Ten simple rules for non-visual, reproducible and accessible bioinformatics
summary
The gist
Bioinformatics workflows rely heavily on visual representations because quality-control plots, cell embeddings, heatmaps, genome-browser tracks…
In short
The paper proposes ten rules for making bioinformatics workflows non-visual and accessible. It argues that treating plots as decision records, using literate programming, and structuring data with metadata ensures transparency and reproducibility for all users, aligning accessibility with FAIR research principles.
Key concepts
- Decision Records
- Treating plots not as mere illustrations but as structured decision records. This means documenting the analysis's aim, the evidence found in text/tables, the rationale behind choices made (including uncertainty), and what data was inspected. It moves beyond just showing a result to explaining *why* that result matters.
- Literate Programming
- Writing code and explanations together in formats like Quarto or R Markdown. This approach ensures that when a screen reader reads the document, it follows a logical flow where code chunks are small and atomic, allowing one pass through the text to resemble visual inspection of results.
- Accessible Environment
- Setting up computing environments so they inherently support assistive technologies like screen readers or braille displays. This means prioritizing command-line interfaces over graphical ones and using version-controlled dotfiles to ensure consistency across different user setups.
- Structured Data and Metadata
- Organizing data outputs with clear structure (like TSV) and rich metadata. This makes data easily inspectable via shell tools or read aloud row by row, ensuring that the underlying numeric summaries are auditable without needing to rerun complex analyses.
Terminology used across episodes
This episode discusses
The paper
Ten simple rules for non-visual, reproducible and accessible bioinformatics · Read on arXiv
Department of Molecular Life Sciences, University of Zurich · Life Science Zurich Graduate School · SIB Swiss Institute of Bioinformatics
Bioinformatics workflows rely heavily on visual representations. Quality-control plots, cell embeddings, heatmaps, genome-browser tracks, and interactive dashboards are not merely illustrations, but instruments for making analytical decisions. For blind and low-vision researchers who use screen readers, braille displays, or audio-based interfaces, these create a barrier: the evidence used to justify an analysis is often encoded in visual form, while the underlying decision remains undocumented. We argue that non-visual accessibility and computational reproducibility are closely aligned, as they both require analyses to be transparent and to record why decisions were made. We present ten simple rules for non-visual bioinformatics, covering plots as decision records, cautious use of AI-generated figure descriptions, accessible computing environments, text-first literate programming, structured data and metadata, compact object summaries, accessible publication formats, collaboration practices, shared community infrastructure, and accessibility as part of FAIR research. The intended audience is computational biologists and developers. Using single-cell RNA-seq as a running example, we show that the accessible equivalent of a plot is a structured decision record. That is, a plot companion that goes beyond storing the underlying data by also stating the purpose of the analysis and the resulting quantitative evidence and uncertainty. We argue that treating accessibility in this way makes bioinformatics more inclusive and also more transparent and auditable.
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: I'm Ines.
Marcus: And I'm Marcus, here with Yuki.
Ines: We're looking at the paper: "Ten simple rules for non-visual, reproducible and accessible bioinformatics".
Marcus: Here's the gist of the paper. Bioinformatics workflows rely heavily on visual representations because quality-control plots, cell embeddings, heatmaps, genome-browser tracks, and interactive dashboards are not merely illustrations but instruments for making analytical decisions The evidence used to justify an analysis is often encoded in visual form while the underlying decision remains undocumented The Core Argument The paper argues that non-visual accessibility and computational reproducibility are closely aligned because both require analyses to be transparent and to record why decisions were made They present ten simple rules for non-visual bioinformatics covering plots as decision records, cautious use of AI-generated figure descriptions, accessible computing environments, text-first literate programming, structured data and metadata, compact object summaries, accessible publication formats, collaboration practices, shared community infrastructure and accessibility as part of FAIR research They show that the accessible equivalent of a plot is a structured decision record which goes beyond storing the underlying data by also stating the purpose of the analysis and the resulting quantitative evidence and uncertainty Structuring Decisions as Records Rule 1 mandates treating plots as decision records, not illustrations, where supporting data should be closer to decision records, keeping track of the purpose of the diagnostic, identifying the data being inspected and reporting numerical page 2 A decision record structures information at several levels including stating the analysis aim, reporting evidence in text and tables, recording the decision and its rationale and preserving ambiguity or uncertainty For instance, it enriches statements such as “The UMAP looked good after integration” with “After integration, donor-associated separation decreased while known cell-type marker structure was preserved.
Ines: No major cluster was dominated by a single donor.
Marcus: The selected integration was retained for downstream clustering, but small donor-specific differences in cluster 7 should be rechecked during annotation” Establishing an Accessible Workflow Rule 3 advises setting up an accessible environment first, where the operating system should already manage a screen reader or braille display predictably Level 1 barriers include tools with a graphical interface only and terminals that expose text differently on each operating system Facilitators at this level include screen readers and braille displays, Emacs with Emacspeak, version-controlled dotfiles, and a command line in place of a web interface Rule 4 promotes literate programming for reproducibility by placing explanations and code in the same source file which after rendering also contains results This works better when code chunks are small and atomic with one intention per chunk and using formats with plain text source such as Quarto markdown or R markdown Data Representation and Publication Rule 5 calls for structuring data and metadata for non-visual exploration, suggesting that command line interface (CLI) tools can reduce the burden encountered on browsing web portals by making structured data queries easier Rule 6 facilitates data inspection and fingerprinting by suggesting summaries in R and Python designed for visual scanning, including properties like in-memory size and count of missing values for object fingerprinting Rule 7 recommends publishing in accessible formats, suggesting that reports should be generated as a tagged PDF/UA-1 file validated by veraPDF or an HTML page checked by axe against WCAG 2.
Title and authors: Marcus: We’re diving into this paper today, "Ten simple rules for non-visual, reproducible and accessible bioinformatics," which really focuses on how we do the actual analysis work day-to-day, not just the data portals. We'll start by talking about who wrote this and what they’re trying to fix.
Ines: The authors are Jacqueline Kientsch, Stephan Neuhauss, and Izaskun Mallona from Zurich. They come from a background where they use screen readers and braille displays to navigate code and literature.
Yuki: That lived experience is key because it shows why these rules matter—it’s not just an abstract technical problem; it’s about real researchers facing real barriers every day.
Marcus: Right. They point out that the problem isn't just with big data portals, but with the day-to-day practice of computational analysis itself, like inspecting data and justifying thresholds.
Ines: They set up ten specific rules to bridge that gap between visual output and a justifiable decision record. These aren't suggestions; they are guidelines for making the process accessible to everyone who doesn't work visually.
Yuki: So, if we look at this paper, it’s essentially proposing a framework for making bioinformatics transparent by structuring how we record our reasoning alongside the data.
Marcus: That’s right. It moves beyond just storing the underlying data and focuses on stating the purpose of the analysis and reporting that resulting quantitative evidence and uncertainty.
The paper's summary: Ines: So, what do these ten rules actually suggest? They start by treating plots as decision records instead of just illustrations, where you keep track of why you made a diagnostic choice.
Marcus: That means if you have a UMAP plot that looks good after integration, you don't just write "the UMAP looked good"; you need to add the purpose of the diagnostic and the resulting quantitative evidence.
Yuki: For example, instead of just saying "the donor separation decreased," you’d need to explain *why* that matters biologically and report what that change implies for downstream clustering.
Ines: That's a good point. And they talk about establishing an accessible workflow right from the start, like setting up an accessible computing environment before you even write your code.
Marcus: They list Level one barriers, like tools with only graphical interfaces or terminals that behave differently depending on your operating system, and then suggest facilitators like screen readers or Emacs with Emacspeak <ref:2608.14400#pg1>.
Yuki: I wonder if that makes sense for someone who is used to a standard desktop setup; moving to a command line as the primary interface sounds like a big shift.
Ines: It does, but the point is that it removes friction for assistive technology users, and they pair that with literate programming so explanations and code live in the same file.
Marcus: So when you render that document, it should resemble a visual inspection through a screen reader pass, which they suggest works best when code chunks are small and atomic.
The paper's improvements: Ines: Moving into the more advanced stuff, the paper talks about improving how we deal with AI descriptions and how we structure our data for non-visual exploration.
Marcus: They caution against taking AI descriptions at face value; they suggest cross-checking those textual outputs from different models against what colleagues assess to catch things like hallucinated labels or misread axis ranges.
Yuki: That sounds practical because AI can be really confident about trends that just don't hold up when you look closely at the underlying data structure.
Ines: Exactly. And for data itself, they push for structuring metadata in formats like TSV so that comma-separated values are easy to inspect using shell tools and can be read aloud row by row.
Marcus: That addresses a major barrier where output meaning depends entirely on column alignment, which is a huge pain point for anyone not looking at the data visually.
Yuki: I’m curious about their idea for object fingerprints, suggesting summaries in R and Python that include things like the in-memory size and the count of missing values.
Ines: That replaces those dense columnar outputs where you have to align everything manually; it gives you a quick snapshot of the object's characteristics right away.
Marcus: They also recommend publishing using accessible formats, like PDF/UA-one files validated by tools like veraPDF, or HTML pages checked by axe against WCAG two point two standards for logical reading order and structural tags <ref:2608.14400#pg1>.
Conclusion: Ines: So wrapping this up on "Ten simple rules for non-visual, reproducible and accessible bioinformatics," the main thrust is making analysis transparent by treating plots as decision records, building accessible environments first, and using literate programming to keep code and explanation together.
Marcus: The authors argue that when you structure your data with metadata in formats like TSV and use summaries for fingerprinting, you make the process much easier for people who rely on screen readers or other assistive technologies.
Yuki: It’s about ensuring that the same plain-text outputs built for non-visual use also serve every reader, meaning tables and logs are diffable so you can audit the numeric summaries without rerunning everything.
Ines: That auditing capability is crucial because it allows you to question a decision based on the evidence right there in the text, not just looking back at a picture.
Marcus: The paper gives us a very concrete set of actions—from choosing command lines over web interfaces to validating your final PDF output—to build that chain of trust.
Yuki: It’s about making bioinformatics FAIR by embedding accessibility directly into the workflow rather than treating it as an afterthought.
Ines: That’s the essence of it, ensuring that the decisions made in complex analyses are traceable and justifiable to anyone who needs to understand them.
More episodes
- 2610.33881-Sensorimotor behavior from an untrained brain-and-nerve-cord connectome
- 2610.40344-Morphological Features of Murine Pulmonary Arterial Networks for Diagnosing Vascular Disease
- Daily Summary for 2026-11-19
- 2609.10193-Are You Learning Biological Signal or Shortcuts? Auditing and Mitigating Bias in Protein-Protein Interaction Datasets
- 2610.38236-Generative discovery of chemical reaction networks
- 2507.11176-An Interpretable AI framework Quantifying Traditional Chinese Medicine Principles Towards Enhancing and Integrating with Modern Biomedicine
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- 2504.03590-Towards a Unified Framework for Determining Conformational Ensembles of Disordered Proteins
- 2603.15217-A multiscale discrete-to-continuum framework for structured population models