1 of 25

Openscapes for better science in less time, together

Julia Stewart Lowndes, PhD

Openscapes Founder & Co-Director Senior Fellow, National Center for Ecological Analysis and Synthesis (NCEAS), UC Santa Barbara

lowndes@nceas.ucsb.edu; @juliesquid

Erin Robinson, MSc

Openscapes Co-Director

Co-Founder, Metadata Game Changers

erin@metadatagamechangers.com;

@connector_erin

Slides: openscapes.org/media

Artwork by Allison Horst

CZI Seed Networks Computational Biology Meeting – April 6, 2021

2 of 25

Openscapes helps research groups reimagine data analysis, develop modern skills that are of immediate value to them,

and cultivate collaborative and inclusive research communities.

Engage • Empower • Amplify

openscapes.org

@openscapes

Supercharge your research (Lowndes et al. 2019, Nature)

Open software means kinder science (Lowndes 2019, Scientific American)

3 of 25

“I didn’t think to find solutions, because I didn’t realize that there was a problem; I thought that this was just the nature of doing [ecological] research – an endless pile of confusingly named csv files and R scripts.”

Openscapes faculty alumni, 2021

Now that I can see all the ways that our lab has profited from this initial investment, I cannot imagine how I managed without Openscapes’ principles. Our lab is producing substantially more science of higher quality and we are making more of our underlying data and code publicly available as a direct result of [Openscapes’] intervention.

4 of 25

Data-intensive science can be really lonely and demoralizing. Like Luke Skywalker here, we’ve all stared at challenges that we can’t solve with the skillsets we have. Despite all of his skills, abilities, and past successes, his experiences have not set him to deal with this effectively. If he relied on whatever pulleys, rope, or duct tape he had on hand, it wouldn’t be something he’d be proud of or could reuse and learn from, and it wouldn’t have gotten him to where he needed to get to on time.

----

We might not even realize that we are facing problems that we shouldn’t have to tackle with whatever we have on hand.

This can be trying to manage csv and R scripts, or trying to understand how ML applies to our work, or trying to run someone else’s code.

We’ve all felt like this when it comes to data analysis. When you are lonely, demoralized, staring at a challenge that we can’t solve with the skillsets we have.

We like to ask people: what is their Luke Skywalker moment? for me it was realizing that I couldn’t use Excel for my PhD thesis because my data files were too big to even open.

---

Data-intensive research can be really lonely & demoralizing. Like Luke Skywalker here, we all have moments where we are staring at a challenge that we can’t solve with the skillsets we have.

I’ve had 2 big Luke Skywalker moments like this: the first was in grad school when I discovered my data were too big to open in Excel and had to learn to code in a total panic, and the second was with my research team with the Ocean Health Index at UCSB when we realized all our personal, homegrown approaches to coding and data management meant that we couldn’t collaborate easily or reproduce our own work. It was demoralizing wake-up call that forwarded emails with attachments like “data_analysis_final_finalv3.R” was not reproducible research.

5 of 25

6 of 25

7 of 25

We learned the hard way we couldn’t reproduce our own work
But not just about reproducibility –

Difficult to iterate or build from our own work (so much reinventing!)
Difficult to onboard new people and offboard (so much knowledge lost when people leave)

Our path to better science in less time using open data science tools

(Lowndes et al. 2017, Nature Ecology & Evolution)

We published our story as Our path to better science in less time using open data science tools. This figure shows our Yoda moment, which happened over years, as we introduced tooling and practices to address our issues of repeatability, collaboration, etc, and how the R programming community helped helped us reimagine what was possible.

This story still resonates with researchers because we not only shared about the tools but we talked about how hard this transition was.

It was hard because we had to learn not only new skillsets like switching to R and GitHub, but new and mindsets like letting go of our personally crafted systems and sharing imperfect work with each other. This meant changing the way we had always worked for our whole careers, which was very intimidating.

But it was possible because we trusted each other as a team and we had strong vertical leadership from our PI and horizontal leadership within our team.

We’ve had enduring benefits to this approach well beyond efficiency in data analysis

we’ve been able to reimagine public engagement because R has made communication part of our analytical workflow. And we’ve reimagined collaboration and learning through this as well.

[animate] What I became focused on after this was published was how to help operationalize this and scale this approach so that other researchers could also do better science in less time.

Transition: So that question was the key motivator for Openscapes, which I was able to build through a fellowship from Mozilla.

8 of 25

Reproducibility

Data preparation
Modelling
Version control
Organization

Communication outside the project

Sharing data and code
Sharing methods and instruction

Our path to better science in less time using open data science tools

(Lowndes et al. 2017, Nature Ecology & Evolution)

Collaboration within our team

Coding collaboratively
Our (simplified) GitHub workflow
Internal communication

9 of 25

Our path to better science in less time using open data science tools

(Lowndes et al. 2017, Nature Ecology & Evolution)

connection to broader communities

listen
learn
contribute

open coding language

shared practices
streamlined tools

collaboration platform

bookkeeping
display, publication, & distribution

docs, slides, sheets

coordinate
co-develop
share

Taught by the community, practiced as a team

10 of 25

We learned the hard way we couldn’t reproduce our own work
But not just about reproducibility –

Difficult to iterate or build from our own work (so much reinventing!)
Difficult to onboard new people and offboard (so much knowledge lost when people leave)

Our path to better science in less time using open data science tools

(Lowndes et al. 2017, Nature Ecology & Evolution)

Critical part of the OHI story: teamwork & PI support.

Open data science transition was rooted in trust.

>> Key design elements for Openscapes