SwiftRepertoire: Few-Shot Immune-Signature Synthesis via Dynamic Kernel Codes

Chunlei Meng; Jiaxuan Lu; Jiekai Wu; Juntao Gao; Li Bao; Muge Qi; Qi Zhao; Rong Fu; Simon Fong; Wei Luo

arxiv: 2602.01051 · v4 · submitted 2026-02-01 · 💻 cs.LG

SwiftRepertoire: Few-Shot Immune-Signature Synthesis via Dynamic Kernel Codes

Rong Fu , Muge Qi , Yang Li , Yabin Jin , Jiekai Wu , Jiaxuan Lu , Chunlei Meng , Youjin Wang

show 6 more authors

Zeli Su Juntao Gao Li Bao Qi Zhao Wei Luo Simon Fong

This is my paper

Pith reviewed 2026-05-16 09:08 UTC · model grok-4.3

classification 💻 cs.LG

keywords few-shot learningimmune repertoireT-cell receptorsadapter synthesisdynamic kernel codesmotif discoveryfew-shot adaptation

0 comments

The pith

A framework synthesizes compact adapters from repertoire data to adapt frozen models for new immune tasks with few examples.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a synthesis framework for creating small task-specific adapters for analyzing T cell receptor repertoires. These adapters are generated from a dictionary of prototypes using lightweight descriptors from the data. This allows adaptation to new tasks with only a few examples while keeping the main model frozen. It matters because it reduces the need for large labeled datasets and heavy computation in clinical immune monitoring, while linking predictions back to biological motifs.

Core claim

The paper claims that synthesizing compact task-specific parameterizations from a learned dictionary of prototypes conditioned on lightweight task descriptors derived from repertoire probes and pooled embedding statistics produces small adapter modules. These modules can be applied to a frozen pretrained backbone to enable immediate adaptation to novel tasks with only a handful of support examples without full model fine-tuning. Interpretability is preserved through motif-aware probes and a calibrated motif discovery pipeline that links decisions to sequence signals.

What carries the argument

The synthesis of adapter modules from a dictionary of prototypes conditioned on task descriptors from repertoire probes and pooled statistics.

If this is right

Enables practical deployment in clinical settings with scarce labeled data.
Conserves computational resources by freezing the pretrained backbone.
Maintains interpretability by connecting predictions to sequence motifs.
Supports adaptation across heterogeneous cohorts without retraining.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This could be extended to other types of biological sequence data for few-shot tasks.
It opens possibilities for real-time model adaptation in resource-limited environments like mobile diagnostics.
The approach highlights how prototype dictionaries can serve as a general mechanism for task conditioning in sequence models.

Load-bearing premise

That lightweight task descriptors derived from repertoire probes and pooled embedding statistics are sufficient to condition synthesis of effective task-specific parameterizations across diverse clinical tasks.

What would settle it

A demonstration that synthesized adapters fail to achieve competitive performance compared to fully fine-tuned models on a diverse set of held-out immune repertoire tasks with high heterogeneity would falsify the claim.

Figures

Figures reproduced from arXiv: 2602.01051 by Chunlei Meng, Jiaxuan Lu, Jiekai Wu, Juntao Gao, Li Bao, Muge Qi, Qi Zhao, Rong Fu, Simon Fong, Wei Luo, Yabin Jin, Yang Li, Youjin Wang, Zeli Su.

**Figure 2.** Figure 2: t-SNE projection of adapter vectors and learned prototypes. Points are colored by disease / task ID; black [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Prototype activation heatmap across disease types. Rows correspond to prototypes and columns to disease [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Performance as a function of support-set size. Curves show AUC versus support set size (5, 10, 20, 50) for [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Stability of key metrics across random seeds. Boxplots summarize the distribution of AUC, F1 and ECE [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Health score (defined as 1− predicted cancer probability) versus subject age. Scatter points show individual libraries; the dashed curve is a fitted smoothing polynomial. Reported Spearman correlation quantifies the immunosenescence trend. 3.4 Bootstrap estimation for prototype coverage and for the selected dimension r Construct a PCA projection Πr : R dθ → R r , (4) where Πr denotes the PCA projection op… view at source ↗

**Figure 7.** Figure 7: Calibration plot comparing predicted probabilities to observed outcome frequencies. The diagonal indicates [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

**Figure 8.** Figure 8: Training and validation AUC across epochs. The plot visualizes convergence behaviour and early-stopping [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗

read the original abstract

Repertoire-level analysis of T cell receptors offers a biologically grounded signal for disease detection and immune monitoring, yet practical deployment is impeded by label sparsity, cohort heterogeneity, and the computational burden of adapting large encoders to new tasks. We introduce a framework that synthesizes compact task-specific parameterizations from a learned dictionary of prototypes conditioned on lightweight task descriptors derived from repertoire probes and pooled embedding statistics. This synthesis produces small adapter modules applied to a frozen pretrained backbone, enabling immediate adaptation to novel tasks with only a handful of support examples and without full model fine-tuning. The architecture preserves interpretability through motif-aware probes and a calibrated motif discovery pipeline that links predictive decisions to sequence-level signals. Together, these components yield a practical, sample-efficient, and interpretable pathway for translating repertoire-informed models into diverse clinical and research settings where labeled data are scarce and computational resources are constrained.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper sketches a conditioning mechanism to synthesize few-shot adapters for immune repertoire models from prototype dictionaries, but the available text supplies no experiments or metrics so the practical payoff stays unverified.

read the letter

The central move is to learn a dictionary of prototypes and then synthesize compact adapter modules on the fly, conditioned only on lightweight descriptors extracted from repertoire probes plus pooled embedding statistics. These adapters plug into a frozen backbone so new tasks can be handled with a handful of examples and no full fine-tuning. The motif-aware probes are meant to keep some link back to sequence-level signals for interpretability. That framing directly targets the label-sparsity and adaptation-cost issues that slow down clinical use of repertoire models. The dynamic-kernel conditioning angle is a concrete design choice that is not just a re-labeling of standard prompt or adapter tuning. It could be useful if the descriptors turn out to carry enough task signal. The obvious gap is that the text gives no numbers, no ablations on descriptor choice or pooling, and no held-out task results. Without those, it is impossible to know whether the synthesized adapters actually match or beat simple baselines on distribution-shifted cohorts or whether the conditioning collapses when motifs vary. The assumption that pooled statistics plus probes are sufficient is stated but not checked. This is the sort of paper that computational immunologists or ML-for-biology groups working on sample-efficient sequence models might want to read once the experiments are in. A reader already thinking about prototype-based or kernel-conditioned adapters could extract the conditioning recipe if it proves sound. I would send it for peer review once the authors add concrete validation on real cohorts and comparisons; as it stands the lack of evidence makes it too early for a full referee process.

Referee Report

2 major / 1 minor

Summary. The paper introduces SwiftRepertoire, a framework for few-shot immune-signature synthesis via dynamic kernel codes. It synthesizes compact task-specific adapter modules from a learned prototype dictionary conditioned on lightweight task descriptors derived from repertoire probes and pooled embedding statistics. These adapters are applied to a frozen pretrained backbone to enable immediate adaptation to novel tasks using only a handful of support examples without full model fine-tuning, while preserving interpretability through motif-aware probes and a calibrated motif discovery pipeline.

Significance. If the central claims hold, the approach could provide a practical, sample-efficient, and interpretable pathway for deploying repertoire-informed models in clinical settings with scarce labels and limited compute, addressing key barriers in T-cell receptor analysis.

major comments (2)

[Abstract and §3] Abstract and §3 (framework description): The claim that lightweight descriptors from repertoire probes and pooled embedding statistics suffice to condition the prototype dictionary for effective task-specific adapters across diverse clinical tasks is load-bearing but unsupported; no ablation on descriptor dimensionality, no analysis of information loss from pooling, and no held-out task evaluations on motif-diverse or distribution-shifted cohorts are provided to verify that the conditioning avoids collapse.
[Results] Results section: No empirical results, validation metrics, implementation details, or comparisons to full fine-tuning baselines are reported, leaving the few-shot adaptation performance and superiority claims without verifiable support.

minor comments (1)

[Abstract] The term 'dynamic kernel codes' is introduced in the title and abstract without a precise definition, equation reference, or section detailing its construction relative to the prototype dictionary.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We agree that the manuscript requires additional empirical support and analyses to substantiate the central claims, and we will incorporate the suggested revisions.

read point-by-point responses

Referee: [Abstract and §3] Abstract and §3 (framework description): The claim that lightweight descriptors from repertoire probes and pooled embedding statistics suffice to condition the prototype dictionary for effective task-specific adapters across diverse clinical tasks is load-bearing but unsupported; no ablation on descriptor dimensionality, no analysis of information loss from pooling, and no held-out task evaluations on motif-diverse or distribution-shifted cohorts are provided to verify that the conditioning avoids collapse.

Authors: We acknowledge that the current manuscript presents the framework at a conceptual level without these supporting analyses. In the revised version we will add ablations varying descriptor dimensionality, quantify information loss from the pooling operation, and report held-out evaluations on motif-diverse and distribution-shifted cohorts to demonstrate that the conditioning mechanism does not collapse. revision: yes
Referee: [Results] Results section: No empirical results, validation metrics, implementation details, or comparisons to full fine-tuning baselines are reported, leaving the few-shot adaptation performance and superiority claims without verifiable support.

Authors: We agree that the absence of empirical results leaves the performance claims unsupported. We will expand the results section with concrete validation metrics, full implementation details, and direct comparisons against full fine-tuning baselines on the few-shot adaptation tasks. revision: yes

Circularity Check

0 steps flagged

Derivation chain self-contained; no reductions to fitted inputs or self-citations

full rationale

The paper presents a synthesis framework that conditions a prototype dictionary on lightweight descriptors (repertoire probes plus pooled embedding statistics) to generate adapter modules for a frozen external pretrained backbone. No equations, derivations, or load-bearing steps are shown that reduce the claimed few-shot performance to self-definitional quantities, fitted parameters renamed as predictions, or self-citation chains. The architecture is described as building on independent pretrained components and motif-aware probes without invoking uniqueness theorems or ansatzes from prior author work. This leaves the central claim of sample-efficient adaptation as an empirical construction rather than a tautological reduction, consistent with a self-contained method.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only review; ledger populated from stated components. The framework assumes a pretrained backbone and learned prototype dictionary exist upstream.

axioms (1)

domain assumption A pretrained encoder backbone exists that captures general repertoire features sufficiently for adapter-based adaptation.
Invoked by the description of applying adapters to a frozen pretrained backbone.

invented entities (1)

Dynamic kernel codes no independent evidence
purpose: To synthesize task-specific parameterizations from the prototype dictionary conditioned on task descriptors.
Introduced as the core synthesis mechanism; no independent evidence provided in abstract.

pith-pipeline@v0.9.0 · 5485 in / 1164 out tokens · 33794 ms · 2026-05-16T09:08:53.125475+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

[1]

T-cell receptor repertoire analysis for the diagnosis and treatment of solid tumor: A methodology and clinical applications.Cancer Communications, 40(10):473–483, 2020

Na Li, Jiani Yuan, Wenjia Tian, Lin Meng, and Yongyu Liu. T-cell receptor repertoire analysis for the diagnosis and treatment of solid tumor: A methodology and clinical applications.Cancer Communications, 40(10):473–483, 2020

work page 2020
[2]

Liquid biopsy approaches to capture tumor evolution and clinical outcomes during cancer immunotherapy.Journal for immunotherapy of cancer, 11 (1):e005924, 2023

Lavanya Sivapalan, Joseph C Murray, Jenna VanLiere Canzoniero, Blair Landon, Jennifer Jackson, Susan Scott, Vincent Lam, Benjamin P Levy, Mark Sausen, and Valsamo Anagnostou. Liquid biopsy approaches to capture tumor evolution and clinical outcomes during cancer immunotherapy.Journal for immunotherapy of cancer, 11 (1):e005924, 2023

work page 2023
[3]

Liquid biopsy in t-cell lymphoma: biomarker detection techniques and clinical application

Zongyao Huang, Yao Fu, Hong Yang, Yehan Zhou, Min Shi, Qingyun Li, Weiping Liu, Junheng Liang, Liuqing Zhu, Sheng Qin, et al. Liquid biopsy in t-cell lymphoma: biomarker detection techniques and clinical application. Molecular Cancer, 23(1):36, 2024

work page 2024
[4]

De novo prediction of cancer-associated t cell receptors for noninvasive cancer detection.Science translational medicine, 12(557):eaaz3738, 2020

Daria Beshnova, Jianfeng Ye, Oreoluwa Onabolu, Benjamin Moon, Wenxin Zheng, Yang-Xin Fu, James Bru- garolas, Jayanthi Lea, and Bo Li. De novo prediction of cancer-associated t cell receptors for noninvasive cancer detection.Science translational medicine, 12(557):eaaz3738, 2020

work page 2020
[5]

T cell receptor convergence is an indicator of antigen-specific t cell response in cancer immunotherapies.Elife, 11:e81952, 2022

Mingyao Pan and Bo Li. T cell receptor convergence is an indicator of antigen-specific t cell response in cancer immunotherapies.Elife, 11:e81952, 2022

work page 2022
[6]

The deep learning framework icantcr enables early cancer detection using the t-cell receptor repertoire in peripheral blood.Cancer Research, 84(11):1915–1928, 2024

Yideng Cai, Meng Luo, Wenyi Yang, Chang Xu, Pingping Wang, Guangfu Xue, Xiyun Jin, Rui Cheng, Jinhao Que, Wenyang Zhou, et al. The deep learning framework icantcr enables early cancer detection using the t-cell receptor repertoire in peripheral blood.Cancer Research, 84(11):1915–1928, 2024

work page 1915
[7]

Xinyang Qian, Guang Yang, Fan Li, Xuanping Zhang, Xiaoyan Zhu, Xin Lai, Xiao Xiao, Tao Wang, and Jiayin Wang. Deeplion2: deep multi-instance contrastive learning framework enhancing the prediction of cancer- associated t cell receptors by attention strategy on motifs.Frontiers in Immunology, 15:1345586, 2024

work page 2024
[8]

Deeplion: deep multi-instance learning improves the prediction of cancer-associated t cell receptors for accurate cancer detection.Frontiers in genetics, 13:860510, 2022

Ying Xu, Xinyang Qian, Xuanping Zhang, Xin Lai, Yuqian Liu, and Jiayin Wang. Deeplion: deep multi-instance learning improves the prediction of cancer-associated t cell receptors for accurate cancer detection.Frontiers in genetics, 13:860510, 2022

work page 2022
[9]

Deeptcr is a deep learning framework for revealing sequence concepts within t-cell repertoires.Nature communications, 12(1):1605, 2021

John-William Sidhom, H Benjamin Larman, Drew M Pardoll, and Alexander S Baras. Deeptcr is a deep learning framework for revealing sequence concepts within t-cell repertoires.Nature communications, 12(1):1605, 2021

work page 2021
[10]

Identification of antigen-specific tcr sequences based on biological and statistical enrichment in unselected individuals.JCI insight, 6(13):e140028, 2021

Neal P Smith, Bert Ruiter, Yamini V Virkud, Ang A Tu, Brinda Monian, James J Moon, J Christopher Love, and Wayne G Shreffler. Identification of antigen-specific tcr sequences based on biological and statistical enrichment in unselected individuals.JCI insight, 6(13):e140028, 2021

work page 2021
[11]

Berttcr: a bert-based deep learning framework for predicting cancer-related immune status based on t cell receptor repertoire.Briefings in Bioinformatics, 25(5):bbae420, 2024

Min Zhang, Qi Cheng, Zhenyu Wei, Jiayu Xu, Shiwei Wu, Nan Xu, Chengkui Zhao, Lei Yu, and Weixing Feng. Berttcr: a bert-based deep learning framework for predicting cancer-related immune status based on t cell receptor repertoire.Briefings in Bioinformatics, 25(5):bbae420, 2024

work page 2024
[12]

Multiple instance neural networks based on sparse attention for cancer detection using t-cell receptor sequences.BMC bioinformatics, 23(1):469, 2022

Younghoon Kim, Tao Wang, Danyi Xiong, Xinlei Wang, and Seongoh Park. Multiple instance neural networks based on sparse attention for cancer detection using t-cell receptor sequences.BMC bioinformatics, 23(1):469, 2022

work page 2022
[13]

Transmil: Transformer based correlated multiple instance learning for whole slide image classification.Advances in neural information processing systems, 34:2136–2147, 2021

Zhuchen Shao, Hao Bian, Yang Chen, Yifeng Wang, Jian Zhang, Xiangyang Ji, et al. Transmil: Transformer based correlated multiple instance learning for whole slide image classification.Advances in neural information processing systems, 34:2136–2147, 2021

work page 2021
[14]

Learning immune receptor representations with protein language models.arXiv preprint arXiv:2402.03823, 2024

Andreas Dounas, Tudor-Stefan Cotet, and Alexander Yermanos. Learning immune receptor representations with protein language models.arXiv preprint arXiv:2402.03823, 2024

work page arXiv 2024
[15]

Deciphering the language of antibodies using self-supervised learning.Patterns, 3(7), 2022

Jinwoo Leem, Laura S Mitchell, James HR Farmery, Justin Barton, and Jacob D Galson. Deciphering the language of antibodies using self-supervised learning.Patterns, 3(7), 2022

work page 2022
[16]

Pan-peptide meta learning for t-cell receptor–antigen binding recognition.Nature Machine Intelligence, 5(3):236–249, 2023

Yicheng Gao, Yuli Gao, Yuxiao Fan, Chengyu Zhu, Zhiting Wei, Chi Zhou, Guohui Chuai, Qinchang Chen, He Zhang, and Qi Liu. Pan-peptide meta learning for t-cell receptor–antigen binding recognition.Nature Machine Intelligence, 5(3):236–249, 2023

work page 2023
[17]

Meta-learning for t cell receptor binding specificity and beyond

Duolin Wang, Fei He, Yang Yu, and Dong Xu. Meta-learning for t cell receptor binding specificity and beyond. Nature machine intelligence, 5(4):337–339, 2023. 14 SwiftRepertoire

work page 2023
[18]

Prototype-based hyperadapter for sample-efficient multi-task tuning.arXiv preprint arXiv:2310.11670, 2023

Hao Zhao, Jie Fu, and Zhaofeng He. Prototype-based hyperadapter for sample-efficient multi-task tuning.arXiv preprint arXiv:2310.11670, 2023

work page arXiv 2023
[19]

Hyperlora: Efficient cross-task generalization via constrained low-rank adapters generation

Chuancheng Lv, Lei Li, Shitou Zhang, Gang Chen, Fanchao Qi, Ningyu Zhang, and Hai-Tao Zheng. Hyperlora: Efficient cross-task generalization via constrained low-rank adapters generation. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 16376–16393, 2024

work page 2024
[20]

Pacia: Parameter-efficient adapter for few-shot molecular property prediction.arXiv preprint arXiv:2310.00614, 2023

Shiguang Wu, Yaqing Wang, and Quanming Yao. Pacia: Parameter-efficient adapter for few-shot molecular property prediction.arXiv preprint arXiv:2310.00614, 2023

work page arXiv 2023
[21]

Spatial heterogeneity of the t cell receptor repertoire reflects the mutational landscape in lung cancer.Nature medicine, 25(10):1549–1559, 2019

Kroopa Joshi, Marc Robert de Massy, Mazlina Ismail, James L Reading, Imran Uddin, Annemarie Woolston, Emine Hatipoglu, Theres Oakes, Rachel Rosenthal, Thomas Peacock, et al. Spatial heterogeneity of the t cell receptor repertoire reflects the mutational landscape in lung cancer.Nature medicine, 25(10):1549–1559, 2019

work page 2019
[22]

Circulating t-cell receptor repertoire for cancer early detection.NPJ Precision Oncology, 9(1):245, 2025

Yilong Li, Michelle Nahas, Dennis Stephens, Kate Froburg, Emma Hintz, Devin Champagne, Amaneet Lochab, Markus Brown, Jasper Braun, María Antonia Fortuño, et al. Circulating t-cell receptor repertoire for cancer early detection.NPJ Precision Oncology, 9(1):245, 2025

work page 2025
[23]

Sequence-based tcr-peptide representations using cross-epitope contrastive fine-tuning of protein language models

Chiho Im, Ryan Zhao, Scott D Boyd, and Anshul Kundaje. Sequence-based tcr-peptide representations using cross-epitope contrastive fine-tuning of protein language models. InInternational Conference on Research in Computational Molecular Biology, pages 34–48. Springer, 2025

work page 2025
[24]

Protein language models expose viral immune mimicry.Viruses, 17(9):1199, 2025

Dan Ofer and Michal Linial. Protein language models expose viral immune mimicry.Viruses, 17(9):1199, 2025

work page 2025
[25]

Contiformer: Continuous- time transformer for irregular time series modeling.Advances in Neural Information Processing Systems, 36: 47143–47175, 2023

Yuqi Chen, Kan Ren, Yansen Wang, Yuchen Fang, Weiwei Sun, and Dongsheng Li. Contiformer: Continuous- time transformer for irregular time series modeling.Advances in Neural Information Processing Systems, 36: 47143–47175, 2023

work page 2023
[26]

Yang Yue, Yihua Cheng, Céline Marquet, Chenguang Xiao, Jingjing Guo, Shu Li, and Shan He. Meta-learning enables complex cluster-specific few-shot binding affinity prediction for protein–protein interactions.Journal of Chemical Information and Modeling, 65(2):580–588, 2025

work page 2025
[27]

Tip-adapter: Training-free adaption of clip for few-shot classification

Renrui Zhang, Wei Zhang, Rongyao Fang, Peng Gao, Kunchang Li, Jifeng Dai, Yu Qiao, and Hongsheng Li. Tip-adapter: Training-free adaption of clip for few-shot classification. InEuropean conference on computer vision, pages 493–510. Springer, 2022

work page 2022
[28]

Methpriorgcn: a deep learning tool for inferring dna methylation prior knowledge and guiding personalized medicine.Briefings in Bioinformatics, 26(2):bbaf131, 2025

Jie Ni, Bin Li, Shumei Miao, Xinting Zhang, Donghui Yan, Shengqi Jing, Shan Lu, Zhuoying Xie, Xin Zhang, and Yun Liu. Methpriorgcn: a deep learning tool for inferring dna methylation prior knowledge and guiding personalized medicine.Briefings in Bioinformatics, 26(2):bbaf131, 2025

work page 2025
[29]

Biformer: Vision transformer with bi-level routing attention

Lei Zhu, Xinjiang Wang, Zhanghan Ke, Wayne Zhang, and Rynson WH Lau. Biformer: Vision transformer with bi-level routing attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10323–10333, 2023. 6 Approximation statement and concise proof Assume the geometric approximation condition dist(θt,S)≤ϵ app,(20) where...

work page 2023

[1] [1]

T-cell receptor repertoire analysis for the diagnosis and treatment of solid tumor: A methodology and clinical applications.Cancer Communications, 40(10):473–483, 2020

Na Li, Jiani Yuan, Wenjia Tian, Lin Meng, and Yongyu Liu. T-cell receptor repertoire analysis for the diagnosis and treatment of solid tumor: A methodology and clinical applications.Cancer Communications, 40(10):473–483, 2020

work page 2020

[2] [2]

Liquid biopsy approaches to capture tumor evolution and clinical outcomes during cancer immunotherapy.Journal for immunotherapy of cancer, 11 (1):e005924, 2023

Lavanya Sivapalan, Joseph C Murray, Jenna VanLiere Canzoniero, Blair Landon, Jennifer Jackson, Susan Scott, Vincent Lam, Benjamin P Levy, Mark Sausen, and Valsamo Anagnostou. Liquid biopsy approaches to capture tumor evolution and clinical outcomes during cancer immunotherapy.Journal for immunotherapy of cancer, 11 (1):e005924, 2023

work page 2023

[3] [3]

Liquid biopsy in t-cell lymphoma: biomarker detection techniques and clinical application

Zongyao Huang, Yao Fu, Hong Yang, Yehan Zhou, Min Shi, Qingyun Li, Weiping Liu, Junheng Liang, Liuqing Zhu, Sheng Qin, et al. Liquid biopsy in t-cell lymphoma: biomarker detection techniques and clinical application. Molecular Cancer, 23(1):36, 2024

work page 2024

[4] [4]

De novo prediction of cancer-associated t cell receptors for noninvasive cancer detection.Science translational medicine, 12(557):eaaz3738, 2020

Daria Beshnova, Jianfeng Ye, Oreoluwa Onabolu, Benjamin Moon, Wenxin Zheng, Yang-Xin Fu, James Bru- garolas, Jayanthi Lea, and Bo Li. De novo prediction of cancer-associated t cell receptors for noninvasive cancer detection.Science translational medicine, 12(557):eaaz3738, 2020

work page 2020

[5] [5]

T cell receptor convergence is an indicator of antigen-specific t cell response in cancer immunotherapies.Elife, 11:e81952, 2022

Mingyao Pan and Bo Li. T cell receptor convergence is an indicator of antigen-specific t cell response in cancer immunotherapies.Elife, 11:e81952, 2022

work page 2022

[6] [6]

The deep learning framework icantcr enables early cancer detection using the t-cell receptor repertoire in peripheral blood.Cancer Research, 84(11):1915–1928, 2024

Yideng Cai, Meng Luo, Wenyi Yang, Chang Xu, Pingping Wang, Guangfu Xue, Xiyun Jin, Rui Cheng, Jinhao Que, Wenyang Zhou, et al. The deep learning framework icantcr enables early cancer detection using the t-cell receptor repertoire in peripheral blood.Cancer Research, 84(11):1915–1928, 2024

work page 1915

[7] [7]

Xinyang Qian, Guang Yang, Fan Li, Xuanping Zhang, Xiaoyan Zhu, Xin Lai, Xiao Xiao, Tao Wang, and Jiayin Wang. Deeplion2: deep multi-instance contrastive learning framework enhancing the prediction of cancer- associated t cell receptors by attention strategy on motifs.Frontiers in Immunology, 15:1345586, 2024

work page 2024

[8] [8]

Deeplion: deep multi-instance learning improves the prediction of cancer-associated t cell receptors for accurate cancer detection.Frontiers in genetics, 13:860510, 2022

Ying Xu, Xinyang Qian, Xuanping Zhang, Xin Lai, Yuqian Liu, and Jiayin Wang. Deeplion: deep multi-instance learning improves the prediction of cancer-associated t cell receptors for accurate cancer detection.Frontiers in genetics, 13:860510, 2022

work page 2022

[9] [9]

Deeptcr is a deep learning framework for revealing sequence concepts within t-cell repertoires.Nature communications, 12(1):1605, 2021

John-William Sidhom, H Benjamin Larman, Drew M Pardoll, and Alexander S Baras. Deeptcr is a deep learning framework for revealing sequence concepts within t-cell repertoires.Nature communications, 12(1):1605, 2021

work page 2021

[10] [10]

Identification of antigen-specific tcr sequences based on biological and statistical enrichment in unselected individuals.JCI insight, 6(13):e140028, 2021

Neal P Smith, Bert Ruiter, Yamini V Virkud, Ang A Tu, Brinda Monian, James J Moon, J Christopher Love, and Wayne G Shreffler. Identification of antigen-specific tcr sequences based on biological and statistical enrichment in unselected individuals.JCI insight, 6(13):e140028, 2021

work page 2021

[11] [11]

Berttcr: a bert-based deep learning framework for predicting cancer-related immune status based on t cell receptor repertoire.Briefings in Bioinformatics, 25(5):bbae420, 2024

Min Zhang, Qi Cheng, Zhenyu Wei, Jiayu Xu, Shiwei Wu, Nan Xu, Chengkui Zhao, Lei Yu, and Weixing Feng. Berttcr: a bert-based deep learning framework for predicting cancer-related immune status based on t cell receptor repertoire.Briefings in Bioinformatics, 25(5):bbae420, 2024

work page 2024

[12] [12]

Multiple instance neural networks based on sparse attention for cancer detection using t-cell receptor sequences.BMC bioinformatics, 23(1):469, 2022

Younghoon Kim, Tao Wang, Danyi Xiong, Xinlei Wang, and Seongoh Park. Multiple instance neural networks based on sparse attention for cancer detection using t-cell receptor sequences.BMC bioinformatics, 23(1):469, 2022

work page 2022

[13] [13]

Transmil: Transformer based correlated multiple instance learning for whole slide image classification.Advances in neural information processing systems, 34:2136–2147, 2021

Zhuchen Shao, Hao Bian, Yang Chen, Yifeng Wang, Jian Zhang, Xiangyang Ji, et al. Transmil: Transformer based correlated multiple instance learning for whole slide image classification.Advances in neural information processing systems, 34:2136–2147, 2021

work page 2021

[14] [14]

Learning immune receptor representations with protein language models.arXiv preprint arXiv:2402.03823, 2024

Andreas Dounas, Tudor-Stefan Cotet, and Alexander Yermanos. Learning immune receptor representations with protein language models.arXiv preprint arXiv:2402.03823, 2024

work page arXiv 2024

[15] [15]

Deciphering the language of antibodies using self-supervised learning.Patterns, 3(7), 2022

Jinwoo Leem, Laura S Mitchell, James HR Farmery, Justin Barton, and Jacob D Galson. Deciphering the language of antibodies using self-supervised learning.Patterns, 3(7), 2022

work page 2022

[16] [16]

Pan-peptide meta learning for t-cell receptor–antigen binding recognition.Nature Machine Intelligence, 5(3):236–249, 2023

Yicheng Gao, Yuli Gao, Yuxiao Fan, Chengyu Zhu, Zhiting Wei, Chi Zhou, Guohui Chuai, Qinchang Chen, He Zhang, and Qi Liu. Pan-peptide meta learning for t-cell receptor–antigen binding recognition.Nature Machine Intelligence, 5(3):236–249, 2023

work page 2023

[17] [17]

Meta-learning for t cell receptor binding specificity and beyond

Duolin Wang, Fei He, Yang Yu, and Dong Xu. Meta-learning for t cell receptor binding specificity and beyond. Nature machine intelligence, 5(4):337–339, 2023. 14 SwiftRepertoire

work page 2023

[18] [18]

Prototype-based hyperadapter for sample-efficient multi-task tuning.arXiv preprint arXiv:2310.11670, 2023

Hao Zhao, Jie Fu, and Zhaofeng He. Prototype-based hyperadapter for sample-efficient multi-task tuning.arXiv preprint arXiv:2310.11670, 2023

work page arXiv 2023

[19] [19]

Hyperlora: Efficient cross-task generalization via constrained low-rank adapters generation

Chuancheng Lv, Lei Li, Shitou Zhang, Gang Chen, Fanchao Qi, Ningyu Zhang, and Hai-Tao Zheng. Hyperlora: Efficient cross-task generalization via constrained low-rank adapters generation. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 16376–16393, 2024

work page 2024

[20] [20]

Pacia: Parameter-efficient adapter for few-shot molecular property prediction.arXiv preprint arXiv:2310.00614, 2023

Shiguang Wu, Yaqing Wang, and Quanming Yao. Pacia: Parameter-efficient adapter for few-shot molecular property prediction.arXiv preprint arXiv:2310.00614, 2023

work page arXiv 2023

[21] [21]

Spatial heterogeneity of the t cell receptor repertoire reflects the mutational landscape in lung cancer.Nature medicine, 25(10):1549–1559, 2019

Kroopa Joshi, Marc Robert de Massy, Mazlina Ismail, James L Reading, Imran Uddin, Annemarie Woolston, Emine Hatipoglu, Theres Oakes, Rachel Rosenthal, Thomas Peacock, et al. Spatial heterogeneity of the t cell receptor repertoire reflects the mutational landscape in lung cancer.Nature medicine, 25(10):1549–1559, 2019

work page 2019

[22] [22]

Circulating t-cell receptor repertoire for cancer early detection.NPJ Precision Oncology, 9(1):245, 2025

Yilong Li, Michelle Nahas, Dennis Stephens, Kate Froburg, Emma Hintz, Devin Champagne, Amaneet Lochab, Markus Brown, Jasper Braun, María Antonia Fortuño, et al. Circulating t-cell receptor repertoire for cancer early detection.NPJ Precision Oncology, 9(1):245, 2025

work page 2025

[23] [23]

Sequence-based tcr-peptide representations using cross-epitope contrastive fine-tuning of protein language models

Chiho Im, Ryan Zhao, Scott D Boyd, and Anshul Kundaje. Sequence-based tcr-peptide representations using cross-epitope contrastive fine-tuning of protein language models. InInternational Conference on Research in Computational Molecular Biology, pages 34–48. Springer, 2025

work page 2025

[24] [24]

Protein language models expose viral immune mimicry.Viruses, 17(9):1199, 2025

Dan Ofer and Michal Linial. Protein language models expose viral immune mimicry.Viruses, 17(9):1199, 2025

work page 2025

[25] [25]

Contiformer: Continuous- time transformer for irregular time series modeling.Advances in Neural Information Processing Systems, 36: 47143–47175, 2023

Yuqi Chen, Kan Ren, Yansen Wang, Yuchen Fang, Weiwei Sun, and Dongsheng Li. Contiformer: Continuous- time transformer for irregular time series modeling.Advances in Neural Information Processing Systems, 36: 47143–47175, 2023

work page 2023

[26] [26]

Yang Yue, Yihua Cheng, Céline Marquet, Chenguang Xiao, Jingjing Guo, Shu Li, and Shan He. Meta-learning enables complex cluster-specific few-shot binding affinity prediction for protein–protein interactions.Journal of Chemical Information and Modeling, 65(2):580–588, 2025

work page 2025

[27] [27]

Tip-adapter: Training-free adaption of clip for few-shot classification

Renrui Zhang, Wei Zhang, Rongyao Fang, Peng Gao, Kunchang Li, Jifeng Dai, Yu Qiao, and Hongsheng Li. Tip-adapter: Training-free adaption of clip for few-shot classification. InEuropean conference on computer vision, pages 493–510. Springer, 2022

work page 2022

[28] [28]

Methpriorgcn: a deep learning tool for inferring dna methylation prior knowledge and guiding personalized medicine.Briefings in Bioinformatics, 26(2):bbaf131, 2025

Jie Ni, Bin Li, Shumei Miao, Xinting Zhang, Donghui Yan, Shengqi Jing, Shan Lu, Zhuoying Xie, Xin Zhang, and Yun Liu. Methpriorgcn: a deep learning tool for inferring dna methylation prior knowledge and guiding personalized medicine.Briefings in Bioinformatics, 26(2):bbaf131, 2025

work page 2025

[29] [29]

Biformer: Vision transformer with bi-level routing attention

Lei Zhu, Xinjiang Wang, Zhanghan Ke, Wayne Zhang, and Rynson WH Lau. Biformer: Vision transformer with bi-level routing attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10323–10333, 2023. 6 Approximation statement and concise proof Assume the geometric approximation condition dist(θt,S)≤ϵ app,(20) where...

work page 2023