Ranomics
ColabFold AlphaFold2 predicted protein structure colored by per-residue pLDDT confidence score
Structure prediction

ColabFold fast protein structure prediction

ColabFold swaps the slow homology-search step for an MMseqs2-based MSA pipeline that runs roughly an order of magnitude faster, with no measurable loss in accuracy on well-represented targets. Built for throughput when you have hundreds of designs to triage.

Monomer or multimer. Single-sequence or full MSA. Templates optional. Per-residue pLDDT and PAE on every output.

Hosted at tools.ranomics.com. Free tier available on signup.

How it works

From sequence to folded model in one pass

01

Submit sequence

Paste a FASTA sequence or upload a multi-sequence file. Monomers run as-is; multimers use a colon to separate chains.

02

MMseqs2 search

Fast homology search against UniRef30 and an environmental database. Returns a deep MSA without the overhead of a classical alignment search.

03

AF2 inference

Run the AlphaFold2 weights with the assembled MSA, optional PDB70 templates, and configurable recycle iterations.

04

Folded model

Download a ranked PDB ensemble with per-residue pLDDT, PAE matrices, and predicted TM-score for multimer interfaces.

Methodology

What ColabFold changes inside the AF2 pipeline

ColabFold keeps the AlphaFold2 neural-network weights unchanged. The throughput gain comes from replacing the homology-search frontend and exposing inference knobs that most pipelines hide.

Fast MSA

MMseqs2 MSA search

Replaces a classical jackhmmer + HHblits search against the full BFD with a precomputed MMseqs2 server hitting UniRef30 and ColabFoldDB. Same MSA depth, far less overhead.

3 modes

MSA modes

Full MSA for well-represented sequences. Single-sequence mode for designed binders and orphan proteins. Paired mode aligns columns across chains for AF2-multimer.

pLDDT

Per-residue confidence

pLDDT 0-100 per residue tells you which regions of the model to trust. PAE matrix tells you which inter-residue distances are confidently resolved.

Optional

PDB70 templates

Optional template input from PDB70. Useful for sequences with close homologs in the PDB; usually skipped for de novo designed scaffolds.

3 default

Recycle iterations

Each recycle feeds the previous prediction back as input. Default 3, configurable up to 12, with diminishing returns past 6 for most targets.

Beyond a single fold

Multimer, model choice, and batch throughput

Multimer mode

Predicts protein complexes using AF2-multimer weights or the residue-index-jump trick on the monomer model. Returns a predicted interface TM-score and pAE matrix so you can quantify confidence in the inter-chain contacts, not just the chain folds.

When to pick which fold

ColabFold for throughput on natural sequences with MSAs. Full-MSA AF2 for the final, highest-accuracy fold on a single chosen design. ESMFold for the fastest single-sequence pass on designed binders, orphan sequences, or anything where an MSA would not help.

Batch throughput

ColabFold is the throughput tool of choice for triaging large design pools. Fold an entire RFdiffusion or BindCraft output set in a single sitting, then promote the highest-confidence handful to full-MSA AF2 or wet-lab validation.

When to use ColabFold

When speed beats the last decimal of accuracy

Full-MSA AlphaFold2 is the gold standard for a single, definitive structure. But most real campaigns generate hundreds or thousands of candidate sequences, and you cannot afford to fold them all with the slow path. ColabFold is the throughput tool that bridges the gap between sequence generation and final-model validation.

Use ColabFold to rank a design pool, screen mutants, or survey a target family. Promote the top handful to full-MSA AF2 or experimental validation.

Triaging a pool of designed binder sequences before committing to wet-lab validation

Folding all variants in a deep mutational scanning library to flag misfolded designs

Surveying a target protein family by folding all orthologs in a single pass

Running a design-validation pass on RFdiffusion, BindCraft, or BoltzGen outputs

Building a structural prior for downstream docking, MD, or interface analysis

Quick sanity-check folds where 1-2 angstrom accuracy is enough for the decision

Fold your first sequence today

Create a free account on tools.ranomics.com and run ColabFold on a designed sequence. No shared queues, no install.

Method: Mirdita et al., ColabFold: making protein folding accessible to all. Nat Methods 19, 679-682 (2022). See also all Ranomics technology.