Research map / Architectures and efficiency / Attention and transformers
Quantifying Cross-Attention Interaction in Transformers for Interpreting TCR-pMHC Binding
Jiarui Li, Zixiang Yin, Haley Smith, Zhengming Ding, Samuel J Landry, Ramgopal R. Mettu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use QCAI with cross-attention instead of best self-attention baseline | TULIP · TCR-XAI peptide perturbation | LOdds: -0.736 -> -1.470 | Improved |
| use QCAI instead of Rollout | TULIP · TCR-XAI CDR3a perturbation | AOPC: 0.022 -> 0.014 | Hurt |
| use QCAI instead of best baseline | TULIP · TCR-XAI BRHR top-25% CDR3a | hit rate (%): 68.5 -> 79.1 | Improved |
| use QCAI instead of other methods | TULIP · TCR-XAI ROC-AUC peptide | ROC-AUC: 0.6024 | Improved |
| apply QCAI to cross-attention only vs both cross and self attention (ablation) | QCAI · TCR-XAI CDR3b perturbation | AOPC: 0.045 -> 0.048 | No clear effect |
| apply QCAI to self-attention only (Rollout) instead of cross-attention (ablation) | QCAI · TCR-XAI peptide perturbation | LOdds: -1.470 -> -0.044 | Hurt |
| use max instead of average aggregation across cross-attention layers (ablation) | QCAI · TCR-XAI CDR3a | BRHR.25 (%): 72.6 -> 79.1 | Improved |
| use max instead of average aggregation across cross-attention layers (ablation) | QCAI · TCR-XAI peptide | BRHR.25 (%): 76.7 -> 74.3 | Hurt |
| use QCAI instead of TokenTM on a larger integrated dataset | TULIP · Integrated peptide perturbation | AOPC: 0.026 -> 0.036 | Improved |
| replace faster baselines with QCAI | QCAI · per-sample runtime | ms/sample: 0.11 -> 2.19 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AttnLRP — AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformers (ICML 2024) · better on 7 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DecoderTCR: Compositional Pretraining and Entropy-Guided Decoding for TCR-pMHC Interactions (ICML 2026 · similarity 0.89)
- Relation-Aware Equivariant Graph Networks for Epitope-Unknown Antibody Design and Specificity Optimization (AAAI 2025 · similarity 0.80)
- Cross-Gate MLP with Protein Complex Invariant Embedding Is a One-Shot Antibody Designer (AAAI 2024 · similarity 0.80)
- A Label Disambiguation-Based Multimodal Massive Multiple Instance Learning Approach for Immune Repertoire Classification (AAAI 2024 · similarity 0.80)
- Learning Identifiable Factorized Causal Representations of Cellular Responses (NeurIPS 2024 · similarity 0.80)
- Learning Cross-Domain Representations for Transferable Drug Perturbations on Single-Cell Transcriptional Responses (AAAI 2025 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
