Image
UVA Researchers Introduce SNIFF to Help Scientists Choose Trustworthy Software
Not all software is created equal.
Many medical discoveries, drug development efforts, genomic analyses, and public health studies depend on computational software. But how trustworthy are the tools behind those discoveries?
Researchers at the University of Virginia School of Data Science have developed a practical framework to help scientists answer that question.
In a recent paper published in PLOS Computational Biology, the team offers “Ten quick tips to SNIFF out sustainable and secure scientific software,” a guide to help researchers become more informed consumers of scientific software and better protect the integrity of their research.
"This work spun out of a Research Interest Group supported by the School of Data Science, with a few of us comparing notes on software we inherit rather than software we write ourselves," said Stephen Turner, assistant dean of research at the UVA School of Data Science. "SNIFF is our attempt to turn some hard-won instincts into something a researcher can check before they commit."
Think about downloading a smartphone app. Most people will instinctively ask questions like: Who made it? How many people use it? Is it still being updated? If the app requests unusual permissions or has poor reviews, many people think twice before installing it.
The authors argue scientists should apply that same level of scrutiny before choosing software for research.
"SNIFF encourages researchers be intentional about the software they rely on," said lead author V.P. Nagraj, a Ph.D. data science candidate at UVA. "In data science and other computational fields, the number of available tools is growing rapidly, yet there's far more guidance on building software than on selecting it,” he said. “SNIFF helps identify practical indicators of a tool's sustainability, security, and overall health before you commit your research to it. That kind of careful evaluation can prevent technical debt later."
Scientific software can become unreliable for many reasons. A program may be abandoned after the original graduate student who created it leaves the lab. Software can be poorly documented, no longer compatible with newer computers, or vulnerable to security risks. Some software simply produces unreliable results.
When researchers unknowingly rely on software like this, the consequences can be significant. Flawed or outdated software can lead to irreproducible research, incorrect analyses, wasted funding, or setbacks in scientific progress.
The SNIFF framework helps researchers identify software they can trust, improving reliability and reproducibility of scientific discovery. Ultimately, this benefits patients, researchers, policymakers, and society by making scientific discoveries more trustworthy and reproducible.
SNIFF is an acronym for five areas to evaluate before adopting a piece of scientific software:
- Source – Who created the software, and is it still actively maintained?
- Network – Is it part of a respected scientific community or ecosystem?
- Interaction – Are the developers responsive to users and feedback?
- Fit – Is it the right tool for the research question?
- Fragility – Does it rely on outdated or insecure components that could fail or introduce vulnerabilities?
The framework arrives at an important moment. Science is becoming increasingly computational, while AI-assisted coding tools are dramatically accelerating software development. Tasks that once took days or weeks can now often be completed in hours, making it easier than ever to create new software. At the same time, researchers are under pressure to produce reproducible, transparent results.
The authors hope SNIFF will encourage researchers to think more critically about the software they use and help strengthen the foundation on which future discoveries are made.
The broader takeaway is simple: Trustworthy science begins with trustworthy software.
The paper was published on July 15, 2026, by PLOS Computational Biology and co-authored by V. P. Nagraj, Karsten H. Siller, Thomas Stewart, Neal Magee, and Stephen D. Turner.
