ColabFold

ColabFold made AlphaFold usable. By replacing the original pipeline’s slow database search with the far faster MMseqs2 and packaging the whole thing in a Google Colab notebook, it turned a multi-day, cluster-dependent job into something a researcher could run in a browser over lunch.

What it does

The critical change is the multiple sequence alignment step, which dominated AlphaFold’s runtime. ColabFold’s remote MMseqs2 search cuts that dramatically with little accuracy cost, then runs prediction on Colab’s free GPU. It also supports local installation for batch work and complex prediction for multimeric assemblies.

Strengths

• Free, including the compute, via Google Colab.
• Orders-of-magnitude faster alignment than the reference pipeline.
• No local installation or cluster access required to start.
• Widely cited, well maintained, and supports batch and complex modes.

Limitations

Free Colab imposes session limits and variable GPU allocation, which makes large batches awkward without a paid tier or local install. Sequences go to a remote MMseqs2 server, so it is unsuitable for confidential targets unless self-hosted. It packages AlphaFold rather than replacing it, and inherits its limitations.