TULIP: A transformer-based unsupervised language model for interacting peptides and T cell receptors that generalizes to unseen epitopes.

The accurate prediction of binding between T cell receptors (TCR) and their cognate epitopes is key to understanding the adaptive immune response and developing immunotherapies. Current methods face two significant limitations: the shortage of comprehensive high-quality data and the bias introduced by the selection of the negative training data commonly used in the supervised learning approaches. We propose a method, Transformer-based Unsupervised Language model for Interacting Peptides and T cell receptors (TULIP), that addresses both limitations by leveraging incomplete data and unsupervised learning and using the transformer architecture of language models. Our model is flexible and integrates all possible data sources, regardless of their quality or completeness. We demonstrate the existence of a bias introduced by the sampling procedure used in previous supervised approaches, emphasizing the need for an unsupervised approach. TULIP recognizes the specific TCRs binding an epitope, performing well on unseen epitopes. Our model outperforms state-of-the-art models and offers a promising direction for the development of more accurate TCR epitope recognition models.

DRAG in situ barcoding reveals an increased number of HSPCs contributing to myelopoiesis with age.

Ageing is associated with changes in the cellular composition of the immune system. During ageing, hematopoietic stem and progenitor cells (HSPCs) that produce immune cells are thought to decline in their regenerative capacity. However, HSPC function has been mostly assessed using transplantation assays, and it remains unclear how HSPCs age in the native bone marrow niche. To address this issue, we present an in situ single cell lineage tracing technology to quantify the clonal composition and cell production of single cells in their native niche. Our results demonstrate that a pool of HSPCs with unequal output maintains myelopoiesis through overlapping waves of cell production throughout adult life. During ageing, the increased frequency of myeloid cells is explained by greater numbers of HSPCs contributing to myelopoiesis rather than the increased myeloid output of individual HSPCs. Strikingly, the myeloid output of HSPCs remains constant over time despite accumulating significant transcriptomic changes throughout adulthood. Together, these results show that, unlike emergency myelopoiesis post-transplantation, aged HSPCs in their native microenvironment do not functionally decline in their regenerative capacity.

Inspecting the interaction between human immunodeficiency virus and the immune system through genetic turnover.

Chronic infections of the human immunodeficiency virus (HIV) create a very complex coevolutionary process, where the virus tries to escape the continuously adapting host immune system. Quantitative details of this process are largely unknown and could help in disease treatment and vaccine development. Here we study a longitudinal dataset of ten HIV-infected people, where both the B-cell receptors and the virus are deeply sequenced. We focus on simple measures of turnover, which quantify how much the composition of the viral strains and the immune repertoire change between time points. At the single-patient level, the viral-host turnover rates do not show any statistically significant correlation, however, they correlate if one increases the amount of statistics by aggregating the information across patients. We identify an anti-correlation: large changes in the viral pool composition come with small changes in the B-cell receptor repertoire. This result seems to contradict the naïve expectation that when the virus mutates quickly, the immune repertoire needs to change to keep up. However, a simple model of antagonistically evolving populations can explain this signal. If it is sampled at intervals comparable with the sweep time, one population has had time to sweep while the second cannot start a counter-sweep, leading to the observed anti-correlation. This article is part of the theme issue 'Interdisciplinary approaches to predicting evolutionary biology'.

Neoantigen quality predicts immunoediting in survivors of pancreatic cancer.

Cancer immunoediting1 is a hallmark of cancer2 that predicts that lymphocytes kill more immunogenic cancer cells to cause less immunogenic clones to dominate a population. Although proven in mice1,3, whether immunoediting occurs naturally in human cancers remains unclear. Here, to address this, we investigate how 70 human pancreatic cancers evolved over 10 years. We find that, despite having more time to accumulate mutations, rare long-term survivors of pancreatic cancer who have stronger T cell activity in primary tumours develop genetically less heterogeneous recurrent tumours with fewer immunogenic mutations (neoantigens). To quantify whether immunoediting underlies these observations, we infer that a neoantigen is immunogenic (high-quality) by two features-'non-selfness'  based on neoantigen similarity to known antigens4,5, and 'selfness'  based on the antigenic distance required for a neoantigen to differentially bind to the MHC or activate a T cell compared with its wild-type peptide. Using these features, we estimate cancer clone fitness as the aggregate cost of T cells recognizing high-quality neoantigens offset by gains from oncogenic mutations. With this model, we predict the clonal evolution of tumours to reveal that long-term survivors of pancreatic cancer develop recurrent tumours with fewer high-quality neoantigens. Thus, we submit evidence that that the human immune system naturally edits neoantigens. Furthermore, we present a model to predict how immune pressure induces cancer cell populations to evolve over time. More broadly, our results argue that the immune system fundamentally surveils host genetic changes to suppress cancer.

Probing T-cell response by sequence-based probabilistic modeling.

With the increasing ability to use high-throughput next-generation sequencing to quantify the diversity of the human T cell receptor (TCR) repertoire, the ability to use TCR sequences to infer antigen-specificity could greatly aid potential diagnostics and therapeutics. Here, we use a machine-learning approach known as Restricted Boltzmann Machine to develop a sequence-based inference approach to identify antigen-specific TCRs. Our approach combines probabilistic models of TCR sequences with clone abundance information to extract TCR sequence motifs central to an antigen-specific response. We use this model to identify patient personalized TCR motifs that respond to individual tumor and infectious disease antigens, and to accurately discriminate specific from non-specific responses. Furthermore, the hidden structure of the model results in an interpretable representation space where TCRs responding to the same antigen cluster, correctly discriminating the response of TCR to different viral epitopes. The model can be used to identify condition specific responding TCRs. We focus on the examples of TCRs reactive to candidate neoantigens and selected epitopes in experiments of stimulated TCR clone expansion.

Single molecule microscopy reveals key physical features of repair foci in living cells.

In response to double strand breaks (DSB), repair proteins accumulate at damaged sites, forming membrane-less sub-compartments or foci. Here we explored the physical nature of these foci, using single molecule microscopy in living cells. Rad52, the functional homolog of BRCA2 in yeast, accumulates at DSB sites and diffuses ~6 times faster within repair foci than the focus itself, exhibiting confined motion. The Rad52 confinement radius coincides with the focus size: foci resulting from 2 DSBs are twice larger in volume that the ones induced by a unique DSB and the Rad52 confinement radius scales accordingly. In contrast, molecules of the single strand binding protein Rfa1 follow anomalous diffusion similar to the focus itself or damaged chromatin. We conclude that while most Rfa1 molecules are bound to the ssDNA, Rad52 molecules are free to explore the entire focus reflecting the existence of a liquid droplet around damaged DNA.

Contribution of resident and circulating precursors to tumor-infiltrating CD8+ T cell populations in lung cancer.

Tumor-infiltrating lymphocytes (TILs), in general, and especially CD8+ TILs, represent a favorable prognostic factor in non-small cell lung cancer (NSCLC). The tissue origin, regenerative capacities, and differentiation pathways of TIL subpopulations remain poorly understood. Using a combination of single-cell RNA and T cell receptor (TCR) sequencing, we investigate the functional organization of TIL populations in primary NSCLC. We identify two CD8+ TIL subpopulations expressing memory-like gene modules: one is also present in blood (circulating precursors) and the other one in juxtatumor tissue (tissue-resident precursors). In tumors, these two precursor populations converge through a unique transitional state into terminally differentiated cells, often referred to as dysfunctional or exhausted. Differentiation is associated with TCR expansion, and transition from precursor to late-differentiated states correlates with intratumor T cell cycling. These results provide a coherent working model for TIL origin, ontogeny, and functional organization in primary NSCLC.

RBM-MHC: A Semi-Supervised Machine-Learning Method for Sample-Specific Prediction of Antigen Presentation by HLA-I Alleles.

The recent increase of immunopeptidomics data, obtained by mass spectrometry or binding assays, opens up possibilities for investigating endogenous antigen presentation by the highly polymorphic human leukocyte antigen class I (HLA-I) protein. State-of-the-art methods predict with high accuracy presentation by HLA alleles that are well represented in databases at the time of release but have a poorer performance for rarer and less characterized alleles. Here, we introduce a method based on Restricted Boltzmann Machines (RBMs) for prediction of antigens presented on the Major Histocompatibility Complex (MHC) encoded by HLA genes-RBM-MHC. RBM-MHC can be trained on custom and newly available samples with no or a small amount of HLA annotations. RBM-MHC ensures improved predictions for rare alleles and matches state-of-the-art performance for well-characterized alleles while being less data demanding. RBM-MHC is shown to be a flexible and easily interpretable method that can be used as a predictor of cancer neoantigens and viral epitopes, as a tool for feature discovery, and to reconstruct peptide motifs presented on specific HLA molecules.

Generative models of T-cell receptor sequences.

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 autoencoder 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.

Exploiting B Cell Receptor Analyses to Inform on HIV-1 Vaccination Strategies.

The human antibody repertoire is generated by the recombination of different gene segments as well as by processes of somatic mutation. Together these mechanisms result in a tremendous diversity of antibodies that are able to combat various pathogens including viruses and bacteria, or malignant cells. In this review, we summarize the opportunities and challenges that are associated with the analyses of the B cell receptor repertoire and the antigen-specific B cell response. We will discuss how recent advances have increased our understanding of the antibody response and how repertoire analyses can be exploited to inform on vaccine strategies, particularly against HIV-1.

Detecting T cell receptors involved in immune responses from single repertoire snapshots.

Hypervariable T cell receptors (TCRs) play a key role in adaptive immunity, recognizing a vast diversity of pathogen-derived antigens. Our ability to extract clinically relevant information from large high-throughput sequencing of TCR repertoires (RepSeq) data is limited, because little is known about TCR-disease associations. We present Antigen-specific Lymphocyte Identification by Clustering of Expanded sequences (ALICE), a statistical approach that identifies TCR sequences actively involved in current immune responses from a single RepSeq sample and apply it to repertoires of patients with a variety of disorders - patients with autoimmune disease (ankylosing spondylitis [AS]), under cancer immunotherapy, or subject to an acute infection (live yellow fever [YF] vaccine). We validate the method with independent assays. ALICE requires no longitudinal data collection nor large cohorts, and it is directly applicable to most RepSeq datasets. Its results facilitate the identification of TCR variants associated with diseases and conditions, which can be used for diagnostics and rational vaccine design.

Evidence for Shaping of Light Chain Repertoire by Structural Selection.

The naïve immunoglobulin (IG) repertoire in the blood differs from the direct output of the rearrangement process. These differences stem from selection that affects the germline gene usage and the junctional nucleotides. A major complication obscuring the details of the selection mechanism in the heavy chain is the failure to properly identify the D germline and determine the nucleotide addition and deletion in the junction region. The selection affecting junctional diversity can, however, be studied in the light chain that has no D gene. We use probabilistic and deterministic models to infer and disentangle generation and selection of the light chain, using large samples of light chains sequenced from healthy donors and transgenic mice. We have previously used similar models for the beta chain of T-cell receptors and the heavy chain of IGs. Selection is observed mainly in the CDR3. The CDR3 length and mass distributions are narrower after selection than before, indicating stabilizing selection for mid-range values. Within the CDR3, proline and cysteine undergo negative selection, while glycine undergoes positive selection. The results presented here suggest structural selection maintaining the size of the CDR3 within a limited range, and preventing turns in the CDR3 region.

A model for the integration of conflicting exogenous and endogenous signals by dendritic cells.

Cells of the immune system are confronted with opposing pro- and anti-inflammatory signals. Dendritic cells (DC) integrate these cues to make informed decisions whether to initiate an immune response. Confronted with exogenous microbial stimuli, DC endogenously produce both anti- (IL-10) and pro-inflammatory (TNFα) cues whose joint integration controls the cell's final decision. Backed by experimental measurements we present a theoretical model to quantitatively describe the integration mode of these opposing signals. We propose a two step integration model that modulates the effect of the two types of signals: an initial bottleneck integrates both signals (IL-10 and TNFα), the output of which is later modulated by the anti-inflammatory signal. We show that the anti-inflammatory IL-10 signaling is long ranged, as opposed to the short-ranged pro-inflammatory TNFα signaling. The model suggests that the population averaging and modulation of the pro-inflammatory response by the anti-inflammatory signal is a safety guard against excessive immune responses.

Fluctuating fitness shapes the clone-size distribution of immune repertoires.

The adaptive immune system relies on the diversity of receptors expressed on the surface of B- and T cells to protect the organism from a vast amount of pathogenic threats. The proliferation and degradation dynamics of different cell types (B cells, T cells, naive, memory) is governed by a variety of antigenic and environmental signals, yet the observed clone sizes follow a universal power-law distribution. Guided by this reproducibility we propose effective models of somatic evolution where cell fate depends on an effective fitness. This fitness is determined by growth factors acting either on clones of cells with the same receptor responding to specific antigens, or directly on single cells with no regard for clones. We identify fluctuations in the fitness acting specifically on clones as the essential ingredient leading to the observed distributions. Combining our models with experiments, we characterize the scale of fluctuations in antigenic environments and we provide tools to identify the relevant growth signals in different tissues and organisms. Our results generalize to any evolving population in a fluctuating environment.