Systems Immunology: Learning the Rules of the Immune System

Abstract

Given the many cell types and molecular components of the human immune system, along with vast variations across individuals, how should we go about developing causal and predictive explanations of immunity? A central strategy in human studies is to leverage natural variation to find relationships among variables, including DNA variants, epigenetic states, immune phenotypes, clinical descriptors, and others. Here, we focus on how natural variation is used to find patterns, infer principles, and develop predictive models for two areas: (a) immune cell activation-how single-cell profiling boosts our ability to discover immune cell types and states-and (b) antigen presentation and recognition-how models can be generated to predict presentation of antigens on MHC molecules and their detection by T cell receptors. These are two examples of a shift in how we find the drivers and targets of immunity, especially in the human system in the context of health and disease.

Keywords: T cell receptor; antigen presentation; immune cell types; single-cell RNA sequencing; single-cell genomics; systems immunology.

Figure 1

Single-cell experimental tools and datasets needed for generating models of cell types, states, regulatory networks, and disease/therapy signatures. The boxes in the outer circle represent data types (yellow) as well as experimental and analytical frameworks (pink) needed for describing the properties of a cell and for creating the prediction models described in the center, which include classifying cells and states, learning regulatory networks, and identifying predictive signatures of treatment and disease. Abbreviations: BCR, B cell receptor; CRISPR, clustered regularly interspaced short palindromic repeat; TCR, T cell receptor.

Figure 2

Experimental tools and datasets needed to develop predictive models of antigen presentation and TCR recognition. The boxes in the outer circle represent steps in the process of MHC-I antigen presentation and recognition, and the experimental strategies and types of datasets that need to be collected to train predictive models. The inner circle lists some of the analytical methods for learning the underlying rules that govern each step or the integrated process. The box in the center shows the output (or goals) of the predictive models. Abbreviations: ANN, artificial neural networks; GLM, generalized linear models; HMM, hidden Markov models; HT, high throughput; kNN, k-nearest neighbors; MHC-IP, immunoprecipitation of MHC proteins from cells; MS, mass spectrometry; pMHC, peptide-MHC complex; PTMs, posttranslational modifications; SMM, stabilized matrix method; SVM, support vector machines; TAP, transporter associated with antigen processing.