Skip to content

Getting started

CRAIC runs on macOS, Linux, and Windows. It needs Python 3.9+. The Rust acceleration core is optional — without it CRAIC falls back to an identical NumPy implementation, just slower.

Install and launch

cd path/to/craic
./build_rust.sh      # makes a .venv, installs deps, builds the Rust core if Rust is present
./craic.command      # or:  .venv/bin/python -m craic

python3 -m venv .venv
.venv/bin/python -m pip install numpy biopython PySide6
.venv/bin/python -m craic examples/coding_genes.fasta

macOS, Homebrew, and conda

If your python3 is Homebrew's, a system-wide pip install fails with externally-managed-environment (PEP 668). The virtualenv above avoids that, and build_rust.sh also unsets CONDA_PREFIX for its own run so maturin won't refuse to build inside a conda shell.

A double-clickable app (macOS)

Launched with craic or python -m craic, macOS shows the Python executable (e.g. python3.13) in the Dock and app switcher, because a bare script isn't an application bundle. To get a proper CRAIC name and icon everywhere, and a Finder/Dock launcher, build a small .app:

craic make-app        # writes CRAIC.app into ~/Applications

It wires the app to the Python you ran it with, so run it from the environment that has CRAIC installed (e.g. .venv/bin/craic make-app). Re-run it if you move or delete that environment. Because it is built on your Mac rather than downloaded, macOS doesn't block it. See craic make-app.

The status bar shows core: rust once the native core is built, otherwise core: numpy.

Your first alignment

  1. Open a file (FASTA, PHYLIP, Clustal, Stockholm, or NEXUS) — or load the bundled examples/coding_genes.fasta.
  2. Choose an engine and press Align. Tick align as protein for protein-coding nucleotides — CRAIC translates, aligns the amino acids, and threads the nucleotides back through, keeping the reading frame intact and switching to the protein view.
  3. Toggle View between nucleotide, codon, and amino acid.
  4. Use the Track dropdown to overlay a per-column statistic (Conservation, Reliability, Aligner agreement, …); it computes whatever it needs on demand the first time you pick it.
  5. Click-drag a column range to drive the realignment sandbox and the posterior explorer; single-click a residue to probe where else it could align.

Then try it with the answer known

The fastest way to understand what the reliability score is claiming is to check it against a case where the right answer exists. Teach → Generate dataset with known answer… evolves sequences down a random tree and loads them unaligned. Align them, then compare the Reliability track — a prediction — against the Reference correctness track, which is the outcome. Colour by Correct placement to see which individual residues went wrong, and open the Column inspector to read what the right answer was.

See Teaching with a known answer.

Optional: external aligners

CRAIC auto-detects MAFFT, MUSCLE, Clustal Omega, ProbCons, PRANK and ClustalW on your PATH and offers them in the engine menu and the disagreement comparison. None are required — the built-in progressive aligner always works.