AIXC BioSciences Platform Structure Prediction
Structure Prediction

Boltz-2AI Structure Prediction.
Multiple Backends.
GPU.

State-of-the-art protein structure prediction with Boltz-2our primary engine and three complementary backends. Template conditioning, multi-chain co-folding, and per-residue confidence metrics. 5-6x speedup.

Request Structure Prediction View Sample Output
PREDICTION BACKENDS

Boltz-2Primary Engine + Three Complementary Engines

Our primary engine is Boltz-2a state-of-the-art generative model, the latest generative structure prediction model. Three additional backends provide cross-validation and specialized capabilities.

Boltz-2Primary Engine
Primary Engine
State-of-the-art generative model. Handles proteins, nucleic acids, small molecules, and covalent modifications. Template conditioning and confidence scoring built in.
Chai-1Multi-Solver Engine
Multi-Solver Backend
Six solver backends for cross-validation. Combines deep learning with physics-based refinement. Excellent for complex multi-chain assemblies.
ESM-2 / ESM-3Language Model Engine
Language Model Embeddings
Meta's protein language models provide evolutionary-scale embeddings. ESM-2Variant model for variant effect prediction, ESM-3generation model for fast structure generation from sequence alone.
CAPABILITIES

Advanced Folding Features

Beyond simple monomer prediction. Our pipeline handles the complex scenarios that real drug discovery projects require.

Template Conditioning
Provide experimental structures as templates to guide prediction. Particularly valuable for homologous targets where partial structural information exists.
Multi-Chain Co-Folding
Predict complex assemblies with multiple protein chains, nucleic acids, and ligands simultaneously. Essential for antibody-antigen and multi-subunit complexes.
Per-Residue Confidence
pLDDT scores for every residue, pTM for overall model quality, and PAE (predicted aligned error) matrices for inter-domain confidence assessment.
QUALITY METRICS

Confidence You Can Trust

Every prediction comes with quantitative confidence scores. Know exactly which regions are reliable and which need experimental validation.

pLDDT
Predicted Local Distance Difference Test
Per-residue confidence score (0-100). Scores above 90 indicate very high confidence. Below 50 indicates likely disordered regions.
0 Low507090+ High
pTM
Predicted Template Modeling Score
Global model quality metric (0-1). Scores above 0.5 suggest a correct overall fold. Scores above 0.8 indicate high-quality predictions suitable for downstream analysis.
0.00.30.50.8+
PERFORMANCE

GPU Inference

Our infrastructure runs on dedicated GPU clusters, delivering significant speedups over CPU-only inference.

GPU Inference

Dedicated GPU compute enables high-throughput inference. Boltz-2Primary and ESMlanguage models run natively on high-bandwidth memory, eliminating transfer bottlenecks.

80GB
GPU Memory
6,912
GPU Cores
312
Tensor Cores
5-6x
Speedup vs CPU
COMPARISON

Structure Prediction Methods

How Boltz-2our primary engine compares to other leading structure prediction tools across key capabilities.

Feature Comparison Matrix
FeatureAlphaFold2Method AESMFoldMethod BBoltz-2AIXC Primary (AIXC)
Multi-chain co-foldingLimitedNoFull
Small molecule dockingNoNoYes
Nucleic acid supportNoNoYes
Template conditioningYesNoYes
MSA-free modeNoYesYes
Covalent modificationsNoNoYes
Confidence metricspLDDT, pTMpLDDTpLDDT, pTM, PAE, ipTM
GPU supportNoCPU onlyFull GPU
GET STARTED

Need a Structure Predicted?

Submit your protein sequence and we will deliver high-confidence 3D structure predictions with full confidence metrics and downloadable PDB/CIF files.

Request Structure Prediction View Pricing