VDJdb in 2026: boosting T-cell receptor recognition evidence using paratope embeddings and AI-based structure prediction.

Abstract

We present a significant update to VDJdb, introducing substantial enhancements to both the data content and the technical infrastructure. The integration of steadily accumulating T-cell receptor (TCR):epitope recognition data, together with advances in high-throughput experimental techniques, has expanded the landscape for machine learning-based prediction of TCR specificity, covering foreign antigens, neoantigens, and autoimmunity-associated epitopes. However, these advances have introduced new challenges, most notably in data quality-large-scale assays have heightened concerns about measurement reliability and exacerbated existing issues such as HLA and epitope coverage biases. To address these issues, we have adopted state-of-the-art artificial intelligence approaches, including protein sequence embeddings and AI-driven structure prediction. The updated VDJdb resource now features TCR specificity records annotated with embedding-derived paratope features, advanced noise filtering capabilities, and predicted three-dimensional protein structures. These improvements enable more comprehensive interrogation of TCR-epitope recognition and provide novel lines of evidence to support the reliability of high-throughput assay records, which are frequently limited by insufficient independent validation. VDJdb can be accessed at https://vdjdb.com and https://github.com/antigenomics/vdjdb-db.