enFoldX: Turning AlphaFold3 Structural Noise into TCR Recognition Predictions

Recently, researcher quaidmorris detailed an innovative framework named enFoldX, which transforms the structural noise of AlphaFold3 (AF3) into predictive signals for T-cell receptor (TCR) recognition. The study notes that sequence-based models often struggle to generalize when predicting which T cells recognize specific peptides. However, utilizing structural ensembles generated by AF3 provides more robust generalization than relying on a single predicted structure.

Core Mechanism and Workflow

enFoldX shifts away from traditional structural prediction approaches. Instead of extracting a single structure during AF3 predictions for TCRα, TCRβ, MHC, and the peptide, the complete workflow generates an entire ensemble to use directly as classification input. By analyzing this structural ensemble, the model better captures the true characteristics of TCR and pMHC interactions.

Hallucination Frequency as a Signal

The study specifically highlights AF3's "hallucination" phenomenon. Using the CMV epitope `NLVPMVATV` as an example, AF3 produces hallucinations for both true binders and non-binders, but the frequency of these hallucinations is itself informative. The researchers found that AF3 hallucinates in about 7 out of 10 predictions on non-cognate TCR:pMHC pairs. enFoldX leverages this high-frequency noise characteristic seen in non-binders to effectively predict T-cell specificity.

Framework Extensibility

The authors emphasize that the core to robust generalization for TCR:pMHC lies in structural ensembles rather than single predictions. Therefore, this ensemble-based framework is not only applicable to AlphaFold3 but should theoretically extend smoothly to other structural prediction models like RoseTTAFold.

2026-07-22 ~ 2026-07-22 · 11 related posts

• enFoldX uses noisy AlphaFold3 ensembles to predict T cell specificity — quaidmorris · 2026-07-22
• AF3 hallucinations appear in 7 of 10 seeds for a non-cognate TCR:pMHC pair — quaidmorris · 2026-07-22
• enFoldX gains accuracy as AF3 ensemble disagreement rises for non-binders — quaidmorris · 2026-07-22
• enFoldX classifies TCR–pMHC pairs from AF3 ensembles instead of one structure — quaidmorris · 2026-07-22
• [source] enFoldX extracts 106 interface and confidence features from AF3 ensembles — quaidmorris · 2026-07-22
• [source] enFoldX reaches AUC 0.82 on human VDJdb and transfers to mouse at 0.76 — quaidmorris · 2026-07-22
• enFoldX tops 8 neoantigen scans and an unseen-peptide benchmark — quaidmorris · 2026-07-22
• [source] enFoldX turns AlphaFold3 ensemble noise into a TCR–peptide–MHC predictor — quaidmorris · 2026-07-22
• Structural ensembles beat single predictions in TCR:pMHC generalization study — quaidmorris · 2026-07-22
• Structural ensembles, not single predictions, drive robust TCR:pMHC generalization — quaidmorris · 2026-07-22
• Turning Noise into Signal: Predicting TCR Binding Using AlphaFold3 Hallucinations — quaidmorris · 2026-07-22