Bi-directional state space model for long-range DNA sequence modeling
Caduceus introduces bi-directional Mamba state space architectures with reverse-complement equivariance, outperforming transformers on long genomic sequence prediction while reducing compute footprint.
Bi-directional state space (Mamba) architecture overcoming causal autoregressive limitationsGuaranteed reverse-complement (RC) equivariance essential for double-stranded DNA modelingSub-quadratic scaling allowing 131k base pair context on single standard GPUsSuperior performance on variant effect and chromatin accessibility benchmarks
Caduceus introduces bi-directional Mamba state space architectures with reverse-complement equivariance, outperforming transformers on long genomic sequence prediction while reducing compute footprint.
Information checked against an official source; not a hands-on test. Source · Last reviewed: 20/09/2026, 11:10:15
Key Features
Bi-directional state space (Mamba) architecture overcoming causal autoregressive limitations
Guaranteed reverse-complement (RC) equivariance essential for double-stranded DNA modeling
Sub-quadratic scaling allowing 131k base pair context on single standard GPUs
Superior performance on variant effect and chromatin accessibility benchmarks
Academic Context & Research Evidence
Biological & Workflow Fit
Biological Application
Long-range gene regulation, promoter-enhancer mapping, and variant impact scoring
Research Workflow
Genomic DNA sequence -> predict regulatory activity with RC equivariance
Use this citation format when referencing Caduceus in scientific publications and benchmark papers.
@software{caduceus_2026,
title = {{Caduceus}},
author = {{Stanford University & Princeton University}},
year = {2026},
url = {https://github.com/kuleshov-group/caduceus},
note = {Indexed on aibioatlas - AI for Biology and Drug Discovery}
}
Peer-Reviewed Literature & Preprints
Live scientific citations streamed from Europe PMC and PubMed for Caduceus.
⏳ Fetching real-time literature from Europe PMC & PubMed...
Technical / Product Information
Missing values mean the catalog has no recorded information. They do not mean a feature is absent.
Entry typeAI Model
Access modeOpen Source
AI roleEquivariant State Space Modeling
Input dataNot recorded
Output dataNot recorded
Licence conditionsApache 2.0
Commercial eligibilityOpen source release
Compute requirements1x GPU (>= 16GB VRAM)
ValidationNot recorded
TypeBi-directional Mamba genomic model
Intended useNot recorded
CompatibilityNot recorded
ManufacturerStanford University & Princeton University
Biological applicationLong-range gene regulation, promoter-enhancer mapping, and variant impact scoring
Research workflowGenomic DNA sequence -> predict regulatory activity with RC equivariance