In brief: ESMFold2 focuses on all-atom structure prediction from sequence inputs. Its output is a model-derived structural hypothesis and should be interpreted alongside confidence measures and other evidence.
Start with the task
Folding an existing sequence is different from designing a new one. ESMFold2 provides a prediction workflow for sequences and complexes. Check its current support for proteins, DNA, RNA, ligands and optional MSA input before assuming any combination is accepted in your environment.
Connect with the documented client
The Biohub model page lists SequenceStructureForgeInferenceClient and a current example model ID. The snippet below initializes a client and is not a complete inference call:
import os
from esm.sdk.forge import SequenceStructureForgeInferenceClient
client = SequenceStructureForgeInferenceClient(
model="esmfold2-fast-2026-05",
url="https://biohub.ai",
token=os.environ["ESM_API_KEY"],
)Use the official folding notebook for request formatting and interpretation of returned structure objects.
Confidence is not verification
Low confidence may suggest uncertainty, but high confidence does not certify that a protein adopts one unique conformation or functions in a cell. Validate interfaces, clashes and domain orientations, and compare with appropriate experimental information.
What to read next
Explore the official ESMFold2 documentation, the primary technical preprint and the ESM Atlas. For novel biological claims, document the target, prediction settings and validation strategy.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.