tcr-repertoire-analysis-agent

---

name: 'tcr-repertoire-analysis-agent' description: 'AI-powered T-cell receptor repertoire analysis for cancer diagnosis, immunotherapy response prediction, and therapeutic TCR selection using deep learning and multi-layer ML approaches.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools:

• read_file
• run_shell_command

---

TCR Repertoire Analysis Agent

The TCR Repertoire Analysis Agent provides comprehensive T-cell receptor repertoire analysis for cancer immunology applications. It leverages deep learning and multi-layer machine learning approaches to analyze TCR diversity, predict immunotherapy response, identify tumor-reactive TCRs, and support therapeutic TCR selection for cancer immunotherapy.

When to Use This Skill

• When analyzing TCR repertoire for cancer diagnosis and staging.
• For predicting immunotherapy (anti-PD-1/PD-L1) response from TCR profiles.
• To identify tumor-reactive TCRs for adoptive cell therapy.
• When monitoring treatment response through TCR clonality changes.
• For selecting therapeutic TCRs for TCR-T cell therapy development.

Core Capabilities

1. Repertoire Diversity Analysis: Quantify TCR diversity, clonality, and convergence.

2. Cancer Diagnosis: Distinguish cancer types from TCR signatures.

3. Immunotherapy Response Prediction: Predict checkpoint inhibitor response.

4. Tumor-Reactive TCR Identification: Find neoantigen-specific TCRs.

5. TCR-pMHC Binding Prediction: Predict TCR epitope specificity.

6. Clonal Dynamics Tracking: Monitor TCR clones during treatment.

TCR Repertoire Metrics

Metric	Definition	Clinical Significance
Clonality	Gini coefficient of clone sizes	Immune focusing
Shannon Entropy	Diversity measure	Immune breadth
Richness	Unique clonotypes	Repertoire depth
Top Clone %	Largest clone fraction	Dominant response
Convergent TCRs	Shared across patients	Public epitope response
Tumor-Infiltrating %	TIL-derived TCRs	Tumor reactivity

Workflow

1. Input: TCR-seq data (bulk or single-cell), clinical metadata.

2. Preprocessing: CDR3 extraction, error correction, clustering.

3. Repertoire Analysis: Calculate diversity, clonality, convergence.

4. ML Classification: Cancer type, stage, response prediction.

5. TCR Prioritization: Rank tumor-reactive TCR candidates.

6. TCR-pMHC Prediction: Predict epitope specificity.

7. Output: Repertoire metrics, predictions, therapeutic candidates.

Example Usage

User: "Analyze the TCR repertoire from this melanoma patient's tumor and blood to predict immunotherapy response and identify tumor-reactive TCRs."

Agent Action:

python3 Skills/Immunology_Vaccines/TCR_Repertoire_Analysis_Agent/tcr_repertoire_analysis.py \
    --tumor_tcr tumor_tils.tsv \
    --blood_tcr pbmc_tcrs.tsv \
    --cancer_type melanoma \
    --hla_type HLA-A*02:01,HLA-B*07:02 \
    --neoantigens patient_neoantigens.fasta \
    --task response_prediction,tcr_identification \
    --output tcr_analysis/

Input Formats

Format	Source	Fields
AIRR-seq	Standardized	CDR3, V/J genes, count
MiXCR	MiXCR pipeline	Clone info, counts
10x VDJ	Single-cell	CDR3, cell barcode
Custom TSV	Any pipeline	Flexible mapping

Output Components

Output	Description	Format
Repertoire Metrics	Diversity scores	.json
Response Prediction	Immunotherapy probability	.json
Cancer Classification	Type/stage prediction	.json
Tumor-Reactive TCRs	Ranked candidates	.csv
TCR-pMHC Predictions	Epitope specificity	.csv
Clonal Tracking	Dynamics over time	.csv
Visualizations	Repertoire plots	.png, .pdf

Response Prediction Features

Feature Category	Features	Importance
Diversity	Shannon, Gini, richness	High
Clonality	Top clones, expansion	High
Convergence	Public TCRs, sharing	Moderate
Sequence Features	CDR3 length, motifs	Moderate
TIL Characteristics	TIL fraction, phenotype	High

AI/ML Components

Cancer Classification:

• Multi-layer ensemble (XGBoost, RF, SVM)
• TCR embedding networks
• Attention-based sequence models

Response Prediction:

• Cox regression with TCR features
• Deep survival analysis
• Multi-task learning (response + survival)

TCR-pMHC Prediction:

• AlphaFold3-based structural prediction
• Transformer models (TCR-BERT)
• Contrastive learning embeddings

Clinical Applications

Application	TCR Biomarker	Clinical Utility
Diagnosis	Cancer-specific TCRs	Early detection
Staging	Clonality patterns	Disease extent
Prognosis	Intratumoral diversity	Survival prediction
Response	Baseline clonality	IO response
Monitoring	Clone dynamics	Treatment tracking
Therapy	Tumor-reactive TCRs	TCR-T development

Performance Benchmarks

Task	Dataset	Performance
Cancer vs Normal	Digestive cancers	AUC 0.91
Metastasis Detection	CRC	AUC 0.85
IO Response	Melanoma	AUC 0.78
TCR-pMHC Prediction	IEDB benchmark	AUC 0.82

Prerequisites

• Python 3.10+
• MiXCR, TRUST4 for TCR calling
• immunarch, tcrdist3
• PyTorch, transformers
• AlphaFold3 (optional, for structure)

Related Skills

• TCR_pMHC_Prediction_Agent - Detailed TCR-epitope prediction
• Neoantigen_Prediction_Agent - Neoantigen identification
• TME_Immune_Profiling_Agent - Broader immune context
• TCell_Exhaustion_Analysis_Agent - T cell phenotyping

TCR Sequence Analysis

CDR3 Feature	Analysis	Meaning
Length Distribution	Histogram	V(D)J usage
Amino Acid Usage	Positional frequency	Binding properties
Hydrophobicity	CDR3 profile	MHC interaction
Charge	Net charge	Peptide binding
Motif Enrichment	k-mer analysis	Epitope specificity

Therapeutic TCR Selection Criteria

Criterion	Threshold	Rationale
Tumor Enrichment	>10-fold vs blood	Tumor specificity
Clone Size	Top 1% in tumor	Functional expansion
Neoantigen Binding	Predicted positive	Target specificity
Safety (Cross-react)	No self-peptide hits	Safety
HLA Restriction	Common alleles	Broad applicability

Special Considerations

1. Sample Quality: Fresh samples preferred for TIL analysis
2. Sequencing Depth: Sufficient depth for rare clones
3. Batch Effects: Normalize across sequencing runs
4. HLA Context: TCR analysis requires HLA typing
5. Paired Chains: Single-cell for alpha-beta pairing

Cancer-Specific TCR Signatures

Cancer Type	Key TCR Features	Public TCRs
Melanoma	High clonality, MAA-reactive	Yes
NSCLC	Moderate diversity	Limited
CRC-MSI	Neoantigen-reactive	Variable
HPV+ HNSCC	HPV-E6/E7 reactive	Yes

Author

AI Group - Biomedical AI Platform