Biology subjects

Pavlovic, M.

Publications and source records attributed to Pavlovic, M..

At least 19 recordsLinked to original sources

bioRxiv2026

What Do Generative Models Learn About Adaptive Immune Receptor Repertoires? A Benchmark Study

Generative models are increasingly used to model adaptive immune receptor repertoire (AIRR) sequence distributions, promising to decode the sequence diversity shaping immune responses and accelerate the design of therapeutic antibodies and T-cell receptors. Yet it remains unclear whether these models produce biologically meaningful outputs or merely capture surface-level sequence statistics while missing features driven by receptor generation and selection. Rigorous evaluation is needed, but the field lacks established standards, as existing machine learning metrics do not all translate directly to the AIRR domain, given the complex structure of the data and the lack of biological ground truth. Consequently, researchers face difficulties in evaluating the models and selecting appropriate ones, which can critically affect downstream clinical applications. Here, we apply a suite of evaluation metrics tailored to AIRR sequence data and present a systematic comparison of popular generative model families proposed for the AIRR field, including variational autoencoders, long short-term memory networks, antibody language models, selection models, and simple statistical baselines. We focus specifically on the task of learning individual-specific immune receptor repertoires, a clinically relevant challenge with direct implications for personalized immunotherapy, disease monitoring, and vaccine response studies. By analyzing the sequences generated by each model, we identify memorization risks, innovation capabilities, and sensitivity to hyperparameter tuning. Taken together, these results advance the understanding of how current generative models reproduce the biology of individual immune repertoires and lay the groundwork for more principled model development and evaluation.

immunology↗

bioRxiv2026

Unsupervised Machine Learning for Adaptive Immune Receptors with immuneML

Machine learning (ML) enables adaptive immune receptor repertoires (AIRRs) analyses for biomarker identification and therapeutic development. With the majority of AIRR data partially or imperfectly labeled, unsupervised ML is essential for motif discovery, biologically meaningful clustering, and generation of novel receptor sequences. However, no unified framework for unsupervised ML exists in the AIRR field, hindering the assessment of model robustness and generalizability. Here, we present an immuneML release advancing unsupervised ML in the AIRR field through unified clustering workflows, interpretable generative modeling, integration with protein language model embeddings, dimensionality reduction, and visualization. We demonstrate immuneMLs utility in three use cases: (i) benchmarking generative models for epitope-specific sequence generation, assessing specificity and novelty, (ii) systematic evaluation of clustering approaches on experimental receptor sequences against biological properties, such as epitope specificity and MHC, and (iii) unsupervised analysis of an experimental AIRR dataset to examine potential confounding, a practice widespread in related fields but unexplored in AIRR analyses.

bioinformatics↗

bioRxiv2025

Quantitative mapping of antigen specificity in adaptive immune repertoire embedding spaces

The adaptive immune receptor repertoire (AIRR) encompasses an immense diversity of antibody and T-cell receptor sequences, whose collective organization - how receptors are distributed, clustered, and interrelated across sequence and functional (e.g., antigen-binding) dimensions - remains poorly characterized. Representing AIRRs in continuous representation spaces that capture sequence, biochemical, and structural similarity between receptors may enable comparisons beyond discrete sequence features. Using both one-hot encodings and protein language model (PLM) embeddings, we developed a quantitative framework to map immune receptor organization at global (sequence-set-level) and local (single-sequence-level) scales. Applying the geometry-aware Wasserstein-2 distance, we show that the global structure of the AIRR space can be recovered from as few as [~]105 sequence embeddings, at least 10 orders of magnitude smaller than the theoretical immune receptor diversity. We found that immune receptor sequences annotated with different antigen specificities occupy distinct regions of representation space. To resolve local relationships, we introduce a spatial homogeneity metric that quantifies the extent of functional clustering. We found higher spatial homogeneity in embedding spaces than in sequence space for diverse antigen-specific datasets. Our framework establishes a foundation for quantitative mapping of adaptive immune repertoire organization.

immunology↗

bioRxiv2025

Nischarin agonist rilmenidine shows antimetastatic potential in pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is one of the most aggressive cancer types with a dismal prognosis. Early metastatic dissemination and rich desmoplastic stroma limit therapeutic efficacy, and the discovery of new treatment approaches that will target these obstacles is of great importance. Nischarin (NISCH)/Imidazoline-1 receptor (IR-1) is a potential tumor suppressor, that is involved in the regulation of cell migration, invasion, and cytoskeletal organization. Importantly, several FDA-approved agonists target this receptor. This study aimed to examine the expression of NISCH in PDAC and the effects of its agonists with the intent of drug repurposing. NISCH was expressed in PDAC tumor tissue, in both the epithelial and stromal compartments of tumors, and higher NISCH expression was associated with longer patient survival. Out of the three tested NISCH agonists, rilmenidine was the most potent and could induce cancer cell apoptosis. Rilmenidine treatment induced transcriptional changes in cancer cells associated with so far reported NISCH function and decreased their adhesion and migration in vitro. In CAFs, rilmenidine treatment decreased the expression of activation markers and limited cancer cell-CAF cytokine communication in co-culture. Ultimately, in the in vivo tumor xenograft zebrafish model, rilmenidine reduced the metastatic spread of pancreatic cancer cells. Our results suggest that NISCH agonist rilmenidine is a promising candidate for drug repurposing as an antimetastatic agent in PDAC and that NISCH can be a potential target for the development of new PDAC therapeutics.

cancer biology↗

bioRxiv2024

Revealing the hidden sequence distribution of epitope-specific TCR repertoires and its influence on machine learning model performance

Numerous efforts have been made to decipher the epitope-T cell receptor (TCR) recognition code. Both simple machine learning techniques and deep learning strategies have been used to train models to predict the binding of epitopes by TCR sequences. A good training data set rests at the basis of every accurate prediction model, yet little attention has been given to the composition of these data sets. In this paper, we studied the natural distribution of TCR sequences within epitope-specific TCR repertoires, i.e. a set of TCRs binding the same epitope, and its impact on the predictability of TCR-epitope interactions. We found that the observed diversity of these repertoires can result from a smaller set of core binding motifs constrained by TCR generation. Moreover, a clear relationship was found between the sequence distribution of the training data and performance metrics, emphasizing the importance of the used ground-truth data when using machine learning models in this domain. Taken together, these findings inform data set composition to help push epitope-TCR prediction models to the next level.

bioinformatics↗

bioRxiv2023

Simulation of adaptive immune receptors and repertoires with complex immune information to guide the development and benchmarking of AIRR machine learning

Machine learning (ML) has shown great potential in the adaptive immune receptor repertoire (AIRR) field. However, there is a lack of large-scale ground-truth experimental AIRR data suitable for AIRR-ML-based disease diagnostics and therapeutics discovery. Simulated ground-truth AIRR data are required to complement the development and benchmarking of robust and interpretable AIRR-ML methods where experimental data is currently inaccessible or insufficient. The challenge for simulated data to be useful is incorporating key features observed in experimental repertoires. These features, such as antigen or disease-associated immune information, cause AIRR-ML problems to be challenging. Here, we introduce LIgO, a software suite, which simulates AIRR data for the development and benchmarking of AIRR-ML methods. LIgO incorporates different types of immune information both on the receptor and the repertoire level and preserves native-like generation probability distribution. Additionally, LIgO assists users in determining the computational feasibility of their simulations. We show two examples where LIgO supports the development and validation of AIRR-ML methods: (1) how individuals carrying out-of-distribution immune information impacts receptor-level prediction performance and (2) how immune information co-occurring in the same AIRs impacts the performance of conventional receptor-level encoding and repertoire-level classification approaches. LIgO guides the advancement and assessment of interpretable AIRR-ML methods.

bioinformatics↗

bioRxiv2023

BrainLine: An Open Pipeline for Connectivity Analysis of Heterogeneous Whole-Brain Fluorescence Volumes

Whole-brain fluorescence images require several stages of computational processing to fully reveal the neuron morphology and connectivity information they contain. However, these computational tools are rarely part of an integrated pipeline. Here we present BrainLine, an open-source pipeline that interfaces with existing software to provide registration, axon segmentation, soma detection, visualization and analysis of results. By implementing a feedback based training paradigm with BrainLine, we were able to use a single learning algorithm to accurately process a diverse set of whole-brain images generated by light-sheet microscopy. BrainLine is available as part of our Python package brainlit: http://brainlit.neurodata.io/.

bioengineering↗

bioRxiv2022

BioNumPy: Fast and easy analysis of biological data with Python

Python is a popular and widespread programming language for scientific computing, in large part due to the powerful array programming library NumPy, which makes it easy to write clean, vectorized and efficient code for handling large datasets. A challenge with using array programming for biological data is that the data is often non-numeric and variable-length (such as DNA sequences), inhibiting out-of-the-box use of standard array programming techniques. Thus, a tradition in bioinformatics has been to use low-level languages like C and C++ to write efficient code. This makes the tools less transparent to the average computational biologist - making them harder to understand, modify and contribute to. We here present a new Python package BioNumPy, which adds a layer on top of NumPy in order to enable intuitive array programming on biological datasets. BioNumPy is able to efficiently load biological datasets (e.g. FASTQ-files, BED-files and BAM-files) into NumPy-like data structures, so that NumPy operations like indexing, vectorized functions and reductions can be applied to the data. We show that BioNumPy is considerably faster than vanilla Python and other Python packages for common bioinformatics tasks, and in many cases as fast as tools written in C/C++. BioNumPy thus bridges a long-lasting gap in bioinformatics, allowing the same programming language (Python) to be used across the full spectrum from quick and simple scripts to computationally efficient processing of large-scale data.

bioinformatics↗

bioRxiv2021

In silico proof of principle of machine learning-based antibody design at unconstrained scale

Generative machine learning (ML) has been postulated to be a major driver in the computational design of antigen-specific monoclonal antibodies (mAb). However, efforts to confirm this hypothesis have been hindered by the infeasibility of testing arbitrarily large numbers of antibody sequences for their most critical design parameters: paratope, epitope, affinity, and developability. To address this challenge, we leveraged a lattice-based antibody-antigen binding simulation framework, which incorporates a wide range of physiological antibody binding parameters. The simulation framework enables both the computation of antibody-antigen 3D-structures as well as functions as an oracle for unrestricted prospective evaluation of the antigen specificity of ML-generated antibody sequences. We found that a deep generative model, trained exclusively on antibody sequence (1D) data can be used to design native-like conformational (3D) epitope-specific antibodies, matching or exceeding the training dataset in affinity and developability variety. Furthermore, we show that transfer learning enables the generation of high-affinity antibody sequences from low-N training data. Finally, we validated that the antibody design insight gained from simulated antibody-antigen binding data is applicable to experimental real-world data. Our work establishes a priori feasibility and the theoretical foundation of high-throughput ML-based mAb design. HighlightsO_LIA large-scale dataset of 70M [3 orders of magnitude larger than the current state of the art] synthetic antibody-antigen complexes, that reflect biological complexity, allows the prospective evaluation of antibody generative deep learning C_LIO_LICombination of generative learning, synthetic antibody-antigen binding data, and prospective evaluation shows that deep learning driven antibody design and discovery at an unconstrained level is feasible C_LIO_LITransfer learning (low-N learning) coupled to generative learning shows that antibody-binding rules may be transferred across unrelated antibody-antigen complexes C_LIO_LIExperimental validation of antibody-design conclusions drawn from deep learning on synthetic antibody-antigen binding data C_LI Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=200 SRC="FIGDIR/small/451480v1_ufig1.gif" ALT="Figure 1"> View larger version (64K): org.highwire.dtl.DTLVardef@1d947bcorg.highwire.dtl.DTLVardef@1f4c51corg.highwire.dtl.DTLVardef@b8df67org.highwire.dtl.DTLVardef@1530ea5_HPS_FORMAT_FIGEXP M_FIG We leverage large synthetic ground-truth data to demonstrate the (A,B) unconstrained deep generative learning-based generation of native-like antibody sequences, (C) the prospective evaluation of conformational (3D) affinity, paratope-epitope pairs, and developability. (D) Finally, we show increased generation quality of low-N-based machine learning models via transfer learning. C_FIG

immunology↗

bioRxiv2021

A billion synthetic 3D-antibody-antigen complexes enable unconstrained machine-learning formalized investigation of antibody specificity prediction

Machine learning (ML) is a key technology for accurate prediction of antibody-antigen binding. Two orthogonal problems hinder the application of ML to antibody-specificity prediction and the benchmarking thereof: The lack of a unified ML formalization of immunological antibody specificity prediction problems and the unavailability of large-scale synthetic benchmarking datasets of real-world relevance. Here, we developed the Absolut! software suite that enables parameter-based unconstrained generation of synthetic lattice-based 3D-antibody-antigen binding structures with ground-truth access to conformational paratope, epitope, and affinity. We formalized common immunological antibody specificity prediction problems as ML tasks and confirmed that for both sequence and structure-based tasks, accuracy-based rankings of ML methods trained on experimental data hold for ML methods trained on Absolut!-generated data. The Absolut! framework thus enables real-world relevant development and benchmarking of ML strategies for biotherapeutics design. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=195 SRC="FIGDIR/small/451258v3_ufig1.gif" ALT="Figure 1"> View larger version (86K): org.highwire.dtl.DTLVardef@4c3a6aorg.highwire.dtl.DTLVardef@8f1deborg.highwire.dtl.DTLVardef@1f55ec6org.highwire.dtl.DTLVardef@2135dd_HPS_FORMAT_FIGEXP M_FIG The software framework Absolut! enables (A,B) the generation of virtually arbitrarily large numbers of synthetic 3D-antibody-antigen structures, (C,D) the formalization of antibody specificity as machine learning (ML) tasks as well as the exploration of ML strategies for real-world antibody-antigen binding or paratope-epitope prediction. C_FIG HighlightsO_LISoftware framework Absolut! to generate an arbitrarily large number of synthetic 3D-antibody-antigen structures that contain biological layers of antibody-antigen binding complexity that render ML predictions challenging C_LIO_LIImmunological antibody specificity prediction problems formalized as machine learning tasks for which the in silico complexes are immediately usable as benchmark datasets C_LIO_LIExploration of machine learning prediction accuracy as a function of architecture, dataset size, choice of negatives, and sequence-structure encoding C_LIO_LIRelative ML performance learnt on Absolut! datasets transfers to experimental datasets C_LI

immunology↗

bioRxiv2021

Integrative genomic and epigenomic analyses identified IRAK1 as a novel target for chronic inflammation-driven prostate tumorigenesis

The impacts of many inflammatory genes in prostate tumorigenesis remain understudied despite the increasing evidence that associates chronic inflammation with prostate cancer (PCa) initiation, progression, and therapy resistance. The overarching goal of this study was to identify dysregulated inflammatory genes that correlate with PCa progression and decipher their molecular mechanisms as well as clinical significance in PCa using integrative genomics, transcriptomics, and epigenomics approach. Our Weighted Gene Co-expression Analysis (WGCNA) and multivariate analysis identified 10 inflammatory genes: IRAK1, PPIL5/LRR1, HMGB3, HMGB2, TRAIP, IL1F5/IL36RN, ILF2, TRIM59, NFKBIL2/TONSL, and TRAF7 that were significantly associated with PCa progression. We explored the potentials of IRAK1 and other inflammatory genes as diagnostic and/or prognostic biomarkers by performing both KM survival and AUROC curve analyses. Our results indicate the clinical significance of these inflammatory genes in predicting the development and progression of PCa. IRAK1 was found to be overexpressed and hypomethylated in most PCa samples. A significantly high percentage of castration-resistant PCa (CRPC) and neuroendocrine PCa (NEPC) samples display copy number variations, especially amplification of the IRAK1 gene compared to the indolent prostate adenocarcinoma (PRAD) samples. Furthermore, we identified missense and frameshift mutations of IRAK1 in a few PRAD samples with potential functional implications. In conclusion, the results from this study suggest that IRAK1 dysregulation may be an important contributor to chronic prostatitis (inflammation) and PCa progression.

cancer biology↗

bioRxiv2021

Profiling the baseline performance and limits of machine learning models for adaptive immune receptor repertoire classification

BackgroundMachine learning (ML) methodology development for classification of immune states in adaptive immune receptor repertoires (AIRR) has seen a recent surge of interest. However, so far, there does not exist a systematic evaluation of scenarios where classical ML methods (such as penalized logistic regression) already perform adequately for AIRR classification. This hinders investigative reorientation to those scenarios where further method development of more sophisticated ML approaches may be required. ResultsTo identify those scenarios where a baseline method is able to perform well for AIRR classification, we generated a collection of synthetic benchmark datasets encompassing a wide range of dataset architecture-associated and immune state-associated sequence pattern (signal) complexity. We trained {approx}1300 ML models with varying assumptions regarding immune signal on{approx}850 datasets with a total of {approx}210000 repertoires containing {approx}42 billion TCR{beta} CDR3 amino acid sequences, thereby surpassing the sample sizes of current state-of-the-art AIRR ML setups by two orders of magnitude. We found that L1-penalized logistic regression achieved high prediction accuracy even when the immune signal occurs only in 1 out of 50000 AIR sequences. ConclusionsWe provide a reference benchmark to guide new AIRR ML classification methodology by: (i) identifying those scenarios characterised by immune signal and dataset complexity, where baseline methods already achieve high prediction accuracy and (ii) facilitating realistic expectations of the performance of AIRR ML models given training dataset properties and assumptions. Our study serves as a template for defining specialized AIRR benchmark datasets for comprehensive benchmarking of AIRR ML methods.

bioinformatics↗

bioRxiv2021

Individualized VDJ recombination predisposes the available Ig sequence space

The process of recombination between variable (V), diversity (D), and joining (J) immunoglobulin (Ig) gene segments determines an individuals naive Ig repertoire, and consequently (auto)antigen recognition. VDJ recombination follows probabilistic rules that can be modeled statistically. So far, it remains unknown whether VDJ recombination rules differ between individuals. If these rules differed, identical (auto)antigen-specific Ig sequences would be generated with individual-specific probabilities, signifying that the available Ig sequence space is individual-specific. We devised a sensitivity-tested distance measure that enables inter-individual comparison of VDJ recombination models. We discovered, accounting for several sources of noise as well as allelic variation in Ig sequencing data, that not only unrelated individuals but also human monozygotic twins and even inbred mice possess statistically distinguishable immunoglobulin recombination models. This suggests that, in addition to genetic, there is also non-genetic modulation of VDJ recombination. We demonstrate that population-wide individualized VDJ recombination can result in orders of magnitude of difference in the probability to generate (auto)antigen-specific Ig sequences. Our findings have implications for immune receptor-based individualized medicine approaches relevant to vaccination, infection, and autoimmunity.

immunology↗

bioRxiv2021

immuneML: an ecosystem for machine learning analysis of adaptive immune receptor repertoires

Adaptive immune receptor repertoires (AIRR) are key targets for biomedical research as they record past and ongoing adaptive immune responses. The capacity of machine learning (ML) to identify complex discriminative sequence patterns renders it an ideal approach for AIRR-based diagnostic and therapeutic discovery. To date, widespread adoption of AIRR ML has been inhibited by a lack of reproducibility, transparency, and interoperability. immuneML (immuneml.uio.no) addresses these concerns by implementing each step of the AIRR ML process in an extensible, open-source software ecosystem that is based on fully specified and shareable workflows. To facilitate widespread user adoption, immuneML is available as a command-line tool and through an intuitive Galaxy web interface, and extensive documentation of workflows is provided. We demonstrate the broad applicability of immuneML by (i) reproducing a large-scale study on immune state prediction, (ii) developing, integrating, and applying a novel method for antigen specificity prediction, and (iii) showcasing streamlined interpretability-focused benchmarking of AIRR ML.

bioinformatics↗

bioRxiv2020

Predicting Residues Involved in Anti-DNA Autoantibodies with Limited Neural Networks

Computer-aided rational vaccine design (RVD) and synthetic pharmacology are rapidly developing fields that leverage existing datasets for developing compounds of interest. Computational proteomics utilizes algorithms and models to probe proteins for functional prediction. A potentially strong target for such a computational approach is autoimmune antibodies which are the result of broken tolerance in the immune system where it cannot distinguish "self" from "non-self" resulting in attack of its own structures (proteins and DNA, mainly). The information on structure, function and pathogenicity of autoantibodies may assist in engineering RVD against autoimmune diseases. Current computational approaches exploit large datasets curated with extensive domain knowledge, most of which include the need for many computational resources and have been applied indirectly to problems of interest for DNA, RNA, and monomer protein binding. Here, we present a novel method for discovering potential binding sites. We employed long short-term memory (LSTM) models trained on FASTA primary sequences directly to predict protein binding in DNA-binding hydrolytic antibodies (abzymes). We also employed CNN models applied to the same dataset. While the CNN model outperformed the LSTM on the primary task of binding prediction, analysis of internal model representations of both models showed that the LSTM models highlighted sub-sequences that were more strongly correlated with sites known to be involved in binding. These results demonstrate that analysis of internal processes of recurrent neural network models may serve as a powerful tool for primary sequence analysis.

bioinformatics↗

bioRxiv2020

DeepRC: Immune repertoire classification with attention-based deep massive multiple instance learning

A central mechanism in machine learning is to identify, store, and recognize patterns. How to learn, access, and retrieve such patterns is crucial in Hopfield networks and the more recent transformer architectures. We show that the attention mechanism of transformer architectures is actually the update rule of modern Hop-field networks that can store exponentially many patterns. We exploit this high storage capacity of modern Hopfield networks to solve a challenging multiple instance learning (MIL) problem in computational biology: immune repertoire classification. Accurate and interpretable machine learning methods solving this problem could pave the way towards new vaccines and therapies, which is currently a very relevant research topic intensified by the COVID-19 crisis. Immune repertoire classification based on the vast number of immunosequences of an individual is a MIL problem with an unprecedentedly massive number of instances, two orders of magnitude larger than currently considered problems, and with an extremely low witness rate. In this work, we present our novel method DeepRC that integrates transformer-like attention, or equivalently modern Hopfield networks, into deep learning architectures for massive MIL such as immune repertoire classification. We demonstrate that DeepRC outperforms all other methods with respect to predictive performance on large-scale experiments, including simulated and real-world virus infection data, and enables the extraction of sequence motifs that are connected to a given disease class. Source code and datasets: https://github.com/ml-jku/DeepRC

bioinformatics↗

bioRxiv2019

Eomes and Brachyury control pluripotency exit and germ layer segregation by changes of chromatin state

The first lineage specification of pluripotent mouse epiblast segregates neuroectoderm (NE) from mesoderm and endoderm (ME) by currently poorly understood mechanisms. Here we demonstrate that the induction of any ME-gene programs critically relies on the T-box (Tbx) transcription factors Eomes and Brachyury that concomitantly repress pluripotency and NE gene programs. Tbx-deficient cells retain pluripotency and differentiate to NE lineages despite the presence of ME-inducing signals TGF{beta}/Nodal and WNT. Pluripotency and NE gene networks are additionally repressed by Tbx-induced ME factors, demonstrating a remarkable redundancy in program regulation to safeguard mutually exclusive lineage specification. Chromatin analyses revealed that accessibility of ME-gene enhancers depends on Tbx-binding, while NE-gene enhancers are accessible and activation-primed already at pluripotency state. This asymmetry of chromatin landscape thus explains the default differentiation of pluripotent cells to NE in the absence of ME-induction mediated through the activating and repressive functions of early Tbx factors Eomes and Brachyury.

developmental biology↗

bioRxiv2019

A finite vocabulary of antibody-antigen interaction enables predictability of paratope-epitope binding

Antibody-antigen binding relies on the specific interaction of amino acids at the paratope-epitope interface. The predictability of antibody-antigen binding is a prerequisite for de novo antibody and (neo-)epitope design. A fundamental premise for the predictability of antibody-antigen binding is the existence of paratope-epitope interaction motifs that are universally shared among antibody-antigen structures. In the largest set of non-redundant antibody-antigen structures, we identified structural interaction motifs, which together compose a commonly shared structure-based vocabulary of paratope-epitope interactions. We show that this vocabulary enables the machine learnability of antibody-antigen binding on the paratope-epitope level using generative machine learning. The vocabulary (i) is compact, less than 104 motifs, (ii) distinct from non-immune protein-protein interactions, and (iii) mediates specific oligo- and polyreactive interactions between paratope-epitope pairs. Our work successfully leveraged combined structure- and sequence-based learning showing that machine-learning-driven predictive paratope and epitope engineering is feasible.

immunology↗