Maxim Zaslavsky

In single-cell datasets, patient labels indicating disease status (e.g., “sick” or “not sick”) are typically available, but individual cell labels indicating which of a patient’s cells are associated with their disease state are generally unknown. To address this, we introduce mixture modeling for multiple-instance learning (MMIL), an expectation-maximization approach that trains cell-level binary classifiers using only patient-level labels. Applied to primary samples from patients with acute leukemia, MMIL accurately separates leukemia from nonleukemia baseline cells, including rare minimal residual disease (MRD) cells; generalizes across tissues and treatment time points; and identifies biologically relevant features with accuracy approaching that of a hematopathologist. MMIL can also incorporate cell labels when they are available, creating a robust framework for leveraging both labeled and unlabeled cells. MMIL provides a flexible modeling framework for cell classification, especially in scenarios with unknown gold-standard cell labels.

Clinical diagnosis typically incorporates physical examination, patient history, various laboratory tests, and imaging studies but makes limited use of the human immune system’s own record of antigen exposures encoded by receptors on B cells and T cells. We analyzed immune receptor datasets from 593 individuals to develop MAchine Learning for Immunological Diagnosis, an interpretive framework to screen for multiple illnesses simultaneously or precisely test for one condition. This approach detects specific infections, autoimmune disorders, vaccine responses, and disease severity differences. Human-interpretable features of the model recapitulate known immune responses to severe acute respiratory syndrome coronavirus 2, influenza, and human immunodeficiency virus, highlight antigen-specific receptors, and reveal distinct characteristics of systemic lupus erythematosus and type-1 diabetes autoreactivity. This analysis framework has broad potential for scientific and clinical interpretation of immune responses.

Clinical diagnosis typically incorporates physical examination, patient history, and various laboratory tests and imaging studies, but makes limited use of the human system’s own record of antigen exposures encoded by receptors on B cells and T cells. We analyzed immune receptor datasets from 593 individuals to develop MAchine Learning for Immunological Diagnosis (Mal-ID) , an interpretive framework to screen for multiple illnesses simultaneously or precisely test for one condition. This approach detects specific infections, autoimmune disorders, vaccine responses, and disease severity differences. Human-interpretable features of the model recapitulate known immune responses to SARS-CoV-2, Influenza, and HIV, highlight antigen-specific receptors, and reveal distinct characteristics of Systemic Lupus Erythematosus and Type-1 Diabetes autoreactivity. This analysis framework has broad potential for scientific and clinical interpretation of human immune responses.

In this work, we find that CD8 ⁺ T cells expressing inhibitory killer cell immunoglobulin-like receptors (KIRs) are the human equivalent of Ly49 ⁺ CD8 ⁺ regulatory T cells in mice and are increased in the blood and inflamed tissues of patients with a variety of autoimmune diseases. Moreover, these CD8 ⁺ T cells efficiently eliminated pathogenic gliadin-specific CD4 ⁺ T cells from the leukocytes of celiac disease patients in vitro. We also find elevated levels of KIR ⁺ CD8 ⁺ T cells, but not CD4 ⁺ regulatory T cells, in COVID-19 patients, correlating with disease severity and vasculitis. Selective ablation of Ly49 ⁺ CD8 ⁺ T cells in virus-infected mice led to autoimmunity after infection. Our results indicate that in both species, these regulatory CD8 ⁺ T cells act specifically to suppress pathogenic T cells in autoimmune and infectious diseases.

Previous reports show that Ly49 ⁺ CD8 ⁺ T cells can suppress autoimmunity in mouse models of autoimmune diseases. Here we find a markedly increased frequency of CD8 ⁺ T cells expressing inhibitory Killer cell Immunoglobulin like Receptors (KIR), the human equivalent of the Ly49 family, in the blood and inflamed tissues of various autoimmune diseases. Moreover, KIR ⁺ CD8 ⁺ T cells can efficiently eliminate pathogenic gliadin-specific CD4 ⁺ T cells from Celiac disease (CeD) patients’ leukocytes in vitro . Furthermore, we observe elevated levels of KIR ⁺ CD8 ⁺ T cells, but not CD4 ⁺ regulatory T cells, in COVID-19 and influenza-infected patients, and this correlates with disease severity and vasculitis in COVID-19. Expanded KIR ⁺ CD8 ⁺ T cells from these different diseases display shared phenotypes and similar T cell receptor sequences. These results characterize a regulatory CD8 ⁺ T cell subset in humans, broadly active in both autoimmune and infectious diseases, which we hypothesize functions to control self-reactive or otherwise pathogenic T cells. One-Sentence Summary: Here we identified KIR ⁺ CD8 ⁺ T cells as a regulatory CD8 ⁺ T cell subset in humans that suppresses self-reactive or otherwise pathogenic CD4 ⁺ T cells.

Robust quantification of immune cell infiltration into the tumor microenvironment may shed light on why only a small proportion of patients benefit from checkpoint therapy. The immune cells surrounding a tumor have been suggested to mediate an effective response to immunotherapy. However, traditional measurement of immune cell content around a tumor by immunohistochemistry, flow cytometry, or mass cytometry allows measurement of only up to a few dozen markers at a time, limiting the number of immune cell types identified. Immune cell type abundances may instead be estimated in silico by deconvolving gene expression mixtures from bulk RNA sequencing of tumor tissue. By measuring tens of thousands of transcripts at once, bulk RNA-seq provides a rich input to algorithms that quantify cell type abundances in the tumor microenvironment, affording the potential to quantify the states of a greater number of immune cell types (given adequate training data). Here, we first review existing methods for deconvolution and evaluate their performance on synthetic mixtures. Then we develop a Bayesian inference approach, named infino , that learns to distinguish immune cell expression phenotypes and deconvolve mixtures. In contrast to earlier approaches, infino accepts RNA sequencing data, models transcript expression variability, and exploits the relationships between cell types to improve deconvolution accuracy and allow interrogation from the level of broad categories to the level of finest granularity. The resulting probability distributions of immune infiltration could be applied to numerous questions concerning the diverse ecology of immune cell types, including assessment of the association of immune infiltration with response to immunotherapy, and study of the expression profile and presence of elusive T cell subcompartments, such as T cell exhaustion.