Annotating the specificity of TIL products with TCR sequencing: Demonstrating the usefulness of computational algorithms

Our newest peer-reviewed paper in ImmunoInformatics demonstrates how computational TCR epitope annotation can characterise TIL products, screen for viral bystanders, and track tumour-reactive clones in vivo. A great collaboration between ImmuneWatch and the teams of Alexandre Harari and George Coukos from the Ludwig Institute for Cancer Research.

Adoptive cell therapy with tumour-infiltrating lymphocytes — TIL therapy — is one of the most promising frontiers in cancer immunotherapy. The concept is relatively simple on paper: extract T-cells that have already infiltrated a patient's tumour, expand them to billions of cells in the lab, and reinfuse them to mount a powerful anti-tumour attack. The field already has seen successes: consistent 40–50% objective response rates in melanoma, and an FDA-approved TIL product already on the market.

But there is a problem hiding inside every TIL product, and it is one that the field has yet to fully solve: what are these T-cells actually targeting?

The blind spot in TIL manufacturing

A TIL infusion product is a complex, heterogeneous mixture of T-cells that emerged from the tumour microenvironment. Some are reactive against tumour antigens, some against common viral infections, and some with unknown specificity entirely. The manufacturing process amplifies all of them, often more than a thousand-fold.

This matters because not all TILs are created equal. Virus-specific bystander T-cells have been shown to infiltrate tumours in significant numbers, and they can even have higher target affinity than tumour-specific T-cells, causing undesirable skewing during expansion. Once a TIL product has been manufactured, there is currently no standard way to confirm that it actually contains the tumour-reactive T-cells a patient needs. Potency assessment are therefore important to showt the clinical efficacy of TIL therapies.

‍The antigen-specificity of TIL products remains largely unknown. This is a bottleneck.

TCR sequencing meets computational epitope annotation

In a new study published in the peer-reviewed journal ImmunoInformatics, our team at ImmuneWatch, together with collaborators from the Ludwig Institute for Cancer Research in Lausanne (CHUV), demonstrates a proof-of-concept that directly addresses this challenge.

The approach combines two technologies that are individually powerful but transformative together: TCR sequencing of TIL products, and computational epitope annotation using ImmuneWatch DETECT.

TCR sequencing is already routinely performed during TIL manufacturing. Clinical sites and companies developing adoptive cell therapies use it to monitor clonal expansion and product diversity at different stages of production. But until now, the sequencing data has been largely descriptive — telling you what T-cell clones are present, but not what they recognise.

ImmuneWatch DETECT adds that missing layer: it annotates the epitope-specificity of TCRs using machine learning models trained on a curated database of over 2,000 epitope–TCR pairs. It answers the question that TCR sequencing alone cannot: what are these T-cells targeting?

What we found

Working with deeply characterised clinical samples from a melanoma patient treated with TIL therapy in a phase-I clinical trial, we demonstrated three key capabilities.

1. Identifying tumour-reactive T-cell clusters in silico

Clustering of bulk TCR sequencing data from the TIL infusion product revealed a highly oligoclonal population of over 40,000 unique TCRs. When we annotated these with ImmuneWatch DETECT, a clear signal emerged: a dominant cluster of T-cells reactive against the ELAGIGILTV epitope of MelanA, a well-known melanoma-associated antigen. This single cluster represented nearly 10% of the entire TIL repertoire — strong evidence of successful tumour antigen-specific expansion.

2. Screening for viral bystander T-cells

Using the same annotation approach, we screened the TIL product for virus-specific TCRs. Only 0.02% of TCRs in the final product were annotated as virus-specific, compared to 0.36–1.52% in peripheral blood samples and 0.70% in the baseline tumour. This confirmed that the manufacturing process had effectively enriched for tumour-reactive T-cells whilst minimising viral bystander contamination, a critical quality metric.

3. Validating predictions with in vitro experiments

Three TCRs from the MelanA-specific cluster were selected for functional validation. Their paired alpha-beta chains were cloned into a reporter cell line and stimulated with the MelanA peptide. All three showed clear, specific T-cell activation, confirming that the computational predictions were correct.

As a bonus, we demonstrated that the identified MelanA-specific clones could be tracked in the patient's peripheral blood before and after adoptive cell therapy, opening the door to TCR-based immune monitoring.

Why this matters for TIL therapy development

This work demonstrates that combining TCR sequencing with computational epitope annotation can serve as a practical quality control layer for TIL products. Specifically, it enables:

• Tumour-specificity assessment — confirming that the TIL product contains T-cells reactive against known tumour antigens
• Bystander screening — detecting and quantifying virus-specific T-cells that may dilute therapeutic potency
• Immune monitoring — tracking tumour-reactive clones in peripheral blood after treatment to assess persistence and dynamics

These capabilities can be integrated into existing TCR sequencing workflows without additional wet-lab experiments, making them immediately actionable for clinical sites and companies developing adoptive cell therapies.

The road ahead

Current epitope annotation models, including ImmuneWatch DETECT, work best for epitopes where TCR training data already exists, so-called "seen epitope" models. This means the approach is already powerful for well-characterised tumour antigens and viral epitopes, but coverage will continue to expand as the field generates more high-quality TCR–epitope training data through new high-throughput methods.

The challenge of predicting TCR specificity is fundamentally a data problem, not an algorithmic one. As training databases grow, the quality control and immune monitoring applications demonstrated in this paper will become increasingly comprehensive, evolving towards a companion diagnostic for adoptive cell therapy.

Read the full paper

Van Houcke M, Wuyts S, Bosschaerts T, Chiffelle J, Auger A, Coukos G, Harari A, Meysman P. Applications of T-cell receptor specificity annotation models for quality control and immunomonitoring in adoptive T-cell therapies. ImmunoInformatics 22 (2026) 100067.

Interested in annotating the specificity of your TCR repertoires? Learn more about ImmuneWatch DETECT or get in touch with our team.