Being able to access publicly available code linked to a paper or project, and understand the logic well enough to extend it, has accelerated how I engage with research, making reproducibility and rigor easier to ensure, and iteration and sharing across the research community easier too. For work where software is required, that’s when I reach for open-source software first.

In my experience, open-source software lets me act on a task fast, whether that’s predicting structures, making a figure with Inkscape, running quick analysis through Jalview, analyzing a western blot through Fiji, and the list goes on. There’s no wait time beyond the download itself, and beyond anything monetary, one of the best things about open-source software is the long-term continuity and accessibility of projects. I can install the same software on multiple machines and know a script, data file, image, or analysis can be opened and reproduced. I won’t lose access to my own work just because I’ve moved or changed computers.

So what exactly constitutes open source? It distills down to two things: publicly available code and a license that grants everyone the right to use, modify, and redistribute that code. Because of this, the range of usage can be wide, from simply using software as-is to modifying it in creative and positive ways. PyMOL is a good example of this in practice, as there’s both a paid commercial license and an open-source version, and the open version doesn’t prevent researchers from using the core features of 3D structural analysis. That openness is exactly what enabled creative projects like RayMol, which brings 3D structure visualization to a phone, among other things. Additionally, being able to see and modify the underlying code often means community members can contribute directly to documentation and troubleshooting resources.

In the era of computational structure prediction, many models have also published full algorithms and weights, allowing people to build on and improve predictions. This shows up with AlphaFold2, ESMFold, and the many specialized versions spun off from them, where each iteration builds from the last, and is exactly why open-source, and open weights specifically, are so valued here.

So again, as someone who reaches for these programs, I wanted to create an easy way to catalog and install the open-source software that is often used in structural biology. The leader for such software access is SBGrid, a software consortium that labs can subscribe to. However, when SBGrid isn’t an option, that’s the gap this project fills. This is also a deliberately narrow project, only to download genuinely open-source structural biology tools - not a reproducibility framework, a container system, or a replacement for institutional infrastructure.

“Free for academic use” does not equal “open-source”

This catalog is also deliberately strict: everything in it carries an OSI-approved or equivalent open-source license (e.g. MIT, BSD, GPL, Apache-2.0).

“Free for academic/nonprofit use, paid license for commercial use” is categorically different, no matter how generous the nonprofit terms are or how common the tool is in practice. This roughly amounts to hundreds of tools that fit into the free for academic use category versus several dozen for genuinely open-source at the time of writing. Several tools which I currently or in the past have used are not in this project’s catalog by that principle.

Is this prohibitive? Probably not at the academic/nonprofit level, and it’s likely not prohibitive for for-profit groups with the budget for it either. Nevertheless, the goal here is to make it easier to find and download open-source structural biology software, which makes outlining this caveat important. So if there’s something seemingly missing, that’s the reason why.

Hardware is an important consideration

In setting up the software catalog, one thing that became more evident is the distinction between “open-source” and “accessible.” A tool can be strictly open-source and still practically unusable without workstation or cluster support. Crystallography and NMR software is largely built to run on a desktop, but structure prediction using deep learning requires additional hardware.

GPU-required vs laptop-friendly

So hardware is an important factor, and when choosing what to install, I’ve created annotations to highlight considerations: ⚡ = GPU-beneficial (might be slow otherwise), ⚡⚡ = GPU-required.

A note on testing: Several of the CPU-friendly software below I have installed and verified, but that hasn’t been done exhaustively across every single entry in this catalog. So, there may be an install quirk, a missing dependency, or a stale package name here or there. This should still get very close to a working setup though.

---

Crystallography

Almost entirely CPU-friendly. Core algorithms: integration, refinement, density fitting.

• Coot, Model building/validation. There are macOS headaches though! Check out BANDICOOT for a solution.
• DIALS, Diffraction integration
• Servalcat, Refinement
• CCTBX, Core crystallographic algorithm library
• Gemmi, File format library
• DSSP, Secondary structure assignment

Cryo-EM

• Scipion, Coordinates other tools (RELION, EMAN2, etc.)
• ⚡ RELION
• ⚡ EMAN2
• ⚡⚡ cryoDRGN, Neural network-based reconstruction

NMR

The most laptop-friendly category in the whole catalog

• nmrglue
• ssNake
• NMRFx Processor
• COLMARvista, CPU (browser app)

Molecular dynamics

Small systems and short trajectories are fine on CPU but anything else benefits from a GPU.

• ⚡ GROMACS
• ⚡ OpenMM
• AmberTools, Utilities (tleap, cpptraj, antechamber)
• MDAnalysis, Trajectory analysis, doesn’t run simulations itself

Structure & interaction prediction

With deep learning hardware, GPUs are a must. When that isn’t an option, ColabFold (hosted by Google Colab notebook/free-tier GPU) is an alternative. Check out the repository of a specific model - there will likely be a ColabFold option.

• ⚡ ESMFold, Can run on CPU, but GPU is the path for real work
• ⚡⚡ Protenix
• ⚡⚡ Boltz-2
• ⚡⚡ AlphaFold2
• ⚡⚡ OpenFold / OpenFold3
• ⚡⚡ ColabFold (local install)
  • There is also a browser alternative

Docking & cheminformatics

Classical docking and cheminformatics, and mainly CPU-friendly

• AutoDock Vina
• AutoDock4
• Open Babel
• RDKit
• PLIP

Sequence analysis

CPU-friendly

• HMMER, CPU (multi-core helps)
• HH-suite, CPU (benefits from RAM/cores)
• MAFFT
• Clustal Omega
• MMseqs2

Visualization & general libraries

CPU-friendly

• PyMOL
• OpenStructure
• Jmol, NGLView
• Biopython, Biotite, ProDy, ssbio, pdb-tools, FreeSASA, all CPU libraries/utilities

---

How does it work?

Each tool installs into its own isolated environment so nothing conflicts. The first time a tool is installed, a small shortcut gets added to PATH that points to it, so the command runs automatically.

Most installs are through conda (conda-forge and bioconda), while the ML-based structure prediction models like Boltz-2 and Protenix, install via pip, in which case the catalog pins to a known, reviewed version tied to its official PyPI hash.

Within the repository, list shows every tool in the catalog, its license status, and how it gets installed. Both installing and removing tools is intended to be easy with calls like install [tool], remove [tool], remove --all, but there are some exceptions that need more involved setup.

Full details and future updates are in the Github repository README.