Biology subjects

Louzoun, Y.

Publications and source records attributed to Louzoun, Y..

16 recordsLinked to original sources

bioRxiv2026

Beyond eplet counts: machine learning integration of evolutionary distance and electrostatic divergence improves prediction of HLA-DQ donor-specific antibody formation after kidney transplantation

The emergence of donor-specific antibodies (DSA) against HLA-DQ reduces graft survival and limits future transplant options. Current practice predicts DSA development from HLA antigen mismatches or unique eplet counts (HLAMatchmaker), but predictive accuracy remains limited. Improving the prediction accuracy is crucial for solid organ transplants. We analysed a retrospective cohort of 240 kidney transplants (480 donor-DQ samples) in which the donor DQ allele targeted by each post-transplant DSA was identified. We compared the predictive accuracy of unique eplets (HLAMatchmaker), total eplets, amino acid mismatches, PAM genetic distance, and electrostatic mismatch score (EMS), individually and in combination. DSA emerge through the response of two donor and two patient DQ heterodimers. We evaluate the appropriate combination method of patient and donor alleles. Total eplet or amino acid mismatch counts outperformed unique eplets as used in HLAMatchmaker (AUC 0.79 vs. 0.73). Combining eplets with PAM and EMS distances in an XGBoost classifier achieved AUC 0.84, with a hazard ratio above 7. The most informative features were PAM distance and a subset of eplets distributed across the HLA-DQ alpha and beta chains. When extending prediction from a single donor DQ allele to the donor as a whole, the maximum of the two per-allele scores outperformed a probabilistic combination. The unique-eplet approach is suboptimal for predicting anti-DQ DSA in this cohort; combining total eplets with genetic and biochemical distance metrics yields substantially better prediction, essential for the reduction of DSA emergence. External validation in independent cohorts is the necessary next step.

immunology↗

bioRxiv2026

GRIMM-II: A Two-Stage Real-Time Algorithm for Nine-Locus HLA Imputation and Matching with Up to Three Mismatches

BackgroundThe success of hematopoietic stem cell transplantation (HSCT) depends critically on human leukocyte antigen (HLA) matching between donor and recipient. While traditional matching focuses on five classical HLA loci (A, B, C, DRB1, DQB1), clinical practice increasingly considers extended typing at nine loci, including DPA1, DQA1, DPB1, and DRB3/4/5. Furthermore, emerging evidence supports transplantation with up to three HLA mismatches under post-transplant cyclophosphamide (PTCy) regimens. However, current donor search algorithms cannot efficiently identify donors with multiple mismatches across extended HLA loci in real-time. MethodsWe developed GRIMM-II (GRaph IMputation and Matching, version II), which comprises two novel algorithms: ML-GRIM (Multi-Locus GRIM) for HLA imputation across multiple loci, and ML-GRMA (Multi-Locus GRMA) for real-time donor-patient matching with up to three mismatches. Both algorithms employ a two-stage approach that combines efficient candidate reduction through graph-theoretic frameworks with detailed genotype comparison. ML-GRIM partitions genotypes into class I (HLA-A, B, C) and class II (remaining loci) components, enabling memory-efficient storage and rapid candidate identification. ML-GRMA searches a pre-imputed donor graph composed of donor genotypes and their sub-components, then computes asymmetric graft-versus-host (GvH) and host-versus-graft (HvG) mismatch probabilities to provide clinically relevant compatibility assessments. Both imputation and matching tools are available as a web application at https://grimmard.math.biu.ac.il/ and through GitHub repositories at https://github.com/nmdp-bioinformatics/py-graph-imputation (imputation) and https://github.com/nmdp-bioinformatics/py-graph-match (matching). ResultsWe validated ML-GRMA and ML-GRIM using the WMDA3 (World Marrow Donor Association) validation dataset, successfully reproducing all previously reported matches while identifying numerous additional candidate donors not detected by previous algorithms. Further validation of ML-GRMA using 3,000 patients with artificially introduced mismatches (0-3 allele substitutions) demonstrated 100% sensitivity and specificity in identifying matching donors at expected mismatch levels. We validated ML-GRIM using simulated nine-locus typings derived from 8,078,224 US donors in the NMDP registry. The algorithm successfully imputed genotypes across variable numbers of typed loci while incorporating multiethnic haplotype frequencies. The algorithm achieved real-time performance with typical imputation times under one second and matching times of 1-13 seconds per patient for up to three mismatches, even when searching databases exceeding 8 million donors. Notably, ML-GRMA identified substantially more potentially suitable donors than traditional algorithms by accounting for the biological reality that GvH and HvG mismatches often differ, particularly for donors homozygous at specific loci. To evaluate ML-GRIM performance with low-resolution typing, we tested it on simulated 3-locus typings from the same population. The resulting imputation accuracy correlated with the mutual information between typed loci and complete genotypes. ConclusionsGRIMM-II provides a scalable, memory-efficient solution for nine-locus HLA imputation and real-time identification of donors with up to three mismatches. The graph-based framework supports dynamic registry updates and can readily accommodate additional HLA loci and matching criteria as clinical knowledge evolves. By expanding the pool of acceptable donors while maintaining computational efficiency, GRIMM-II addresses a critical need in contemporary transplantation practice, particularly for patients from underrepresented ethnic minorities who face lower probabilities of finding perfectly matched donors.

bioinformatics↗

bioRxiv2025

Nutrition-dependent development of the Oral Microbiome in Early Pregnancy

BackgroundMost studies of the oral microbiome during pregnancy have focused on the second and third trimesters (T2, T3, respectively). Findings remain inconsistent-some report shifts in specific taxa, whereas others observe little change in diversity. To date, no large-scale longitudinal study has examined oral microbiome development across all three trimesters, leaving early gestational dynamics (during the first trimester; T1) largely unexplored. MethodsWe conducted a longitudinal analysis of the oral microbiome in 346 pregnant women from Israel and validated key findings in an independent cohort of 154 pregnant women from Russia. In Israel, saliva samples were collected during T1 (11-14 weeks), T2 (24-28 weeks), and T3 (32-38 weeks) trimesters; in Russia, samples were collected during T2 and T3 with similar ranges of gestational weeks. Alongside sample collection, participants completed dietary and health questionnaires to assess maternal factors that could influence microbial composition. Microbial profiles were analyzed to test for (i) differential abundance across trimesters and (ii) the influence of maternal nutrition and lifestyle factors on these dynamics. ResultsSignificant shifts in oral microbial composition were observed as early as the transition from T1 to T2. Alpha diversity decreased progressively across pregnancy (Shannon index: T1 = 3.261, T2 = 3.173, T3 = 3.109; Kruskal-Wallis p = 0.0023). Notable taxonomic changes included a significant reduction in Verrucomicrobiota (particularly Akkermansia muciniphila) and an increase in Synergistota from T1 to T2 (adjusted p < 0.01), alongside an increase in Gammaproteobacteria and a decrease in Erysipelotrichia, suggesting an ecological shift towards potentially pro-inflammatory communities. Despite these systematic population-level changes, within-subject microbial distances across trimesters were smaller than between-subject distances, indicating that individual women maintained relatively stable microbial profiles over time. Among 54 maternal variables examined, gluten-free diet showed the strongest and most consistent associations with oral microbiome composition across all trimesters, followed by smoking history and conception method. Key findings were validated in an independent cohort of 154 Russian women. ConclusionsThis study provides the first large-scale evidence of significant oral microbiome changes beginning in early pregnancy, characterized by reduced diversity and a directional shift toward potentially pro-inflammatory communities. The strong associations with gluten consumption and smoking suggest a large-scale effect of lifestyle on the pregnancy oral microbiome. The alteration in the microbial composition highlights the oral microbiome as a sensitive marker of gestational physiology.

microbiology↗

bioRxiv2025

Niche Overlap Is Not Enough: Same Overlap, Opposite Dynamics

Niche overlap (NO) is a central concept in coexistence theory, determining the outcome of competitive interactions at equilibrium. Yet, NO is a composite quantity that collapses multiple mechanisms and thus cannot by itself capture community dynamics. We quantify this gap by analyzing one of the simplest out-of-equilibrium observables: temporal correlations in species abundances, widely used in microbial ecology. Within a generalized consumer-resource framework, we separate the mechanistic traits that shape dynamical responses and show that systems with identical NO can display opposite correlation patterns. The decisive predictor is the yield-depletion mismatch (YDM) - the mismatch between species depletion and yield dissimilarities. Analytical theory, stochastic simulations, and a reanalysis of microbial time-series consistently show that YDM, not NO, sets the sign and magnitude of abundance correlations, whereas instantaneous growth-rate correlations increase with NO. Our results suggest that other out-of-equilibrium properties, such as invasion probabilities and coexistence robustness, will likewise be largely independent of NO and instead controlled by distinct out-of-niche parameters.

ecology↗

bioRxiv2024

Prediction of KIR3DL1/Human Leukocyte Antigen binding

KIR3DL1 is a polymorphic inhibitory Natural Killer (NK) cell receptor that recognizes Human Leukocyte Antigen (HLA) class I allotypes that contain the Bw4 motif. Structural analyses have shown that in addition to residues 77-83 that span the Bw4 motif, polymorphism at other sites throughout the HLA molecule can influence the interaction with KIR3DL1. Given the extensive polymorphism of both KIR3DL1 and HLA class I, we built a machine learning prediction model to describe the influence of allotypic variation on the binding of KIR3DL1 to HLA class I. Nine KIR3DL1 tetramers were screened for reactivity against a panel of HLA class I molecules which revealed different patterns of specificity for each KIR3DL1 allotype. Separate models were trained for each of KIR3DL1 allotypes based on the full amino sequence of exons 2 and 3 encoding the 1 and 2 domains of the class I HLA allotypes, the set of polymorphic positions that span the Bw4 motif, or the positions that encode 1 and 2 but exclude the connecting loops. The Multi-Label-Vector-Optimization (MLVO) model trained on all alpha helix positions performed best with AUC scores ranging from 0.74 to 0.974 for the 9 KIR3DL1 allotype models. We show that a binary division into binder and non-binder is not precise, and that intermediate levels exist. Using the same models, within the binder group, high- and low-binder categories can also be predicted, the regions in HLA affecting the high vs low binder being completely distinct from the classical Bw4 motif. We further show that these positions affect binding affinity in a nonadditive way and induce deviations from linear models used to predict interaction strength. We propose that this approach should be used in lieu of simpler binding models based on a single HLA motif.

immunology↗

bioRxiv2024

Efficient test for deviation from Hardy Weinberg Equilibrium with known or ambiguous typing in highly polymorphic loci

The Hardy-Weinberg Equilibrium (HWE) assumption is essential to many population genetics models. Multiple tests were developed to test its applicability in observed genotypes. Current methods are divided into exact tests applicable to small populations and a small number of alleles, and approximate goodness of fit tests. Existing tests cannot handle ambiguous typing in multi-allelic loci. We here present a novel exact test (UMAT - Unambiguous Multi Allelic Test) practically not limited in the number of alleles and population size, based on a perturbative approach around the current observations. We show its accuracy in the detection of deviation from HWE. We then propose an additional model to handle ambiguous typing using either sampling into UMAT or a goodness of fit test with a variance estimate taking ambiguity into account, named ASTA (Asymptotic Statistical Test with Ambiguity). We show the accuracy of ASTA and the possibility to detect of the source of deviation from HWE. We apply these tests to the HLA loci to recover multiple previously reported deviations from HWE, and a large number of new ones.

genetics↗

bioRxiv2022

Counting Is Almost All You Need

The immune memory repertoire encodes the history of present and past infections and immunological attributes of the individual. As such, multiple methods were proposed to use T-cell receptor (TCR) repertoires to detect disease history. We here show that the counting method outperforms all existing algorithms. We then show that the counting can be further improved using a novel attention model to weight the different TCRs. The attention model is based on the projection of TCRs using a Variational AutoEncoder (VAE). Both counting and attention algorithms predict better than any current algorithm whether the host had CMV and its HLA alleles. As an intermediate solution between the complex attention model and the very simple counting model, we propose a new Graph Convolutional Network approach that obtains the accuracy of the attention model and the simplicity of the counting model. The code for the models used in the paper are provided in: https://github.com/louzounlab/CountingIsAlmostAllYouNeed

immunology↗

bioRxiv2021

Progesterone supplementation in mice leads to microbiome alterations and weight gain in a sex-specific manner

BackgroundProgesterone is a steroid hormone produced by the ovaries, involved in pregnancy progression and necessary for successful gestation. We have previously shown that progesterone affects gut microbiota composition and leads to increased relative abundance of Bifidobacterium. ResultsIn non-pregnant female GF mice, levels of progesterone were significantly higher than in SPF mice of the same status. However, no significant differences were observed between GF and SPF males. Females treated with progesterone gained more weight than females treated with a placebo. In contrast to female mice, males treated with progesterone did not gain significantly more weight than males treated with a placebo. Progesterone supplementation led to microbial changes in females but not in males (16S rRNA sequencing). Accordingly, the weight gain observed in female mice treated with progesterone was fully transferable to both male and female germ-free mice via fecal transplantation. ConclusionsWe demonstrate that bacteria play a role in regulating progesterone levels in a female-specific manner. Furthermore, weight gain and metabolic changes associated with progesterone may be mediated by the gut microbiota.

microbiology↗

bioRxiv2020

MIPMLP - Microbiome Pre-processing Machine Learning Pipeline.

16S sequencing results are often used for Machine Learning (ML) tasks. 16S gene sequences are represented as feature counts, which are associated with taxonomic representation. Raw feature counts may not be the optimal representation for ML. We checked multiple preprocessing steps and tested the optimal combination for 16S sequencing-based classification tasks. We computed the contribution of each step to the accuracy as measured by the Area Under Curve (AUC) of the classification. We show that the log of the feature counts is much more informative than the relative counts. We further show that merging features associated with the same taxonomy at a given level, through a dimension reduction step for each group of bacteria improves the AUC. Finally, we show that z-scoring has a very limited effect on the results. These preprocessing steps are integrated into the MIPMLP - Microbiome Preprocessing Machine Learning Pipeline, which is available as a stand alone version at https://github.com/louzounlab/microbiome/tree/master/Preprocess or as a service at http://mip-mlp.math.biu.ac.il/Home ImportanceMicrobiome composition has been proposed as a biomarker (mic-marker) for multiple diseases. However, a clear analysis of the optimal way to represent the gene sequence counts is still lacking. We propose a simple and straight forward method that significantly improves the accuracy of mic-marker studies. This method can be of use to merge two of the most important advances in biology in the last decade: Microbiome analysis, and the introduction of machine learning methods to biological studies.

microbiology↗

bioRxiv2020

Projection of gut microbiome pre and post-bariatric surgery to predict surgery outcome

BackgroundBariatric surgery is often the preferred method to resolve obesity and diabetes, with ~800,000 cases worldwide yearly and high outcome variability. The ability to predict the long-term Body Mass Index (BMI) change following surgery has important implications on individuals and the health care system in general. Given the tight connection between eating habits, sugar consumption, BMI, and the gut microbiome, we tested whether the microbiome before any treatment is associated with different treatment outcomes, as well as other intakes (high-density lipoproteins (HDL), Triglycerides, etc.). ResultsA projection of the gut microbiome composition of obese (sampled before and after bariatric surgery) and slim patients into principal components was performed and the relation between this projection and surgery outcome was studied. The projection reveals 3 different microbiome profiles belonging to slim, obese, and obese who underwent bariatric surgery, with post-surgery more different from the slim than the obese. The same projection allowed for a prediction of BMI loss following bariatric surgery, using only the pre-surgery microbiome. ConclusionsThe gut microbiome can be decomposed into main components depicting the patients development and predicting in advance the outcome. Those may be translated into better clinical management of obese individuals planning to undergo metabolic surgery. ImportanceBMI and diabetes can affect the gut microbiome composition. Bariatric surgery has large variabilities in outcome. The microbiome was previously shown to be a good predictor for multiple diseases. We analyzed here the gut microbiome before and after bariatric surgery and show that: O_LIThe microbiome before surgery can be used to predict surgery outcome. C_LIO_LIPost-surgery microbiome drifts further away from the slim microbiome than pre-surgery obese patients. C_LI These results can lead to a microbiome-based pre-surgery decision whether to perform surgery.

microbiology↗

bioRxiv2020

Autoencoder based local T cell repertoire density can be used to classify samples and T cell receptors

Recent advances in T cell repertoire (TCR) sequencing allow for characterization of repertoire properties, as well as the frequency and sharing of specific TCR. However, there is no efficient measure for the local density of a given TCR. TCRs are often described either through their Complementary Determining region 3 (CDR3) sequences, or theirV/J usage or their clone size. We here show that the local repertoire density can be estimated using a combined representation of these components through distance conserving autoencoders and Kernel Density Estimates (KDE). We present ELATE - an Encoder based LocAl Tcr dEnsity and show that the resulting density of a sample can be used as a novel measure to study repertoire properties. The cross-density between two samples can be used as a similarity matrix to fully characterize samples from the same host. Finally, the same projection in combination with machine learning algorithms can be used to predict TCR-peptide binding through the local density of known TCRs binding a specific target. Code availability- https://github.com/louzounlab/Autoencoder

immunology↗

bioRxiv2019

Multiple Measures Reveal The Value of Both Race And Geographic Ancestry For Self-Identification

There is long-standing tension regarding whether and how to use race or geographic ancestry in biomedical research. We examined multiple self-reported measures of race and ancestry from a cohort of over 100,000 U.S. residents alongside genetic data. We found that these measures are often non-overlapping, and that no single self-reported measure alone provides a better fit to genetic ancestry than a combination including both race and geographic ancestry. We also found that patterns of reporting for race and ancestry appear to be influenced by participation in direct-to-consumer genetic ancestry testing. Our results demonstrate that there is a place for the language of both race and geographic ancestry as we seek to empower individuals to fully describe their family history in research and medicine.\n\nOne Sentence SummarySelf-identification in the United States according to both racial and geographic terms best reflects genetic ancestry in individuals.

genetics↗

bioRxiv2019

Prediction of specific TCR-peptide binding from large dictionaries of TCR-peptide pairs

Current sequencing methods allow for detailed samples of T cell receptors (TCR) repertoires. To determine from a repertoire whether its host had been exposed to a target, computational tools that predict TCR-epitope binding are required. Currents tools are based on conserved motifs and are applied to peptides with many known binding TCRs. Given any TCR and peptide, we employ new NLP-based methods to predict whether they bind. We combined large-scale TCR-peptide dictionaries with deep learning methods to produce ERGO (pEptide tcR matchinG predictiOn), a highly specific and generic TCR-peptide binding predictor. A set of standard tests are defined for the performance of peptide-TCR binding, including the detection of TCRs binding to a given peptide/antigen, choosing among a set of candidate peptides for a given TCR and determining whether any pair of TCR-peptide bind. ERGO significantly outperforms current methods in these tests even when not trained specifically for each test. The software implementation and data sets are available at https://github.com/louzounlab/ERGO

immunology↗

bioRxiv2018

Single haplotype admixture models using large scale HLA genotype frequencies to reproduce human admixture

The Human Leukocyte Antigen (HLA) is the most polymorphic region in humans. Anthropologists use HLA to trace populations migration and evolution. However, recent admixture between populations masks the ancestral haplotype frequency distribution.\n\nWe present an HLA-based method based on high-resolution HLA haplotype frequencies to resolve population admixture using a non-negative matrix factorization formalism and validated using haplotype frequencies from 56 populations. The result is a minimal set of original populations decoding roughly 90% of the total variance in the studied admixtures. These original populations agree with the geographical distribution, phylogenies and recent admixture events of the studied groups.\n\nWith the growing population of multi-ethnic individuals, the matching process for stem-cell and solid organ transplants is becoming more challenging. The presented algorithm provides a framework that facilitates the breakdown of highly admixed populations into original groups, which can be used to better match the rapidly growing population of multi-ethnic individuals worldwide.\n\nAuthor SummaryHuman Leukocyte Antigen (HLA) is known to be the most polymorphic region in the human genome. Anthropologists frequently use HLA to trace migration and evolution of different populations. This is due to the high linkage among HLA genes leading to the transmission of intact haplotypes from parents to offspring, hence preserving key population ancestral features.\n\nWe developed a new HLA-based method to identify admixture models in mixed populations using high-resolution HLA haplotype frequencies. Our results highlight that a single highly polymorphic locus can contain enough information to map clearly human admixture and the population genetics of the different human populations, and reproduces results based on SNP arrays.\n\nThe presented algorithm is validated using haplotype frequencies sampled from 56 worldwide populations. Under such factorization we demonstrate that 90% of the variance in these populations can be explained using a much-reduced set of 8 ethnic groups. We demonstrate that the estimated ethnic groups and admixture models agree with the geographical distribution, population phylogenies and recent historic admixture events of the studied populations.

immunology↗

bioRxiv2018

GRIMM: GRaph IMputation and Matching for HLA Genotypes

Motivation: For over 10 years allele-level HLA matching for bone marrow registries has been performed in a probabilistic context. HLA typing technologies provide ambiguous results in that they could not distinguish among all known HLA allele sequences, therefore registries have implemented matching algorithms that provide lists of donor and cord blood units ordered in terms of the likelihood of allele-level matching at specific HLA loci. With the growth of registry sizes, current match algorithm implementations are unable to provide match results in real time.\n\nResults: We present here novel computationally-efficient open source implementation of an HLA imputation and match algorithm using a graph database platform. Using graph traversal, our algorithm runtime grows slowly with registry size. This implementation generates results that agree with consensus output on a publicly-available match algorithm crossvalidation dataset.\n\nAvailability: The Python, Perl and Neo4jJcode is available at https://git.com/nmdp-bioinformatics/grimm\n\nSupplementary information: Supplementary data are available at Bioinformatics online.

bioinformatics↗

bioRxiv2018

Long term context dependent genetic adaptation of the viral genetic cloud

RNA viruses generate a cloud of genetic variants within each host. This cloud contains high frequency genotypes, and a very large number of rare variants. While the dynamics of frequent variants are affected by the fitness of each variant, the rare variants cloud is affected by more complex genetic factors, including context dependent mutations. It serves as a spearhead for the viral populations movement within the adaptive landscape. We here use an experimental evolution system to show that the genetic cloud surrounding the Coxsackie virus master sequence slowly, but steadily, evolves over hundreds of generations. The evolution of the rare variants cloud often precedes the appearance of high frequency variants. The rare variants clouds evolution is driven by a combination of a context-dependent mutation pattern and selection for and against specific nucleotide compositions.This combination affects the mutated dinucleotide distribution, and eventually leads to a non-uniform dinucleotide distribution in the main viral sequence. We then tested these conclusions on other RNA viruses with similar conclusions.

genetics↗