On generative models of T-cell receptor sequences

Abstract

T-cell receptors (TCR) are key proteins of the adaptive immune system, generated randomly in each individual, whose diversity underlies our ability to recognize infections and malignancies. Modeling the distribution of TCR sequences is of key importance for immunology and medical applications. Here, we compare two inference methods trained on high-throughput sequencing data: a knowledge-guided approach, which accounts for the details of sequence generation, supplemented by a physics-inspired model of selection; and a knowledge-free Variational Auto-Encoder based on deep artificial neural networks. We show that the knowledge-guided model outperforms the deep network approach at predicting TCR probabilities, while being more interpretable, at a lower computational cost.

Haematopoietic Stem Cell Transplantation Results in Extensive Remodelling of the Clonal T Cell Repertoire in Multiple Sclerosis

“…Stochastically, more-readily generated clones are more likely to be public. CD4+ and CD8+ clones were subjected to an analysis utilising a novel modelling tool entitled SONIA to obtain values for P GEN (probability that the sequence is generated by V-D-J recombination) and P POST (probability of the clonotype surviving thymic selection); with lower values indicating a ‘rare’ clone ( 39 , 56 , 57 ).…”

“…The AHSCT baseline CD4 (or CD8) was then used as the reference dataset to analyse the 6- and 12-month timepoints adjusting minimum CDR3 length to 7 and kmers (AA sequences of k length) from 3 to 7 and results were filtered for Fisher’s score <0.001. SONIA ( 39 , 56 , 57 ) was used to infer thymic selection pressures on features of amino acid CDR3 sequences. Following validation in hundreds of thousands of human αβ TCRs, SONIA takes as input TCR CDR3 AA sequences along with V and J genes.…”

Dynamics of B-cell repertoires and emergence of cross-reactive responses in COVID-19 patients with different disease severity

“…We characterized the probability to observe a clonal lineage ancestor in the periphery as 𝑃 post (σ) ∼ 𝑃 gen (σ)𝑒 ' (:features * ( (+) , which deviates from the inferred generation probability of the receptor 𝑃 gen (σ) by selection factors 𝑞 . (𝜎) (Isacchini et al, 2020a;Isacchini et al, 2020b;Sethna et al, 2020) . These selection factors 𝑞 .…”

“…These selection factors 𝑞 . (𝜎) depend on sequence features, including IGHV-gene and IGHJ-gene usages, HCDR3 length, and amino acid preferences at different positions in the HCDR3 (Methods) (Elhanati et al, 2014;Isacchini et al, 2020a;Isacchini et al, 2020b;Marcou et al, 2018;Sethna et al, 2020).…”

“…We inferred a model (Marcou et al, 2018) to characterize a null probability for generation of receptors gen ( ), trained on the set of unproductive inferred naïve BCRs. We inferred a selection model (Isacchini et al, 2020b) to characterize the deviation from the null among inferred naïve productive BCRs using the probability of entry to the periphery post ( ) and characterize selection factors . ( ) dependent on sequence features to differentiate between cohort statistics.…”