Use AlphaFold Online
Commercially Available AlphaFold No-Code Web Server
AlphaFold: Protein Monomer and Multimer Structure Prediction from Sequence
AlphaFold is a neural network that predicts the three-dimensional structure of a protein from its amino acid sequence, and was reported to regularly reach atomic accuracy even when no similar structure is known. It was developed by DeepMind (London) and described by Jumper, Evans, Pritzel and colleagues in Nature in 2021. The open-source inference pipeline for AlphaFold v2 is released under the Apache License 2.0 (code), with model parameters under CC BY 4.0, and it also includes AlphaFold-Multimer for protein complexes.
Knowing a protein's structure is central to understanding its function, but experimental determination is slow and costly. The paper notes that earlier computational methods fell well short of atomic accuracy, especially when no homologous structure was available to use as a template. AlphaFold was evaluated in CASP14, a blind assessment, where it performed competitively with experimental structures in most cases.
The AlphaFold-Multimer extension targets complexes. The official repository supports homomers and heteromers from multi-sequence FASTA input, although it describes the multimer system as a work in progress that is less stable than the monomer system.
Feature | Details |
|---|---|
Task | Protein 3D structure prediction from sequence (monomer or multimer) |
Inputs | Amino acid sequence, MSAs of homologs and optional templates; FASTA file |
Outputs | Five ranked PDB models with per-residue pLDDT; PAE and pTM for pTM models |
CASP14 backbone accuracy | Median 0.96 Å r.m.s.d.95 versus 2.8 Å for the next-best method |
License | Code: Apache 2.0; model parameters: CC BY 4.0 |
Why Accurate Protein Structure Prediction Matters
Reliable structures from sequence alone let researchers form hypotheses about function, binding sites and domain organization before any wet-lab work. Because AlphaFold also reports how confident it is in each region, scientists can decide which parts of a model to trust and which to treat with caution.
How AlphaFold Predicts Structure: Evoformer, Structure Module and Recycling
Inputs: the network takes the target sequence, multiple sequence alignments of homologs, and optional template structures.
Evoformer trunk: repeatedly updates an MSA representation and a residue-pair representation, exchanging information between them in every block (48 blocks in the paper).
Structure module: builds the 3D structure as per-residue rotations and translations using invariant point attention, trained with a frame-aligned point error loss.
Recycling: outputs are fed back into the network repeatedly, which the authors report adds notable accuracy for little extra training time.
Confidence and relaxation: pLDDT is predicted per residue, and a final Amber relaxation step removes stereochemical violations.
AlphaFold Accuracy Results and Confidence Metrics
Benchmark | AlphaFold result |
|---|---|
CASP14 backbone (87 domains) | Median 0.96 Å r.m.s.d.95 (next-best method 2.8 Å) |
CASP14 all-atom | Median 1.5 Å r.m.s.d.95 (best alternative 3.5 Å) |
Recent PDB chains without a close template (n = 3,144) | Median backbone r.m.s.d. of 1.46 Å |
pLDDT versus true lDDT-Cα | Pearson r = 0.76 (10,795 chains) |
Reading the confidence scores
pLDDT: per-residue score from 0 to 100, stored in the B-factor column of the output PDB files; higher means more confident.
PAE: predicted aligned error between residue pairs, where lower is more confident; it indicates how well domains are packed relative to each other (pTM models only).
pTM: a scalar estimate of global fold quality for pTM models.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with AlphaFold on Tamarind Bio
Structure hypotheses for new targets: model proteins that lack an experimental structure to guide function analysis.
Domain architecture analysis: use PAE to judge how confidently separate domains are placed relative to each other.
Homomer and heteromer modeling: predict complexes with the AlphaFold-Multimer system from multi-sequence input.
Starting models for downstream design: use ranked structures as inputs to docking, design or simulation workflows.
How to Use AlphaFold on Tamarind Bio
Log in: sign in to tamarind.bio and open the AlphaFold tool.
Provide the sequence: enter a single amino acid sequence for a monomer, or multiple sequences for a complex (FASTA-style input as the source defines).
Choose monomer or multimer mode: if exposed, select the preset that matches your target.
Set options: if exposed, adjust settings such as the number of predictions per model or relaxation.
Run the job: submit it and wait for the prediction to finish.
Download outputs: retrieve the PDB models, ranked from most to least confident.
Interpret confidence: inspect pLDDT, and PAE where available, before relying on any region.
Use downstream: take the top-ranked model into further analysis or design.
Things to Keep in Mind
Shallow alignments: accuracy drops substantially when the median MSA depth is below about 30 sequences.
Complexes: the original model was weaker for proteins whose shape depends mainly on contacts with other chains, and the repository calls AlphaFold-Multimer a work in progress.
Run-to-run variance: some targets give different results across random seeds.
Compute grows with length: the repository benchmark shows about 4.9 s for 100 residues but about 18,824 s for 5,000 residues on one A100 GPU, excluding MSA and template search.
Relaxation: the final relaxation removes stereochemical violations but does not improve accuracy scores.
Not for clinical use: the repository states outputs are theoretical modeling and not medical advice.
Source: Jumper, J., Evans, R., Pritzel, A., et al. "Highly accurate protein structure prediction with AlphaFold." Nature 596, 583-589 (2021). DOI: 10.1038/s41586-021-03819-2. Code: github.com/google-deepmind/alphafold. AlphaFold-Multimer: Evans, R., et al., bioRxiv 2021, DOI 10.1101/2021.10.04.463034.