An Approach to Identify HLA Class II Immunogenic Epitopes in the Greek Population through Machine Learning Algorithms.

Current pre-transplantation routine matching involves serum anti-HLA antibodies quantification but cannot always preclude unfavorable graft outcomes. Epitope-based matching is proposed as a more precise approach, but to date no epitope-matching algorithm provides a satisfactory predictive tool for transplantation outcomes. In this study, anti-HLA-II loci responses from 1748 patients were analyzed with unsupervised machine learning algorithms, namely principal component analysis (PCA) and antigenic distances, projected as dendrograms. PCA for anti-HLA-DR anti-bodies revealed three main clusters of responses: anti-HLA-DR51 combined with anti-HLA-DRB1*01, anti-HLA-DR52 combined with anti-HLA-DRB1*08 and anti-HLA-DR53 combined with anti-HLA-DRB1*10. The dendrogram for anti-HLA-DR confirmed the pattern and showed further bisection of each cluster. Common epitopes present exclusively in all HLA molecules of each cluster were determined following the HLA epitope registry. Thus, we propose that 19 out of 123 HLA-DR epitopes are those that mainly lead anti-HLA-DR responses in the studied population. Likewise, we identified 22 out of 83 epitopes responsible for anti-HLA-DQ and 13 out of 62 responsible for anti-HLA-DP responses. Interpretation of these results may elucidate mechanisms of interlocus cross-reactivity, providing an alternative way of estimating the significance of each epitope in a population and thus suggesting a novel strategy towards optimal donor selection.

Hidden Patterns of Anti-HLA Class I Alloreactivity Revealed Through Machine Learning.

Detection of alloreactive anti-HLA antibodies is a frequent and mandatory test before and after organ transplantation to determine the antigenic targets of the antibodies. Nowadays, this test involves the measurement of fluorescent signals generated through antibody-antigen reactions on multi-beads flow cytometers. In this study, in a cohort of 1,066 patients from one country, anti-HLA class I responses were analyzed on a panel of 98 different antigens. Knowing that the immune system responds typically to "shared" antigenic targets, we studied the clustering patterns of antibody responses against HLA class I antigens without any a priori hypothesis, applying two unsupervised machine learning approaches. At first, the principal component analysis (PCA) projections of intra-locus specific responses showed that anti-HLA-A and anti-HLA-C were the most distantly projected responses in the population with the anti-HLA-B responses to be projected between them. When PCA was applied on the responses against antigens belonging to a single locus, some already known groupings were confirmed while several new cross-reactive patterns of alloreactivity were detected. Anti-HLA-A responses projected through PCA suggested that three cross-reactive groups accounted for about 70% of the variance observed in the population, while anti-HLA-B responses were mainly characterized by a distinction between previously described Bw4 and Bw6 cross-reactive groups followed by several yet undocumented or poorly described ones. Furthermore, anti-HLA-C responses could be explained by two major cross-reactive groups completely overlapping with previously described C1 and C2 allelic groups. A second feature-based analysis of all antigenic specificities, projected as a dendrogram, generated a robust measure of allelic antigenic distances depicting bead-array defined cross reactive groups. Finally, amino acid combinations explaining major population specific cross-reactive groups were described. The interpretation of the results was based on the current knowledge of the antigenic targets of the antibodies as they have been characterized either experimentally or computationally and appear at the HLA epitope registry.

Patterns of 1,748 Unique Human Alloimmune Responses Seen by Simple Machine Learning Algorithms.

Allele specific antibody response against the polymorphic system of HLA is the allogeneic response marker determining the immunological risk for graft acceptance before and after organ transplantation and therefore routinely studied during the patient's workup. Experimentally, bead bound antigen- antibody reactions are detected using a special multicolor flow cytometer (Luminex). Routinely for each sample, antibody responses against 96 different HLA antigen groups are measured simultaneously and a 96-dimensional immune response vector is created. Under a common experimental protocol, using unsupervised clustering algorithms, we analyzed these immune intensity vectors of anti HLA class II responses from a dataset of 1,748 patients before or after renal transplantation residing in a single country. Each patient contributes only one serum sample in the analysis. A population view of linear correlations of hierarchically ordered fluorescence intensities reveals patterns in human immune responses with striking similarities with the previously described CREGs but also brings new information on the antigenic properties of class II HLA molecules. The same analysis affirms that "public" anti-DP antigenic responses are not correlated to anti DR and anti DQ responses which tend to cluster together. Principal Component Analysis (PCA) projections also demonstrate ordering patterns clearly differentiating anti DP responses from anti DR and DQ on several orthogonal planes. We conclude that a computer vision of human alloresponse by use of several dimensionality reduction algorithms rediscovers proven patterns of immune reactivity without any a priori assumption and might prove helpful for a more accurate definition of public immunogenic antigenic structures of HLA molecules. Furthermore, the use of Eigen decomposition on the Immune Response generates new hypotheses that may guide the design of more effective patient monitoring tests.