Magnifying impact through data sharing

An interview with Dr. Kenneth R. Arnold

For Dr. Kenneth R. Arnold, data sharing is a way to extend the reach and impact of a scientific study.

“The scientific value of a dataset almost always exceeds what any single paper can extract from it,” Arnold said, pointing to a recent investigation as an example.

Reduced genetic diversity doesn’t necessarily mean traits are lost

Arnold is an evolutionary biologist and postdoctoral scholar in the Department of Integrative Biology at Oregon State University.

He studies how populations adapt, using experimental evolution in fruit flies. To do that, researchers impose a consistent selection pressure on laboratory populations of flies, maintain them for many generations, and then measure what changed.

This lets researchers test ideas that are otherwise difficult to evaluate directly, he explained. Evolutionary theory makes clear predictions about how genetic variation should behave under sustained selection. Experimental evolution lets researchers check those predictions against a real population with a known history. The work incorporates a wide range of data types, from time-series genomics and metabolomic profiles, mortality and fecundity records, and developmental timing assays.

In a recent investigation Arnold worked with fruit fly populations selected based on reproduction life stage, with some groups reproducing early in life and others late in life. Many had been maintained that way for more than a thousand generations.

Arnold and colleagues tested the classic theory that intense constant selection exhausts genetic variation, leaving populations unable to respond to new circumstances. Researchers swapped regimes for two fruit fly populations selected for early- and late-stage reproduction and observed their response.

The results, reported in Molecular Biology and Evolution and the dataset “Data from: Beyond fixation: Persistent genetic variation under intense selection,” showed that both populations ‘converged rapidly on the traits and the genetic composition characteristic of the regime they had been moved into,” Arnold said. “Sequencing a founder population at very high depth revealed that many sites that appeared to have no variation left at standard sequencing depth in fact carried rare variants at frequencies too low to detect.”

In conservation genomics, reduced genetic diversity in a threatened population is often read as irreversible loss of adaptive potential. These results suggest that may not be true.

Figure 1
Waves of convergence in selection. Differences in mean instantaneous age-specific mortality rate between population groups, plotted as a function of age from egg (days). Each point represents the mean difference across replicate populations at a given age interval; error bars indicate standard error across replicate populations. Lines show smoothed trends. The purple series in each panel shows the baseline mortality difference between the long-established A- and C-type founder populations, against which trajectory convergence is evaluated. a) C→A trajectory: differences between C→A populations and A-type founder populations after ∼33 (magenta) and ∼143 (red) generations of early-reproduction selection. b) A→C trajectory: differences between A→C populations and C-type founder populations after ∼12 (light blue) and ∼51 (dark blue) generations of delayed-reproduction selection.

A place for everything

The dataset comprises around 1.4 million variable sites, tracked across 40 populations, plus an unusually deeply sequenced sample and years of accompanying lifespan and development data. The researchers answered their own question, “but the data may hold answers to a great many questions we did not ask,” said Arnold.

It may prove useful for researchers looking for candidate genes and shared characteristics with other organisms, and those studying aging, developmental timing, or life-history trade-offs.

“The greatest long-term value, however, may come from data aggregation,” he said. “Open data lower barriers for researchers at institutions without the resources to generate large datasets for themselves. Synthesis across studies is often where the most general conclusions come from.”

In addition to the potential future value of the public dataset, the work was supported by the National Institutes of Health, which mandates data sharing. There’s a moral imperative as well. “When public funds support research, the products of that research are held in trust for the public rather than owned by the investigators who generated them,” said Arnold.

The study generated a huge amount of data. Raw sequence data went to the NCBI Sequence Read Archive. Analysis scripts were posted to GitHub.

“But raw reads alone are far from accessible,” said Arnold.

So, the team deposited processed datasets on Dryad, including variant tables, the phenotypic measurements, and the supporting files that connect the sequence data to the published figures. “That left each component in the place best suited to it, to make our data accessible at multiple levels.”

Dryad was a natural choice for those resources. It’s “well established in evolutionary biology and molecular evolution, to the point where it is often the first place we look when we want the data behind a paper.”

In data publication, consistency is everything

For Arnold, deciding to share the data was the easy part. The data publishing and publication process was a little less straightforward. It “definitively highlighted the importance of consistent labeling,“ Arnold recalled.

With eleven authors across four institutions and several years of data collection, preparing the data for sharing was surprisingly challenging.

“Each of us had developed our own shorthand for many of the same measurements. What one person recorded as “development rate” another had recorded as “pupariation timing,” and reviewers rightly pointed out that these are not interchangeable.”

Similar ambiguities had accumulated around other data points as well. Reconciling it all meant establishing definitions, updating labels, and creating documentation across all the data files — a tedious process at times, but necessary for the data to be interpretable by any future reader.

Reconciling these types of anomalies is a particular specialty of Dryad’s. Our team of professional data publishing experts review each dataset and work one-on-one with authors to prepare it for publication. That expertise and personal care ensures the data we publish meets all requirements, and makes our datasets the most discoverable and impactful around.

But authors can get an early start on the process, beginning with a data management plan and good recording keeping during data collection.

Arnold’s advice for other researchers: “Start earlier than what would seem necessary… Write documentation for someone who has read the paper and nothing else, explaining every column and label,” and “practice good filename hygiene. You never want to have to choose between Data_final.txt and Data_final_REAL.txt!”

“We tend to take for granted how much scientific progress rests on publicly released packages and pipelines, but almost every step of our analysis depended on tools that someone else built and shared freely.”

Share your data story. Have you shared open data or re-used others’ publicly-available data? Share your story with us at partnerships [at] datadryad.org.

Feedback and questions are always welcome, to hello@datadryad.org. 

To keep in touch with the latest updates from Dryad, follow us on LinkedIn, Mastodon, and Bluesky and subscribe to our quarterly newsletter.

NOTE: This blog is based on a case study prepared as part of Dryad’s work with the NIH GREI working group.