jpg" width="150" />
</figure>

Zachary Foster is an Oregon State University graduate student working
with Niklaus Grunwald working on the communities of fungi and oomycetes
inhabiting woody plants in natural and nursery settings. He uses R
extensively for his research and develops code aimed at making
visualization of taxonomic easier and more informative.

[Twitter](https://twitter.com/zacharyfoster19),
[GitHub](https://github.com/zachary-foster)

------------------------------------------------------------------------

#### Kelly Hondula

Ecologists, hydrologists, and soil scientists collaborating in watershed
science use diverse methods to gather data about ecosystems. Each
collaborator's workflow involves a different set of disciplinary
expertise to derive results, such as modifying lab procedures,
developing calibration curves, or nuanced site-specific corrections.
This results in many different strategies for organizing and managing
data based on the relevant units and timescales of observations, and
whether they are based on manually recorded observations, physical
samples, or machine-generated data from sensors or lab equipment. These
workflows present a challenge for reproducibility because they are
necessarily idiosyncratic and highly contextualized. For example, sensor
data often goes through multiple levels of interpretation for quality
control to identify periods where data are compromised from maintenance,
power failures, biofouling, or other malfunctions. Integrating these
time series data with sample-based data has been a particular challenge
in watershed science and is typically handed with time consuming, custom
approaches that are difficult to document and reproduce.

With the rOpenSci fellowship, I'll be creating tools and training
material to help manage these diverse types of data with workflows in R
structured around an information model called ODM2 (the Observations
Data Model Version 2). This data model was developed in the hydrology
and geoscience community, and was specifically designed to make earth
science data more interoperable across networks of research sites. My
goal is to make it easy for ecologists to adopt ODM2's concepts and
vocabulary for projects that involve studying ecosystems through a
combination of field monitoring, sample collection and analysis,
laboratory studies, and computer modeling, even if that project is not
part of a large network with cyberinfrastructure resources and dedicated
data managers. Datasets using this common framework would have greater
re-use potential, and could be shared with a wider community in existing
long-term data repositories.

I'll be developing a set of modules based on common procedures in
aquatic ecology, with examples and guidelines for how to structure a
reproducible workflow and "translate" it into the ODM2 data model. Each
module will have a set of R functions, template data sheets, and
vignettes with conceptual diagrams describing how to use the workflow.
To complement these modules, I'll be using packages like `RShiny`,
`leaflet`, and `dygraphs` to create visualization tools to help
interpret biogeochemistry and hydrology datasets that use the ODM2
framework. My overall goal is to have these tools make it easier for
ecologists, especially those comfortable using R for data analysis, to
seamlessly integrate components of the ODM2 framework into the workflows
they are already using.

#### About Kelly

<figure class="pull-left">
<img src="/img/team/kelly-hondula.jpg" width="150" />
</figure>

Kelly is an aquatic ecologist working on her PhD in the MEES (Marine,
Estuarine, and Environmental Sciences) Program at the University of
Maryland. Her dissertation is on linking hydrology and methane in
wetlands. She also works at the National Socio-Environmental Synthesis
Center, where she provides data science support for SESYNC teams and
fellows.

[Twitter](https://twitter.com/kellyproof),
[GitHub](https://github.com/khondula),
[Website](https://www.sesync.org/users/khondula)

------------------------------------------------------------------------

#### Jon Zelner

Infectious disease models are increasingly able to incorporate
fine-grained spatial, socio-demographic, and genetic information to
reconstruct patterns of transmission. However, this increased capability
creates new difficulties for open-source epidemiology: Publicly
releasing such detailed data in its raw form is clearly unethical
because it compromises participant privacy. At the same time, keeping
the code and data for such analyses proprietary also poses an ethical
quandary, because sharing this information is critical to both verifying
the accuracy of published research results and for allowing other
researchers to either apply published methods to new data, or to
innovate upon existing methods. These issues are particularly acute for
global health researchers, such as myself, who work with partners in
low-and-middle-income countries (LMICs), for whom closed-source
solutions may be prohibitively expensive and open-source analyses are
critical tools for education, policy, and practice.

Right now, there are two gaping holes in the set of tools necessary to
create maximally reproducible spatial analyses in epidemiology:

1.  Open-source tools for data obfuscation that de-identify
    individual-level case data while preserving important spatial motifs
    and descriptive metrics (e.g. clustering), as well as relationships
    between key individual and ecological variables.

2.  A documented, open-source, turn-key software pipeline for such
    analyses.

As part of this fellowship, my team and I are working to create a R
package (working title is `obfuscatr`) that contains implementations of
best-practice approaches to data obfuscation, such as Voronoi masking.
I'm also hoping to explore other approaches based on Approximate
Bayesian Inference (ABC) that can produce dummy datasets that recreate
key spatial features and covariate relationships without conditioning
directly on point locations. I will be documenting examples of the usage
of these tools in the context of real-world analyses using sensitive
epidemiological data, from my own research and in standalone pieces. I
anticipate these tools will show up in both primary research manuscripts
as well as blog posts/publications focusing in depth on the underlying
data obfuscation tools. I am also hoping to use this fellowship as an
opportunity to revive some earlier work I have done creating
containerized environments, e.g. Docker containers, for reproducible
Bayesian spatial analyses. Ideally, these updated containers would
include the `obfuscatr` tools described above, and would focus on
container implementations usable in HPC environments, such as
Singularity.

#### About Jon

<figure class="pull-left">
<img src="/img/team/jon-zelner.jpg" width="150" />
</figure>

Jon Zelner is a social epidemiologist focused on spatial and dynamic
modeling of endemic and epidemic infectious disease transmission. His
research covers a variety of pathogens from tuberculosis, to
drug-resistant Staph Aureus (i.e. MRSA), as well as influenza and the
wide range of pathogens causing diarrheal disease. He is particularly
interested in the ways domestic and global social inequality drive
susceptibility to infection via their effects on the physical
(households, water quality) and social environments (networks) that
people inhabit.

[Twitter](https://twitter.com/jzelner),
[GitHub](https://github.com/jzelner),
[Webpage](https://www.jonzelner.net/)


rOpenSci’s mission is to enable and support a thriving community of researchers who embrace open and reproducible research practices as part of their work. Since our inception, one of the mechanisms through which we have supported the community is by developing high-quality open source tools that lower barriers to working with scientific data.

Introducing the 2018 rOpenSci Research Fellows!

rOpenSci

Stefanie Butland

**Join our [Community Call on Tuesday, January
30th](https://communitycalls.ropensci.org/)** (January 31 for our
Australian friends)

Nick Golding, 2017 rOpenSci Fellow, will talk about two R packages he
has developed recently. [zoon](https://github.com/zoonproject/zoon) aims
to promote open and reproducible research in ecological modeling by
helping researchers share their code in a modular way and produce
reproducible research artifacts. Nick has recently been trying to
bootstrap a community around this idea and says this is a much harder
problem.

[greta](https://github.com/goldingn/greta) lets you write out and fit
statistical models (like Stan or BUGS) but right in R. It uses
tensorflow to make models scale to massive data, and is designed to be
used and extended by other modeling packages. `greta` relies on some
nice R tricks and lots of thinking about designing APIs for both users
and developers.

### Agenda

1.  Welcome (Stefanie Butland, rOpenSci Community Manager, 5 min)
2.  guest: Nick golding (35 min)
3.  Q & A (moderated by Scott Chamberlain, 20 min)

### Speaker bio

**Nick Golding** is an applied computational statistician working as a
Research Fellow in the [Quantitative & Applied Ecology
Group](https://qaeco.com/) at the University of Melbourne, Australia. He
develops statistical models and software to predict the distributions of
species and human diseases like those caused by the Zika and Ebola
viruses. He is particularly interested in improving these models with
information about traits, mechanistic relationships and population
dynamics. [Nick received an rOpenSci
Fellowship](/blog/2016/12/12/ropensci-fellowship-zoon/) to continue
developing a framework for building and sharing reproducible species
distribution models.

Nick on [GitHub](https://github.com/goldingn),
[Twitter](https://twitter.com/_nickgolding_), [Google
Scholar](https://scholar.google.co.uk/citations?user=peoal7wAAAAJ&hl)

### Resources

- [Announcing our first fellowship awarded to Dr. Nick
  Golding](/blog/2016/12/12/ropensci-fellowship-zoon/); blog post Dec
  12, 2016
- Zoon publication: Nick Golding, Tom August, Tim Lucas, David
  Gavaghan, E. Emiel van Loon, Greg McInerny (2017). The zoon r package
  for reproducible and shareable species distribution modeling. Methods
  Ecol Evol <https://doi.org/10.1111/2041-210X.12858>
- Soon after the Community Call, we'll post the
  [video](https://vimeo.com/channels/rocommunitycalls)
- [Past Community Calls](https://communitycalls.ropensci.org/)


<strong>
 Join our Community Call on Tuesday, January 30th
</strong>
(January 31 for our Australian friends)  Nick Golding, 2017 rOpenSci Fellow, will talk about two R packages he has developed recently. zoon aims to promote open and reproducible research in ecological modeling by helping researchers share their code in a modular way and produce reproducible research artifacts.

Community Call - Writing Packages to Support Research Communities - zoon &amp; greta

rOpenSci's mission is to promote a culture of open, transparent, and
reproducible research across various research domains. Everything we do,
from developing high-quality open-source software for data science and,
[software review](https://github.com/ropensci/software-review), to
building community through events like our [community
calls](https://ropensci.org/commcalls/) and [annual
unconference](https://unconf17.ropensci.org/) are all geared toward
lowering barriers to reproducible, open science.

The rOpenSci Fellowship presents a unique opportunity for researchers
who are engaged in open source to have a bigger voice in their
communities. These fellowships are designed to support individual
researchers and collaborative efforts to help them do better science,
build community around projects or best practices, or develop some tools
as part of ongoing research that could impact one or more research
domains. Two areas that are of particular interest to us are:

- **Reproducibility**The first type of fellowship we seek to support
  includes any activities (software development, best practices,
  guidelines) that promote reproducibility with one or more research
  communities. Funds from this fellowship can be used for activities
  such as finishing development of software, documentation or developing
  guidelines and best practices around methods and approaches. Your
  proposal will need to clearly identify how these funds will impact
  others in your research community and beyond.

- **Working groups**The second type of activity we seek to support is
  working groups, based on [a highly successful problem-solving
  model](https://doi.org/10.1525/bio.2011.61.11.9) pioneered by the
  [National Center for Ecological Analysis and
  Synthesis](https://www.nceas.ucsb.edu/). You propose a critical
  challenge in the area of scientific open source and reproducibility,
  and a team of experts, to collaborate intensively for a week at a
  time. The knowledge and insight your team develops must be broadly
  relevant to the research community.

Our first fellowship was awarded to [Dr. Nick
Golding](/blog/2016/12/12/ropensci-fellowship-zoon) in suppport of a
reproducibility project. Each fellowship will be in the range of
\$30,000 to \$50,000. We plan to award 2-3 fellowships (exact number to
be determined by the size, scope, and quality of applications we
receive) in 2017. We expect to have a similar call in 2018. All
successful fellows will also receive additional travel support to
participate in our [annual unconference](/blog/2017/06/02/unconf2017).

### Fellowship eligibility and criteria

1.  Fellows must be affiliated with a research institution.
2.  You must have Principal Investigator (PI) status at your institution
    to receive funds from this fellowship. If you do not have PI status,
    or do not expect to have PI status by the time these funds are
    disbursed, you will need to collaborate with someone who can fulfill
    this role. For this stage of the application, you will only need to
    list your status and if necessary, the name of person who will serve
    as your PI. We will request a formal letter of support in a later
    round.
3.  Fellowship recipients must strive to spend all funds within a year
    of the start date. While we are able to accommodate extensions for
    exceptional circumstances, we strongly prefer to support projects
    that are scoped well enough to be completed within one year.
4.  Funds can only be used to support personnel and travel. We do not
    allow purchase of equipment, or computing services. A detailed
    budget will be required in the final round (IDC capped at 10%).
5.  Funds cannot be used to organize conferences or hackathons.

## Fellowship selection committee

- [**Karthik Ram**](/about/#team), Committee ChairKarthik is a
  co-founder of rOpenSci and the project lead. Karthik is also a fellow
  at the Berkeley Institute for Data Science and the Berkeley Initiative
  in Global Change Biology at UC Berkeley. Prior to joining Berkeley, he
  earned his PhD in ecology & evolution from the University of
  California, Davis. Karthik's interests are focused on reproducible
  research, especially as it applies to global change. Much of his
  recent work focuses on building tools and services around open data
  and growing diverse data science communities.

- [**Mine Cetinkaya-Rundel**](http://www2.stat.duke.edu/~mc301/), Duke
  UniversityMine is Associate Professor of the Practice at Duke
  University and a Professional Educator at RStudio. Her work focuses on
  innovation in statistics pedagogy, with an emphasis on computation,
  reproducible research, open-source education, and student-centered
  learning. She is the author of three open-source introductory
  statistics textbooks as part of the OpenIntro project and teaches the
  popular Statistics with R MOOC on Coursera.

- [**Ken Benoit**](http://www.kenbenoit.net/), London School of
  EconomicsKenneth Benoit is Professor of Quantitative Social Research
  Methods at the London School of Economics and Political Science. He is
  active in data science and natural language processing and
  quantitative text analysis for social research. He is the creator of
  the quanteda, readtext, and spacyr R packages for text analysis.

- [**Matthew Jones**](https://www.nceas.ucsb.edu/ecoinfo#team), National
  Center for Ecological Analysis and SynthesisMatt is the Director of
  Research and Informatics at the National Center for Ecological
  Analysis and Synthesis and has spent more than a decade developing
  various global data sharing infrastructure projects such as the KNB
  Repository and the DataONE federation.

- [**Di Cook**](https://dicook.org/), Monash UniversityDi is Professor
  of Business Analytics at Monash University in Australia. Her expertise
  is in data visualisation, exploratory data analysis, and interactive
  graphics. She is an Ordinary Member of the R Foundation and a Fellow
  of the American Statistical Association.

## Applying for the fellowship

To apply, please submit a 3-page proposal that describes your goals,
expected outcomes, a tentative timeline, collaborators, and a very high
level budget (no more than a paragraph or two) at
ropensci.org/fellowships (now closed). As part of your submission form,
you will also have to provide details about your affiliation, PI status,
and the name of your PI (if you cannot serve in that role). For
applicants that make the short list, we will request a detailed budget
and may also ask for additional clarification.

**Application deadline is September 1, 2017 (midnight PST).**

Questions about the fellowship are welcome at <fellowships@ropensci.org>

Expect a FAQ page shortly as we receive questions.


rOpenSci’s mission is to promote a culture of open, transparent, and reproducible research across various research domains. Everything we do, from developing high-quality open-source software for data science and, software review, to building community through events like our community calls and annual unconference are all geared toward lowering barriers to reproducible, open science.

Announcing the rOpenSci Fellowships Program

rOpenSci's overarching mission is to promote a culture of transparent,
open, and reproducible research across various scientific communities.
All of our activities are geared towards lowering barriers to
participation, and building a community of practitioners around the
world. In addition to developing and maintaining a large suite of open
source tools for data science, we actively support the research
community with expert review on research software development, community
calls, and hosting annual unconferences around the world.

To further promote reproducible research activities within specialized
domains, we will soon be announcing our annual fellowship program. These
fellowships will provide support for collaborative research projects
that strengthen one or more of rOpenSci's core efforts. These activities
may include development of new domain-specific tools, or supporting
projects that are exemplars of reproducible research. Our early
fellowships are by invitation and an open call will be forthcoming in
early 2017.

::: {.float-left}
<figure class="m-0">
<img src="/img/team/nick-golding.jpg"
style="  object-fit: cover; object-position: center;  height: 250px;  width: 250px; margin-right: 15px"
data-zgotmplz="" alt="Dr Nick Golding, rOpenSci Fellow" />
<p>We are pleased to award the first fellowship to <a
href="https://scholar.google.co.uk/citations?user=peoal7wAAAAJ&amp;hl=en">Dr.
Nick Golding</a> to continue developing a framework for building and
sharing reproducible species distribution models (SDM). Nick Golding is
a Research Fellow in the <a href="https://qaeco.com/">Quantitative &amp;
Applied Ecology Group</a> at the University of Melbourne, Australia. He
develops statistical models and software to predict the distributions of
species and human diseases like those caused by the Zika and Ebola
viruses. Nick is particularly interested in improving these models with
information about traits, mechanistic relationships and population
dynamics.</p>
</figure>
:::

We interviewed Nick to learn more about him, the project, and what the
fellowship will help him and his collaborators accomplish.

------------------------------------------------------------------------

### Q: What is the Zoon Project?

The Zoon Project began in the Computer Science department in Oxford as
part of the EPSRC-funded [2020 Science
project](http://www.2020science.net/), and is a collaboration between [a
group of researchers](https://zoonproject.wordpress.com/user-panel/)
with interests in SDM, scientific software and building better research
communities. The name (Zoon) refers to an individual that is part of a
compound organism, like the polyps that make up jellyfish. It's a bit of
a geeky name (you can blame Greg McInerny for that), but it fits well
with the project's aim to improve collaboration in this area of science.

The main piece of the project is an R package:
[zoon](https://github.com/zoonproject/modules); that allows you to
develop your SDM analysis as a reproducible workflow. SDM involves
taking records on where species have been observed and the environmental
characteristics of those locations, representing these mathematically,
and applying algorithms to turn these into predictions of geographical
species distribution. The field of SDM within ecology has boomed in the
last fifteen years^[1](#fn:1){.footnote-ref role="doc-noteref"}^ ([free
pdf](https://www.researchgate.net/profile/Fabiana_Barbosa4/publication/279750919_Characteristics_of_the_top-cited_papers_in_species_distribution_predictive_models/links/559963f108ae21086d25abe2.pdf))
coinciding with the increase in data and the birth of machine learning.
The challenge is that the field is now a crowded and confusing place.
There are hundreds of different candidate models, and probably thousands
of papers on SDM methods, with each new method generally evaluated
against a different dataset. That poses a serious hurdle to the
community objectively evaluating the available methods and can be
baffling to newcomers to the field.

In other applications of machine learning, there are benchmark datasets
used by everyone in the community to evaluate methods (e.g.
[MNIST](http://yann.lecun.com/exdb/mnist/) and [many
more](https://archive.ics.uci.edu/ml/datasets)). There are also software
platforms (like [caret](https://topepo.github.io/caret/)) that enable
many of the candidate methods to be run from the same software, and
compared on the same dataset. We don't currently have either of these
things for SDM, which makes it really difficult to synthesize
information from the literature. The motivation for the Zoon Project is
to address this reproducibility challenge faced by the community.

![Zoon
workflow](/assets/blog-images/2016-12-12-ropensci-fellowship-zoon/zoon_workflow.png)

The Zoon package implements SDM analyses as workflows of bite-sized,
interoperable software modules. Each module represents one of the five
key steps in an SDM analysis.

1.  Get data on where species occur
2.  Get environmental data
3.  Pre-process data
4.  Plug data into a model
5.  Generate some outputs

Multiple modules can be used at each stage, allowing them to be combined
or compared and because modules can be switched in and out it's easy to
quickly see how an analysis would change if a different method or
dataset were used.

What makes the Zoon project unique is that all of these modules are
developed by the community (not the Zoon team) and held together in an
online repository, just like community-contributed R packages (but much
easier). Basically, each software module has to conform to a set of
input and output formats to make sure the modules are interoperable, and
also requires some basic documentation. The
[zoon](https://github.com/zoonproject/modules) R package helps users to
write, document, test and share these modules so that everyone, not just
experienced R developers, can contribute modules.

------------------------------------------------------------------------

### Q: What's an example of a research question that can be addressed using the Zoon R package?

SDM is increasingly being used in epidemiology and public health
research. For example, the recent outbreak of Ebola in West Africa
occurred in a place people weren't really expecting, so being able to
anticipate where future outbreaks might start is pretty important.
During the outbreak, I worked with some colleagues to develop an SDM to
identify areas in Africa that are environmentally similar to those where
previous human outbreaks have started^[2](#fn:2){.footnote-ref
role="doc-noteref"}^
([code](https://github.com/SEEG-Oxford/ebola_zoonotic)). This used
location data for humans and animals infected with Ebola, environmental
covariates like vegetation, elevation, temperature and
evapotranspiration, and a machine learning method known in the SDM world
as Boosted Regression Tree modelling. We didn't have Zoon back then, but
it would have made that whole workflow quicker to put together, and
really easy to quickly evaluate the effects of changing model
assumptions on areas predicted to be at risk of future outbreaks.

![Figure 5B, Pigott et al
2014](/assets/blog-images/2016-12-12-ropensci-fellowship-zoon/elife-04395-fig5b-v2_crop.jpg)**Predicted
geographical distribution of the zoonotic niche for Ebola
virus.**^[3](#fn:3){.footnote-ref role="doc-noteref"}^ *The scale
reflects the relative probability that zoonotic transmission of Ebola
virus could occur at these locations; areas closer to 1 (red) are more
likely to harbour zoonotic transmission than those closer to 0 (blue).
(Figure 5B from Pigott et al 2014. CC BY 4.0)*

------------------------------------------------------------------------

### Q: What would you like to accomplish with your fellowship?

We've already done most of the technical work in setting up the software
and online repository for Zoon, what we really need now is to build a
community of SDM users who can apply and contribute modules to the
online [repository](https://github.com/zoonproject/modules), as well as
SDM experts to help the community start to synthesize all of the
published research. During the term of the fellowship I'll be developing
an online hub for community-contributed SDM resources (SDM guides for
beginners and best practice guidelines around different aspects of SDM)
to make it easier for new SDM users to get started, and a forum where
SDM experts can discuss what these best practice guides should look
like.

**The Zoon Hub will include:**

- A single web-based portal for new users
- A basic introduction with tutorials to enable someone new to SDM to
  fit a model using the R package and share their workflow with others.
  The objective here is to make Zoon easy to use so new SDM users get
  started using software with reproducibility baked in.
- A series of expert-written best practice guides, covering contentious
  or nuanced topics in SDM methodology, like dealing with biases in
  occurrence data. Each of these guides will be backed up by
  reproducible examples in the form of Zoon workflows. We'll be
  soliciting feedback on these guides from SDM methods experts via an
  open online forum. We'll be encouraging any disputes about best
  practice to be resolved in an evidence-based way by sharing workflows
  that demonstrate a point.

We want new SDM users to try out the tutorials and guides and tell us
how they're working. We have a group of experts who will advise us on
best practices, but we'll invite others to tell us what's wrong with our
suggestions. We'd also love to hear advice on infrastructure from people
who live and breathe reproducible software. Our ultimate aim is to build
a self sustaining community suggesting, discussing, and implementing new
elements.

The rOpenSci fellowship allows us to complete this next step of Zoon
development, moving the discussion from one of opinions on which models
are best, to one of testable reproducible evidence-based answers.

------------------------------------------------------------------------

### Q: How did you arrive at working in an environment where you apply computational statistical skills to epidemiology?

I started with an undergraduate degree in ecology from John Moores
University in Liverpool and then worked as research assistant, briefly
doing real biology (even some pipetting in a lab). I then did my PhD in
mosquito ecology at Oxford, during which time I steadily drifted from
field-based research to statistical ecology and particularly SDM. After
three years as a Postdoc in Oxford applying these methods to mapping
diseases, I moved to the University of Melbourne in February 2016, where
I'm a Research Fellow. My fellowship in Melbourne is about developing
new statistical methods and software to improve our ability to map
species and diseases.

------------------------------------------------------------------------

### Q: How did you hear about rOpenSci?

It's hard to say, but I almost certainly first heard about [rOpenSci on
twitter](https://twitter.com/ropensci). I've followed members of the
team on there for a while, particularly because they're part of an
active online community interested in reproducibility in research.
rOpenSci has lots of really useful tools. Packages like
[rgbif](https://github.com/ropensci/rgbif/), which evolved into
[spocc](https://github.com/ropensci/spocc) are part of the core input
that goes into SDM models in Zoon.

------------------------------------------------------------------------

### Q: If you could tell rOpenSci what tool to develop next, what would have the most impact on your research?

Unlike species occurrence data, there's currently no single portal to
access a wide array of different gridded environmental data (the maps of
climatic and environmental conditions that SDM needs to make prediction
maps). There are a couple of different data sources out there, with R
interfaces to some of them, but it would be great to have access to all
of them via a single R API.

------------------------------------------------------------------------

**Want to learn more about Nick Golding and Zoon?**

- Nick will share more details about the Zoon Project, and outline how
  you can get involved, in a Spring 2017 Community Call.
- [Zoon on GitHub](https://github.com/zoonproject/modules)
- Read Nick's papers on [Google
  Scholar](https://scholar.google.ca/citations?user=peoal7wAAAAJ&hl=en).
- Follow Nick on [Twitter](https://twitter.com/_nickgolding_)

------------------------------------------------------------------------

::: {.footnotes role="doc-endnotes"}

------------------------------------------------------------------------

1.  ::: {#fn:1}
    Barbosa, F. G., & Schneck F. (2015). Characteristics of the
    top-cited papers in species distribution predictive models.
    Ecological Modelling, 313: 77-83.
    <https://doi.org/10.1016/j.ecolmodel.2015.06.014> [↩︎](#fnref:1){.footnote-backref
    role="doc-backlink"}
    :::

2.  ::: {#fn:2}
    Pigott DM, Golding N, Mylne A, Huang Z, Henry AJ, Weiss DJ, Brady
    OJ, Kraemer MU, Smith DL, Moyes CL, Bhatt S, Gething PW, Horby PW,
    Bogoch II, Brownstein JS, Mekaru SR, Tatem AJ, Khan K, Hay SI.
    (2014). Mapping the zoonotic niche of Ebola virus disease in Africa.
    Elife, 8(3), e04395.
    <https://doi.org/10.7554/eLife.04395> [↩︎](#fnref:2){.footnote-backref
    role="doc-backlink"}
    :::

3.  ::: {#fn:3}
    Figure 5. Pigott DM, Golding N, Mylne A, Huang Z, Henry AJ, Weiss
    DJ, Brady OJ, Kraemer MU, Smith DL, Moyes CL, Bhatt S, Gething PW,
    Horby PW, Bogoch II, Brownstein JS, Mekaru SR, Tatem AJ, Khan K, Hay
    SI. (2014). Mapping the zoonotic niche of Ebola virus disease in
    Africa. Elife, 8(3), e04395.
    <https://doi.org/10.7554/eLife.04395.009> [↩︎](#fnref:3){.footnote-backref
    role="doc-backlink"}
    :::
:::


rOpenSci’s overarching mission is to promote a culture of transparent, open, and reproducible research across various scientific communities. All of our activities are geared towards lowering barriers to participation, and building a community of practitioners around the world.

Messaggi di Rogue Scholar

Introducing Open Forensic Science in R

Introducing the 2018 rOpenSci Research Fellows!

Community Call - Writing Packages to Support Research Communities - zoon & greta

Announcing the rOpenSci Fellowships Program

Announcing our first fellowship awarded to Dr. Nick Golding