Structural Biology

AlphaFold Structure Prediction: From Amino Acid
Sequence to 3D Protein Structure

BioMate runs AlphaFold 2 and AlphaFold 3 on AWS GPU instances. Submit a UniProt ID, FASTA sequence, or gene name — BioMate returns the predicted structure with pLDDT confidence scores, PAE plots, and a ready-to-dock PDB file, all in a single chat message.

Try free See all structural biology services
AlphaFold 2 vs AlphaFold 3

Choosing the right model for your target

BioMate automatically selects the appropriate AlphaFold version based on your target type. You can override the selection at any time from the parameter panel.

Capability AlphaFold 2 AlphaFold 3
Single-chain protein Best accuracy; recommended for most targets Supported; similar accuracy to AF2
Protein complex (multimer) AlphaFold-Multimer; good for homodimers and heterodimers Improved multimer accuracy; use for large complexes
Protein–DNA / Protein–RNA Not natively supported Supported; transcription factor–DNA, ribosome–RNA
Protein–small molecule Not supported Supported; predict binding pose alongside structure
Compute requirement Single A100 GPU; ~20–60 min for typical protein Higher memory; ~45–120 min for complexes
pLDDT confidence Per-residue (0–100) Per-residue + ipTM for interface quality
Confidence metrics

Understanding pLDDT, PAE, and ipTM

pLDDT (predicted local distance difference test)

Per-residue confidence score from 0–100. Above 90: very high confidence (well-structured region). 70–90: confident. 50–70: low confidence (may be disordered). Below 50: likely intrinsically disordered.

PAE (predicted aligned error)

2D matrix showing the expected positional error between residue pairs. Low PAE values across a domain indicate a well-defined relative arrangement. High PAE between domains indicates flexible or uncertain interdomain geometry.

ipTM (interface predicted TM-score)

AlphaFold 3 metric specific to complexes. Measures confidence in the predicted interface geometry. ipTM above 0.8 indicates a reliable complex prediction; below 0.5 suggests the interaction may not be well-defined.

pLDDT color guide
pLDDT > 90 Very high confidence
pLDDT 70–90 Confident
pLDDT 50–70 Low confidence
pLDDT < 50 Likely disordered
BioMate flags binding site residues with pLDDT < 70 and excludes them from docking grid definition automatically.
Output files

What you receive from each prediction run

File Format Description
Predicted structure PDB / mmCIF 3D coordinates with B-factor column containing pLDDT per residue
Confidence JSON JSON Per-residue pLDDT array and summary statistics
PAE plot PNG Predicted aligned error heatmap; identifies domain boundaries and flexible regions
Multiple sequence alignment A3M MSA used as input; available for review or re-use with custom MSA
Methods report DOCX AlphaFold version, MSA depth, template hits, and all parameters cited for publication
Integrated pipeline

AlphaFold → AutoDock Vina in one request

BioMate connects structure prediction directly to molecular docking. When you ask BioMate to "dock compound X against target Y," it automatically runs AlphaFold if no experimental structure exists, then feeds the high-confidence regions of the predicted structure into the docking pipeline — all in a single workflow run.

  1. AlphaFold 2 or 3 predicts the target structure Sequence fetched from UniProt automatically. GPU instance selected based on sequence length and target type.
  2. pLDDT-based quality filter applied to binding site Residues with pLDDT < 70 in the predicted binding pocket are flagged. BioMate warns if the active site has low-confidence regions that may affect docking reliability.
  3. Docking grid defined on high-confidence residues AutoDock Vina or GLIDE grid centered on the predicted binding site, excluding disordered loops or terminal regions.
  4. Ligand series docked and ranked Binding poses returned with docking scores, interaction fingerprints, and pLDDT-weighted confidence annotation for each contact residue.
FAQ

Common questions about AlphaFold in BioMate

Does BioMate support AlphaFold 2 and AlphaFold 3?

Yes. BioMate runs both AlphaFold 2 (single-chain proteins, high accuracy for most targets) and AlphaFold 3 (protein–DNA, protein–RNA, and protein–small molecule complexes). The version is selected based on your target type.

What input does AlphaFold structure prediction require?

A FASTA sequence, UniProt ID, or gene name. BioMate fetches the sequence automatically from UniProt if you provide an ID.

How do I interpret pLDDT scores?

pLDDT (predicted local distance difference test) ranges from 0–100. Scores above 90 indicate very high confidence; 70–90 are confident; below 50 indicate disordered or low-confidence regions. BioMate color-codes the structure by pLDDT and flags low-confidence binding site residues.

Can I use the predicted structure for molecular docking?

Yes. BioMate feeds AlphaFold output directly into AutoDock Vina or GLIDE in a single integrated pipeline, filtering out low-pLDDT regions from the docking grid automatically.

Get started

Predict your first protein structure today

Provide a UniProt ID or FASTA sequence. BioMate runs AlphaFold on GPU, returns the structure with confidence annotations, and feeds it directly to docking if you need it.

Try free →