Ranomics
Predicted protein structure with cartoon backbone colored by pLDDT confidence, showing blue high-confidence helices and beta sheets alongside orange and yellow loop regions
Structure prediction

AlphaFold2 protein structure prediction

Gold-standard AlphaFold protein structure prediction with full multiple sequence alignment, structural templates, and calibrated per-residue pLDDT and pairwise PAE confidence. Run a monomer or switch to AF2 multimer for protein complexes with pTM and ipTM scores at the interface.

Hosted at tools.ranomics.com with no shared queues and no install. Free tier covers triage runs.

Used by Ranomics internally to validate designed binders, model new targets, and triage candidate scaffolds before wet-lab campaigns.

How it works

From sequence to scored structure

01

Submit sequence

Paste a single FASTA sequence for monomer mode, or up to several chains for multimer. Optional template PDB IDs can be supplied to anchor the prediction.

02

MSA construction

Homologous sequences are searched against UniRef, BFD, and MGnify with JackHMMer and HHblits. The resulting MSA encodes the coevolutionary signal that drives accuracy.

03

Evoformer + structure module

48 Evoformer blocks iteratively refine paired MSA and residue-residue representations. The structure module then folds an explicit 3D backbone with rotation-equivariant attention.

04

Structure + confidence

Output is a relaxed PDB with per-residue pLDDT and a full pairwise PAE matrix. Multimer mode adds pTM and ipTM for global and interface confidence.

Methodology

Architecture and confidence calibration

Jumper et al. (Nature, 2021) introduced AlphaFold2 with median backbone accuracy of 0.96 Å Cα RMSD on CASP14 free-modeling targets, a result that rendered most prior methods obsolete. The accuracy and the calibration of its confidence scores are why AlphaFold2 remains the structure-prediction reference.

Core

Evoformer trunk

48 stacked blocks operating jointly on the MSA representation and a pairwise residue-residue representation. Row, column, and triangular attention enforce structural and evolutionary consistency before any 3D coordinates are produced.

Input

MSA depth

Accuracy scales with the number of effective sequences (Neff). Hundreds of diverse homologs produce confident folds. Orphan sequences and de novo designs lack MSA signal. For those, use ColabFold or ESMFold instead.

Input

Structural templates

Up to four homologous PDB structures inform the pair representation. Templates help when the MSA is shallow or when a known fold should anchor the prediction.

Output

pLDDT confidence

Per-residue 0 to 100 score predicting local accuracy. Above 90 is high confidence and crystal-comparable. 70 to 90 is confident. 50 to 70 is low. Below 50 generally indicates disorder rather than a wrong fold.

Output

PAE matrix

Predicted Aligned Error reports expected error in Å for every residue pair. Low PAE blocks identify rigid domains and validated inter-chain orientations. High PAE between domains flags hinged or flexible regions.

Beyond monomer folds

Multimer mode and method selection

AF2 multimer for complexes

Switch to multimer mode to fold protein complexes end to end. Paired MSAs across chains encode co-evolution at the interface, and the structure module produces a single assembled complex rather than independent chains stitched together.

pTM and ipTM for interfaces

Multimer outputs add two scalar scores. pTM estimates global TM-score for the complex. ipTM is restricted to inter-chain residue pairs and is the standard binder-design metric. A high ipTM indicates a confidently modeled binding interface.

When to skip the MSA

For designed sequences, orphan proteins, or high-throughput triage where MSA construction does not add signal, use ColabFold with reduced MSA depth or ESMFold for single-sequence folds.

When to use AlphaFold2

The reference structure for any campaign

AlphaFold2 is the structure most reviewers, collaborators, and downstream tools expect to see. When you need the highest-confidence single model for a natural protein with available homologs, full-MSA AF2 is still the answer, not a faster surrogate.

We run AF2 as the validation step on designed binders, the modeling step on novel targets before epitope scouting, and the comparison anchor when a crystal structure exists.

Validating de novo designed binders by refolding the sequence and checking ipTM at the interface

Modeling a new target with available homologs before running an Epitope Scout or RFdiffusion campaign

Triaging protein complexes by asking whether AF2 multimer predicts a plausible interface before committing wet-lab resources

Comparing predicted structure to a crystal or cryo-EM structure to flag conformational disagreements

Producing the reference structure for a paper, grant, or pitch deck when ColabFold or ESMFold is not authoritative enough

Generating a high-confidence template for downstream tools (BindCraft, PXDesign) that consume PDB inputs

Predict your first structure today

Create a free account on tools.ranomics.com and submit a sequence. Full-MSA monomer or AF2 multimer with structure and confidence scores returned.