Abstract
Autoimmune diseases are characterized by highly specific immune responses against molecules in self-tissues. Different autoimmune diseases are characterized by distinct immune responses, making autoantibodies useful for diagnosis and prediction. In many diseases, the targets of autoantibodies are incompletely defined. Although the technologies for autoantibody discovery have advanced dramatically over the past decade, each of these techniques generates hundreds of possibilities, which are onerous and expensive to validate. We set out to establish a method to greatly simplify autoantibody discovery, using a pre-filtering step to define subgroups with similar specificities based on migration of radiolabeled, immunoprecipitated proteins on sodium dodecyl sulfate (SDS) gels and autoradiography [Gel Electrophoresis and band detection on Autoradiograms (GEA)]. Human recognition of patterns is not optimal when the patterns are complex or scattered across many samples. Multiple sources of errors - including irrelevant intensity differences and warping of gels - have challenged automation of pattern discovery from autoradiograms. In this paper, we address these limitations using a Bayesian hierarchical model with shrinkage priors for pattern alignment and spatial dewarping. The Bayesian model combines information from multiple gel sets and corrects spatial warping for coherent estimation of autoantibody signatures defined by presence or absence of a grid of landmark proteins. We show the pre-processing creates more clearly separated clusters and improves the accuracy of autoantibody subset detection via hierarchical clustering. Finally, we demonstrate the utility of the proposed methods with GEA data from scleroderma patients.
Introduction
In autoimmune diseases, the body’s immune system mistakenly attacks self-tissues, producing specific proteins called autoantibodies. These antibodies serve as "molecular fingerprints" that can help clinicians diagnose specific disease subsets and predict patient outcomes. A common tool for identifying these fingerprints is Gel Electrophoresis and Autoradiography (GEA). In GEA, proteins are separated by weight on a gel, creating a pattern of bands that represent the unique immune signature of a patient.
However, GEA is notoriously difficult to analyze at scale. Gels frequently warp due to heat, proteins migrate at slightly different speeds across different experiments (batch effects), and irrelevant intensity differences can obscure real biological patterns. Traditionally, experts have attempted to read these patterns by eye, but as datasets grow, human recognition fails to scale and lacks reproducibility. This research addresses these challenges by automating the "cleaning" and alignment of GEA data, enabling standard clustering algorithms to accurately identify patient subgroups.
Methodology
The researchers developed a two-stage statistical pipeline designed to pre-process GEA data and correct for physical distortions.
1. Peak Detection
To handle variations in brightness and background noise, the authors implemented a "local difference scoring" algorithm. Instead of analyzing absolute intensity, this method evaluates the shape of peaks relative to their immediate neighbors. This approach ensures the model is robust against overall brightness differences and prevents high-intensity "blobs" from drowning out smaller, clinically relevant peaks.
2. Bayesian Spatial Dewarping
The core innovation of the paper is a Bayesian hierarchical model that treats physical gel warping as a mathematical function to be "undone." * Reference Alignment: The model first uses known "landmark" proteins (markers of known weight) to perform a rough piecewise linear alignment. * Hierarchical Bayesian Model: Cubic B-splines are used to model non-linear warping across the gel. By employing shrinkage priors, the model "borrows strength" across different lanes. For example, if a common protein like actin appears in many samples, the model uses its position to help align all other lanes more accurately. * Uncertainty Quantification: Unlike standard "best-fit" methods, the Bayesian approach provides a posterior distribution, offering a measure of confidence in the alignment.
Experimental Evaluation
The methods were evaluated using two primary datasets involving patients with Scleroderma, a chronic connective tissue disease:
- Replication Dataset: This included 20 biological samples, each run twice (long vs. short exposure) to create 40 lanes. This served as a "ground truth" to test if the model could correctly pair replicates despite experimental noise.
- Discovery Dataset: This included 76 scleroderma patients with unknown clustering, used to demonstrate the model's ability to find scientifically meaningful subgroups in a real-world clinical setting.
Key Results
The application of the Bayesian dewarping model yielded significant improvements in data quality and clinical discovery:
- Improved Clustering Accuracy: In the replication study, pre-processing significantly increased the Adjusted Rand Index (aRI), demonstrating that the algorithm was far superior at identifying replicate pairs compared to using raw, unaligned data.
- Increased Cluster Separation: The "average silhouette"—a metric for how tightly grouped and well-separated clusters are—increased by 14% to 118% after dewarping.
- Clinical Discovery: In the 76-patient discovery study, the model identified a specific cluster of seven patients with a unique signature (enriched at 32.7 and 27.9 kDa). This subgroup was completely missed or fragmented by standard analysis methods that did not account for gel warping.
Discussion and Future Work
While the model represents a significant step forward in automating autoantibody discovery, the authors note several areas for future refinement. Currently, the peak detection relies on a manually set tuning parameter for locality, which could be automated. Furthermore, the model treats protein bands as independent, whereas proteins often function in complexes. Future iterations may utilize a "Bayesian factorization" approach to identify groups of bands that consistently appear together. Finally, incorporating patient metadata—such as age or cancer status—directly into the alignment process could further refine the discovery of disease-specific signatures.
Summary updated