Repertoire-Based Diagnostics Using Statistical Biophysics

Abstract

9A fundamental challenge in immunology is diagnostic classification based on repertoire se-10 quence. We used the principle of maximum entropy (MaxEnt) to build compact representations 11 of antibody (IgH) and T-cell receptor (TCRβ) CDR3 repertoires based on the statistical biophysi-12 cal patterns latent in the frequency and ordering of repertoires' constituent amino acids. This 13 approach results in substantial advantages in quality, dimensionality, and training speed com-14 pared to MaxEnt models based solely on the standard 20-letter amino-acid alphabet. De-15 scriptor-based models learn patterns that pure amino-acid-based models cannot. We demon-16 strate the utility of descriptor models by successfully classifying influenza vaccination status 17 (AUC=0.97, p=4×10 -3 ), requiring only 31 samples from 14 individuals. Descriptor-based MaxEnt 18 modeling is a powerful new method for dissecting, encoding, and classifying complex reper-19 toires. 20 3 Arora et al. (2019) Repertoire-Based Diagnostics Using Statistical Biophysics Introduction 21 A major challenge in systems immunology is determining how to describe the sequence-level 22 heterogeneity of antibody (immunoglobulin; Ig) and T-cell receptor (TCR) repertoires in ways 23 that facilitate the identification of meaningful patterns. Sequence-frequency distributions-for 24 example, counts of unique IgH or TCRβ CDR3s-are commonly used but not ideal for interper-25 sonal comparisons, since repertoires from different people are largely disjoint (Robins et al., 26 2010; Arnaout et al., 2011). Motif-frequency distributions, which count how often each of the 20 n 27 possible n-mers appears in a repertoire (for some choice of n), are more likely to overlap be-28 tween individuals, but may fail to detect probabilistic or higher-order patterns and are subject to 29 sampling-related bias unless n is small. Comparisons of frequency distributions between reper-30 toires from different individuals have yielded important insights (Parameswaran et al., 2013; 31 Kaplinsky et al., 2014; Emerson et al., 2017; Sun et al., 2017) but the limitations of this ap-32 proach suggest a need for complementary methods. One such method is maximum-entropy 33 (MaxEnt) modeling (Fig. 1). 34 MaxEnt models, which were first developed for statistical physics and information theory 35 (Jaynes, 1957), can be used to describe repertoires (or other complex ensembles of proteins, 36 nucleic acids, etc.) in terms of constraints called biases that determine the ways in which a giv-37 en repertoire differs from a uniform distribution of sequences (Yeo and Burge, 2004; Russ et al., 38 2005; Seno et al., 2008; Mora et al., 2010; Marks et al., 2011). Given a set of features-for ex-39 ample, the frequencies of the 20 amino acids and the 20 2 =400 nearest-neighbor amino-acid 40 pairs ("neighbors" being defined as contiguous N-to-C-terminus amino acids)-a MaxEnt model 41 describes the degree to which each feature is biased away from its value in a uniform repertoire, 42 taking all the other biases into ac...

A compact vocabulary of paratope-epitope interactions enables predictability of antibody-antigen binding

“…These further investigations will elucidate the question of whether paratope sequence motifs may be used as in silico scanning devices for epitope-specific sequence regions in high-throughput antibody repertoire data for at least a subset of medically important antigens. Understanding the diversity and structure of antibody specificity on the repertoire level is also of importance for enhancing the precision of immune receptor-based immunodiagnostics where one of the most pressing current problems is a low signal-to-noise ratio (few antigen-specific sequences within a large pool of unrelated sequences) (Arora et al, 2019;Brown et al, 2019;Emerson et al, 2017;Ostmeyer et al, 2019).…”

“…Adaptive immune receptor repertoires represent a major target area for the application of machine learning in the hope that it may fast-track the in silico discovery and development of immunereceptor based immunotherapies and immunodiagnostics (Brown et al, 2019;Greiff et al, 2012;Mason et al, 2018Mason et al, , 2019Miho et al, 2018). The complexity of sequence dependencies that determine antigen binding (Dash et al, 2017;Glanville et al, 2017), immune receptor publicity (Greiff et al, 2017b) and immune status (immunodiagnostics) (Ostmeyer et al, 2019;Thomas et al, 2014) represent a perfect application ground for machine learning analysis (Arora et al, 2019;Cinelli et al, 2017;Greiff et al, 2017b;Liu et al, 2019;Mason et al, 2019;Sidhom et al, 2018;Sun et al, 2017). As discussed extensively in a recent literature review by us (Brown et al, 2019) the development of ML approaches for immune receptor datasets was and is still hampered by the lack of ground truth datasets.…”

Computational approaches to therapeutic antibody design: established methods and emerging trends

“…Since the immune system is a dynamic reflection of the overall health of the organism, some antigen-specific signatures in the repertoire could be indicative of particular diseases [ 219 ]. It was demonstrated that statistical classifiers can identify immune profiles of patients with chronic lymphocytic leukemia [ 220 ], multiple sclerosis [ 221 ] or influenza [ 222 ] from NGS data alone. Further development of a larger variety of such models could result in versatile diagnostic tools for multiple conditions on the basis of an individual’s sequenced BCR repertoire [ 220 ].…”

immuneSIM: tunable multi-feature simulation of B- and T-cell receptor repertoires for immunoinformatics benchmarking

“…We validated that immuneSIM can generate immune repertoires that are similar to experimental repertoires (native-like) by evaluating a range of repertoire similarity measures. immuneSIM can also generate aberrant immune receptor repertoires to replicate a broad range of experimental, immunological or disease settings ( Arora et al , 2019 ; Brown et al , 2019 ) ( Supplementary Figs S2 – S7 ).…”