Carbohydrate Active eNzyme Domain analYsis tool (CANDy) - automated analysis of domain architectures in carbohydrate-active enzymes

CANDy is a fast, FAIR and seamless protein domain analysis tool for any CAZy family.

CANDy is available as an installable Python package (CLI + Python API), replacing the original Google Colab / Jupyter Notebook implementation. The previous notebook (archive/CANDy v2.0.ipynb) is kept in this repository for reference, and the original is on Google Colab.

pip install candy-cazyme

That's it for most users -- CANDy's default toolchain is fully bundled:

• Clustering: MMseqs2 -- auto-downloaded and cached on first use (no conda needed). On Linux/macOS this just works. On Windows, MMseqs2's clustering workflows internally need a POSIX shell; the official Windows build handles this itself by installing a small helper (busybox) the first time it runs, which may ask for administrator permission once -- never again after that. (This mirrors upstream: MMseqs2's own docs list WSL as the recommended Windows path and this static build as the fallback for anyone who can't use WSL.)
• MSA: FAMSA via pyfamsa -- a real pip dependency, runs in-process, no download needed.
• Phylogenetics: VeryFastTree via veryfasttree -- also a real pip dependency, no download needed. Except on Apple Silicon Macs: veryfasttree has no macOS arm64 wheel at all (as of 4.0.4.1) and its from-source build fails on stock macOS (an upstream OpenMP-detection bug). CANDy detects this automatically and defaults --tree-tool to FastTree instead there -- see below.

There are two distinct, independent issues here -- you may hit either, both, or neither depending on your setup:

1. --tree crashes with illegal hardware instruction, no traceback. This means your Python itself is x86_64 running under Rosetta 2 translation instead of native arm64 -- FAMSA ships native SIMD code, and Rosetta's emulation of some CPU instructions is a known cause of exactly this crash (it can also make other native calls, e.g. Gemini curation, unreliable). CANDy logs a warning about this at startup if detected, but pip can't fix it for you: by the time pip install runs, the interpreter architecture is already fixed.

Step 0, always do this first: confirm your terminal itself is native, not just your hardware:

arch   # must print "arm64", not "i386"

If it prints i386, your terminal app (Terminal/iTerm) is launching under Rosetta -- and anything installed from it, including tools like uv that are supposed to auto-detect the native architecture, will get fooled into installing x86_64 builds too. Fix this first: quit the app, Finder > select it > Cmd+I > uncheck "Open using Rosetta" > relaunch, then re-run arch to confirm.

Once arch says arm64, two ways to get a correct native Python:

Option A -- uv (recommended): uv manages isolated Python installs and defaults to the native architecture -- once its own install wasn't done under a translated shell (see Step 0).

curl -LsSf https://astral.sh/uv/install.sh | sh   # only if you don't already have uv
uv python install 3.12
uv tool install --python 3.12 "candy-cazyme[gemini]"

If you already ran uv tool install before fixing Step 0, it will have cached an x86_64 environment -- force it to redo the install natively:

uv tool uninstall candy-cazyme
uv tool install --force --python 3.12 "candy-cazyme[gemini]"

This gives you a candy command backed by its own isolated, native-arm64 Python -- no venv/PATH management needed.

Option B -- Homebrew, manually:

# Install (or confirm) Homebrew at the Apple Silicon prefix, /opt/homebrew
# (a pre-existing Homebrew at /usr/local is the Intel-only one):
/opt/homebrew/bin/brew --version || arch -arm64 /bin/bash -c \
  "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install Python from that prefix and use it explicitly:
/opt/homebrew/bin/brew install python@3.12
/opt/homebrew/bin/python3.12 -m venv .venv
source .venv/bin/activate
pip install candy-cazyme

Either way, verify before running a real job:

python3 -c "import platform; print(platform.machine())"   # should print "arm64", not "x86_64"

2. --tree fails to install or build veryfasttree (e.g. a CMake/OpenMP compiler error). This is unrelated to Rosetta -- veryfasttree simply has no macOS arm64 wheel at all, for any Python version, so it always falls back to a from-source build there, and that build fails on stock macOS due to an upstream bug (find_package(OpenMP) fails, since Apple's Clang has no OpenMP support out of the box, and veryfasttree's CMake fallback for that case is itself broken). You shouldn't normally hit this: CANDy detects a Mac without a working native veryfasttree build and automatically defaults --tree-tool to fasttree instead, which does have a real arm64 conda-forge/bioconda build (no compiling anything). That needs the bundled conda environment (a real, one-time dependency for this one platform):

conda env create -f environment.yml
conda activate candy
candy GH173 --email you@example.com --tree   # --tree-tool defaults to fasttree here automatically

If you'd rather force veryfasttree anyway (e.g. you've solved the OpenMP build issue yourself), pass --tree-tool veryfasttree explicitly.

If you'd rather use the original CD-HIT/MAFFT/FastTree tools instead (e.g. to reproduce results bit-for-bit against the published notebook), environment.yml provides CD-HIT and FastTree (conda env create -f environment.yml && conda activate candy, then --clustering-software cd-hit --tree-tool fasttree); MAFFT isn't included there (no osx-arm64 build -- see above) and needs a separate install, e.g. brew install mafft on Intel Mac/Linux, then --alignment-tool mafft.

To also enable automated Gemini-based domain-name curation, see Domain-name curation below.

If you'd rather not have CANDy download anything automatically (e.g. air-gapped environments), set CANDY_NO_AUTO_DOWNLOAD=1 -- clustering will then require mmseqs/cd-hit already on PATH.

# Query a CAZy family directly -- TARGET is auto-detected as a family code or a file path
candy GH173 --email you@example.com --tree

# Restrict to a taxonomic subset, and use a stricter clustering cutoff
candy GH173 --email you@example.com --taxonomy Bacteria --cluster-identity 90

# Reprioritize which InterPro database wins when two disagree on a domain boundary
# (only the databases you name move; everything else keeps its default order)
candy GH173 --email you@example.com --db-preference PFAM,SMART

# Analyse your own FASTA file instead
candy my_sequences.fasta --tree

# Pick a specific MSA/phylogenetics backend explicitly (see Installation for
# when you'd want to -- e.g. --tree-tool fasttree needs the conda environment)
candy GH173 --email you@example.com --tree --alignment-tool mafft --tree-tool fasttree

--email falls back to the CANDY_EMAIL environment variable, then an interactive prompt, so export CANDY_EMAIL=you@example.com once and just run candy GH173 from then on. Run candy --help for the full list of options.

Partway through a run, CANDy needs to decide which raw InterPro domain names (often several near-duplicates from different member databases) should be grouped under one umbrella name. By default (--curation-backend manual) it asks you interactively: it prints a numbered list of the domain names still to curate, then prompts twice --

1. Domain name: -- type the umbrella name you want to use (e.g. Catalytic domain)
2. Includes: -- type the comma-separated numbers of the domains that belong under it (e.g. 0,2)

It repeats this until every domain is grouped; type STOP at the Domain name: prompt at any point to leave all remaining domains as their own individual groups.

To skip this entirely, use Gemini to curate automatically instead:

1. Install the extra: pip install "candy-cazyme[gemini]" (for Mac: uv tool install "candy-cazyme[gemini]" --force --python 3.10)
2. Get a free API key at aistudio.google.com/app/api-keys
3. Run with --curation-backend gemini --curation-api-key YOUR_KEY, or set it once via $env:GOOGLE_API_KEY="YOUR_KEY" (PowerShell) / export GOOGLE_API_KEY=YOUR_KEY (bash) and just pass --curation-backend gemini

If this step appears to hang with nothing printing (v3.0.2+): CANDy now logs before and after the Gemini request, and bounds each attempt to a 60-second timeout, so a multi-minute wait during retries is visible and finite rather than silent. On an older version, this step had no request timeout at all and could hang indefinitely on a stalled connection with zero output -- if you hit this, Ctrl+C, upgrade (pip install --upgrade candy-cazyme), and rerun.

If Gemini itself fails (v3.0.3+, e.g. a 503 UNAVAILABLE "high demand" error, a rate limit, or a model being deprecated): CANDy automatically retries once against a different model (gemini-2.5-flash-lite by default) before giving up on Gemini entirely. If that also fails, it automatically falls back to the interactive manual-curation prompt described above, rather than crashing and discarding the (often several-minutes-long) fetching/clustering/domain-detection work already done in that run.

from candy.config import CAZyFamilyInput, PipelineConfig, Taxonomy
from candy.pipeline import run_pipeline

config = PipelineConfig(
    input=CAZyFamilyInput(enzyme_class="GH", family_number=5, email="you@example.com", taxonomy=Taxonomy.ALL),
    jobname="my_gh5_run",
    output_dir="results",
    build_tree=True,
)
result = run_pipeline(config)
print(result.database_path, result.tree_path)

Results are written to {output_dir}/{jobname}/:

• FASTA files for each processing stage
• A SQLite database ({jobname}_db.db) containing the domain annotations -- open it with DB Browser for SQLite
• A protein domain co-occurrence network ({jobname}_domain_cooccurence_network.graphml) -- open it in Cytoscape (yFiles Organic Layout recommended)
• If --tree/build_tree=True: a FAMSA alignment, a VeryFastTree phylogenetic tree (Newick), and iTOL annotation files for the domain architecture and (for CAZy family queries) characterized-enzyme activity

CANDy communicates with and/or references the following separate libraries, packages and tools:

• Biopython
• pandas
• tqdm
• sqlitebrowser
• SQLAlchemy
• requests
• MMseqs2 / CD-HIT (clustering)
• FAMSA via pyfamsa / MAFFT (alignment)
• VeryFastTree via veryfasttree / FastTree (phylogenetics)
• NetworkX
• Matplotlib

If you find CANDy useful, please cite it as:

Windels A, Franceus J, Pleiss J, Desmet T. CANDy: Automated analysis of domain architectures in carbohydrate-active enzymes. PLoS One. 2024 Jul 11;19(7):e0306410. doi: 10.1371/journal.pone.0306410. PMID: 38990885; PMCID: PMC11238990.

CANDy is licensed under MIT.

CANDy and other information provided is for theoretical utilisation only, caution should be exercised in its use. It is provided 'as-is' without any warranty of any kind, whether expressed or implied. Information is not intended to be a substitute for professional medical advice, diagnosis, or treatment, and does not constitute medical or other professional advice.

Use of the third-party software, libraries or code referred to in the Acknowledgements section in the CANDy README may be governed by separate terms and conditions or license provisions. Your use of the third-party software, libraries or code is subject to any such terms and you should check that you can comply with any applicable restrictions or terms and conditions before use.

The following databases are used by CANDy, and are available with reference to the following:

• UniProt: (unmodified), by The UniProt Consortium, available under a Creative Commons Attribution-NoDerivatives 4.0 International License.
• NCBI: (unmodified), by the National Library of Medicine, available under a Creative Commons Attribution-NoDerivatives 4.0 International License.
• CAZy: (unmodified), by http://www.cazy.org/ and Elodie Drula, Marie-Line Garron, Suzan Dogan, Vincent Lombard, Bernard Henrissat, Nicolas Terrapon, The carbohydrate-active enzyme database: functions and literature, Nucleic Acids Research, Volume 50, Issue D1, 7 January 2022, Pages D571–D577, https://doi.org/10.1093/nar/gkab1045, available under a Creative Commons Attribution-NoDerivatives 4.0 International License.
• InterPro: (unmodified), by EMBL-EBI, available under a Creative Commons Attribution-NoDerivatives 4.0 International License.