This work will be presented at the AACC Academy Distinguished Abstracts: Innovative Applications for Today and Emerging Discoveries for Tomorrow scientific session (33110) and the Wednesday poster session (B-145) at the 2023 Annual Scientific Meeting.

Immunosubtraction, sometimes referred to as immunotyping (IT), utilizes capillary separation of patient serum incubated with anti-sera against IgG, IgA, IgM, Kappa, or Lambda for identification of monoclonal proteins. Immunosubtraction can offer better workflow efficiency due to more hands-off time for medical laboratory scientists, higher throughput and is 100% digital information. Unlike immunofixation which relies on identification of bands that appear on a physical gel, immunosubtraction requires end users to look for the disappearance of peaks in the digital IT signal relative to the digital signal without any anti-sera. Despite its advantages, there is hesitancy to adopt immunosubtraction due to the more complex interpretation required and the lack of comprehensive training for reviewers.

A recent proof-of-concept study sought to address these concerns through the creation of an R software application for immunotyping data with enhanced signal processing capable of deconvoluting discrete peaks in overlapping backgrounds and automatic quantitation of monoclonal proteins. The software workflow and graphical user interface are outlined in Figure 1. In addition to overlaying raw signals for each capillary anti-sera channel, the software has four additional screens that display pertinent information to the reviewer. The additional signal processing includes visualization of subtracted signals of each anti-sera channel subtracted from the capillary channel without anti-sera, deconvolution of discrete peaks using a damped least squares algorithm and calculation of peak features for automatic matching of heavy and light chains.

Figure 1. Peak Deconvolution For Each Anti-Sera Capillary

Figure 1 Legend: Visualization of Immunotyping Data in R Shiny. Three screens display the raw immunotyping data, subtraction of individual capillaries from the reference channel, and deconvoluted peaks for each antisera capillary. Peak features are used to match heavy and light chains. The area under the curve is used to quantitate the monoclonal protein.

To evaluate the software performance, 1315 IT samples previously interpreted by a pathologist were compared to the software interpretation and quantitation of the monoclonal protein. To evaluate repeatability of quantitation, the process was repeated 10 consecutive times.

Agreement between the software and reviewer interpretation was 90.6% (11,927/13,150) across all deconvolutions and was consistent for 84.7% of samples in all 10 rounds of deconvolution. Mean and median percent CV for consecutive rounds of monoclonal quantitation was 5.9% and 1.16% respectively (protein range 0.02-4.5 g/dL). Deming regression of reviewer vs. software quantitation had a calculated slope of 0.728 and Pearson’s r coefficient of 0.933.

Automatic interpretation and quantitation aside, the enhanced visualizations provided by the software circumvents the need for reviewers to infer the “negative” space between anti-sera capillaries by directly graphing the subtracted differences. Deconvoluting discrete peaks for each anti-sera channel gives reviewers a training and competency tool to better understand the relationship between the original IT signal and individual monoclonal peaks. The application runs in a standard web browser and allows laboratories to leverage the digital nature of IT information, promoting remote interpretation and second consultations without any loss of signal fidelity or the need for gel scanning.

Despite limitations in matching some pathologist interpretations, this study represents a first step in automating immunotyping interpretation and more accurate quantitation of monoclonal proteins even in a polyclonal background. Deconvolution of discrete peaks may help address quantitation of proteins migrating in the beta region, differentiation of therapeutic proteins and improve inter-reviewer consistency. In contrast to recent machine learning algorithms that have been trained for interpretation of gel electrophoresis immunofixation, this enhanced signal processing and peak deconvolution of immunotyping data hopes to open the door for its potential future use as a digital pathology modality for monoclonal protein identification and quantitation.