The AI Rewrite Dilemma

There have been multiple ongoing efforts (see Appendix below), and debates, regarding porting bioinformatics tools to Rust. These include two rewrites of minimap2. Here are some of my current reflections, which may change in the future.

From a purely technical standpoint, I believe AI rewrites, especially in Rust, benefit the field. Rust will not only speed up bioinformatics tools in Python and R often by an order of magnitude, but will also greatly simplify deployment. With Rust (and a few other modern languages), non-standard dependencies are compiled into the final binary. This eliminates the need to install dependencies at runtime, freeing us from dependency hell. Bioinformatics would be a more enjoyable field if more tools were written in Rust.

Rust is notorious for its steep learning curve. It has historically been unrealistic to ask biologists to develop tools in Rust. However, coding agents such as Claude Code and Codex have transformed code translation. Since last December, a developer familiar with these agents could port a small tool in hours or one of medium complexity (e.g. minimap2) in days, probably tens of times faster than porting by hand. This is happening and is likely to accelerate.

Legally speaking, permission for rewriting is governed by the software license. Maintainers of open-source projects can’t stop others from rewriting their work, no matter whether they like the action or not. One defensive strategy is to fight fire with fire: preemptively rewrite your own tool in Rust by yourself such that another Rust rewrite by a third party would be of little added value. I don’t necessarily like the idea but I could not think of a more effective solution from the maintainer perspective.

For tool rewriters, please be aware that not all maintainers are comfortable with AI rewrites of their tools. The long-term maintenance is also the elephant in the room that needs to be addressed adequately. I haven’t rewritten any tools. Without direct experience, I am not in a position to make suggestions. I just hope the bioinformatics community could collaboratively come up with guidelines and common practices on tool rewriting to advance this field in a friendly manner. After all, there are humans behind the code. Empathy matters more than technical benefit or legal correctness.

Finally, I always welcome forks and rewrites of my tools whether I am involved or not. If you want to rewrite my tools with AI, please feel free to do so.

Appendix: some ongoing rewrites as of 2026-04-17

• pachterlab/edgePython in Python: based on edgeR in R, explained in this blog post
• seqeralabs/RustQC in Rust: based on multiple tools in C, R and Python, explained in this blog post
• fulcrumgenomics/fgumi in Rust: based on part of fgbio in Scala, explained in this blog post
• Huang-lab/fastVEP in Rust: based on VEP in Perl and Nirvana in C# (unmaintained). Preprint at bioRxiv
• COMBINE-lab/sshash-rs in Rust: based on sshash in C++
• huangnengCSU/rust-fastqtl in Rust: based on a fork of fastqtl in C++ (unmaintained)
• Huang-lab/RastQC in Rust: drop-in replacement of FastQC in Java
• henriksson-lab: multiple projects including minimap2 in C, BLAST in C++, HMMER in C, MUSCLE in C++, DIAMOND in C++, SKESA in C++, Kraken2 in C++, FastQC in Java, Prokka in Perl and more including non-bioinformatics tools and libraries.