I worked with Dr. Jason Sahl, Associate Professor of Biology at Northern Arizona University and Assistant Director of its Pathogen and Microbiome Institute, on NASP, a genomic analysis tool he published as part of his federally funded pathogen-detection research. Researchers use it to compare bacterial genomes, for example to tell how closely related outbreak samples are. Its install package had been broken for years. I found out why, fixed it, and got the fixes shipped to bioconda, the main software channel for bioinformatics.
Got the package building and installing again, and shipped the first new NASP builds since August 2020.
View on GitHub → In review · all checks passingUpdates NASP's code so it runs on current Python and current samtools, and pins the Java version GATK needs.
View on GitHub →Running conda install nasp finished without a single error. Every quick check I could run without actually using the tool also passed. But the first time NASP tried to process an alignment, it crashed.
Every alignment in NASP passes through a tool called samtools, and the samtools version the package pinned couldn't even start. So the pipeline died before it processed a single read, and nobody had been able to publish a fixed build since 2020.
The lesson that shaped the whole project: a dry run only checks that packages fit together. It never runs them. From then on I tested by actually running things.
$ conda install nasp done $ samtools --version samtools: error while loading shared libraries: libncurses.so.5: cannot open shared object file: No such file or directory $ ldd $(which samtools) | grep ncurses libncurses.so.5 => not found
The recipe pinned samtools below 1.3, which picked version 1.2. That build needs a system library (ncurses) but never declared it, so nothing installed it and the program couldn't load.
Fix: pin samtools 1.1 instead. Its build declares the library correctly and still understands NASP's commands. I tested this directly against four samtools versions.
The build script overwrote NASP's Go module file with a one-line version that dropped a dependency. Older Go versions quietly fixed that. Go 1.16 made it a hard error, so the build broke without anyone changing NASP.
Fix: delete that one line.
When that build failure hit bioconda's 2021 mass rebuild, NASP was added to a list of recipes the system stops trying to build. That's why the failure stayed invisible for years.
Fix: remove NASP from the blacklist.
NASP found its own bundled files with pkg_resources, which setuptools deleted. I pinned down exactly when: version 81.0.0 has it, 82.0.0 doesn't.
Fix: a temporary version pin, then a real code fix using Python's built-in importlib.resources at all 4 places NASP used it.
NASP sorts alignments with an old samtools command that newer versions reject. The obvious replacement looked fine, but when I tested it on samtools 1.1 it wrote no output file and still reported success. A small extra flag (-T) makes it work on 1.1, 1.9, and 1.24. This is why I tested every fix instead of trusting it.
Some problems aren't packaging bugs. Fixing them would change which SNPs NASP reports, so they're research decisions for the lab, not for me.
Its 2013-era command fails on every bcftools version still packaged. Rewriting it would change results, so with Dr. Sahl's approval I removed it from the menu instead.
NASP's caller doesn't exist in GATK 4. Its replacement calls SNPs differently, so switching would need scientific revalidation.
It looks compatible on paper, but its defaults changed. MUMmer 3.23 works, so I left it rather than guess.
I worked directly with Dr. Sahl from start to finish. I committed to a deadline, delivered a full written diagnosis, brought him the decisions that affected the science, and shipped the fixes once he approved them.
I set the rules for the project, made the scope and shipping decisions, and opened both pull requests. One is already merged into bioconda, the main software channel for bioinformatics.