Rapid T-cell receptor interaction grouping with ting

Felix Mölder; Ulrik Stervbo; Lucie Loyal; Petra Bacher; Nina Babel; Sven Rahmann

doi:10.1093/bioinformatics/btab361

Rapid T-cell receptor interaction grouping with ting

Bioinformatics. 2021 Oct 25;37(20):3444-3448. doi: 10.1093/bioinformatics/btab361.

Authors

Felix Mölder^{1

2}, Ulrik Stervbo³, Lucie Loyal^{4

5

6}, Petra Bacher^{7

8}, Nina Babel^{3

5

6}, Sven Rahmann^{1

9}

Affiliations

¹ Genome Informatics, Institute of Human Genetics, University of Duisburg-Essen, 45147 Essen, Germany.
² Institute of Pathology, University of Duisburg-Essen, 45147 Essen, Germany.
³ Center for Translational Medicine, Medical Clinic I, Marien Hospital Herne, Ruhr-University Bochum, 44623 Herne, Germany.
⁴ Si-M/"Der Simulierte Mensch" a Science Framework of Technische Universität Berlin and Charité - Universitätsmedizin Berlin, 13353 Berlin, Germany.
⁵ Berlin Institute of Health, Berlin-Brandenburg Center for Regenerative Therapies, 13353 Berlin, Germany.
⁶ Charité - Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, 13353 Berlin, Germany.
⁷ Institute of Immunology, Christian-Albrechts Universität zu Kiel and Universitätsklinik Schleswig-Holstein, 24105 Kiel, Germany.
⁸ Institute of Clinical Molecular Biology, Christian-Albrechts Universität zu Kiel, 24118 Kiel, Germany.
⁹ Algorithmic Bioinformatics, Center for Bioinformatics and Department of Computer Science, Saarland University, 66123 Saarbrücken, Germany.

PMID: 33983394
DOI: 10.1093/bioinformatics/btab361

Abstract

Motivation: Clustering T-cell receptor repertoire (TCRR) sequences according to antigen specificity is challenging. The previously published tool GLIPH needs several days to weeks for clustering large repertoires, making its use impractical in larger studies. In addition, the methodology used in GLIPH suffers from shortcomings, including non-determinism, potential loss of significant antigen-specific sequences or inclusion of too many unspecific sequences.

Results: We present an algorithm for clustering TCRR sequences that scales efficiently to large repertoires. We clustered 36 real datasets with up to 62 000 unique CDR3β sequences using both an implementation of our method called ting, GLIPH and its successor GLIPH2. While GLIPH required multiple weeks, ting only needed about one minute for the same task. GLIPH2 is comparably fast, but uses a different grouping paradigm. In addition, we found that in naïve repertoires, where no or very few antigen-specific CDR3 sequences or clusters should exist, our method indeed selects much fewer motifs and produces smaller clusters.

Availability and implementation: Our method has been implemented in Python as a tool called ting. It is available from GitHub (https://github.com/FelixMoelder/ting) or PyPI under the MIT license.

Supplementary information: Supplementary data are available at Bioinformatics online.

Grants and funding

EFRE-0800427/EFRE.NRW program OsteoSys