Biology subjects

Burstein, D.

Publications and source records attributed to Burstein, D..

16 recordsLinked to original sources

bioRxiv2026

Language-Model-Based Detection of Genetic Editing in Bacteria

Recent advances in genome editing allow easy genetic manipulation of bacteria, providing them with new traits, some of which could be hazardous, e.g. enhanced virulence or extended resistance to antibiotics. The ability to detect artificially modified bacteria is crucial for identifying potential bio-threats. However, malicious genome editing could be challenging to trace due to the natural exchange of genes among bacteria through horizontal transfer. After curating extensive datasets including natural genomes and simulated edited genomes, we utilized a natural language processing approach to detect edited genomes. We developed a transformer-encoder-based machine-learning classifier that, instead of analyzing words in sentences, models gene families in genomes. After training the model on our datasets, it is able to accurately detect genes artificially added to bacterial genomes due to their unnatural context. Our approach provides a scalable method for identifying engineered sequences without relying on specific marker genes, with potential applications in biosecurity, agriculture, GMO regulation and more.

bioinformatics↗

bioRxiv2026

Plasmid-encoded host reprogramming promotes plasmid dissemination

Conjugative plasmids are major drivers of antibiotic resistance dissemination, yet how newly transferred plasmids establish in recipient cells remains poorly understood. Here we investigate YfjB, a previously uncharacterized conserved leading-region protein, which is zygotically induced immediately after plasmid entry and acts specifically during the earliest post-transfer stages. Multi-omics analyses reveal that YfjB reprograms host transcription, triggering extensive metabolic rewiring that compensates the transient fitness cost of plasmid acquisition. Structural analyses show that YfjB is a ParB-like protein containing a CTP-binding domain and a helix-turn-helix DNA-binding motif, linked to a previously uncharacterized dimerization module that forms a V-shaped clamp-like architecture compatible with DNA loading. Consistently, live-cell imaging reveals nucleoid-associated foci in transconjugants, and ChIP-seq identifies multiple chromosomal binding sites. We therefore rename the protein HerB (Host Expression Reprogrammer, ParB-like). More broadly, our findings reveal how mobile genetic elements facilitate their dissemination by transiently subverting host physiology.

microbiology↗

bioRxiv2026

Context-dependent determinants of CRISPR-Cas9 editing efficiency revealed through cross-species endogenous editing analysis

Accurate prediction of CRISPR-Cas9 guide RNA (gRNA) editing efficiency remains limited, particularly outside human systems, where models trained on exogenous human datasets show poor generalization. We analyzed Cas9 efficiency and repair outcomes using novel endogenous editing data from four human cell types, two tomato cell types, and cells from giant river prawn and black soldier fly. While integrating publicly available predictors via ensemble frameworks improved performance, our analysis revealed hundreds of novel features affecting activity. Crucially, dominant features related to sites competition for gRNA, and local geometric properties varied across systems, highlighting the strong context dependence of Cas9 efficiency and arguing against a universal model. Interestingly, codon usage bias-based features also emerged as informative predictors, as they are proxies for chromatin accessibility. In contrast, trends in repair outcomes remained conserved. This work provides essential resources for more generalizable CRISPR guide design.

synthetic biology↗

bioRxiv2026

FAMUS: A Few-Shot Learning Framework for Large-Scale Protein Annotation

Predicting gene function is a pivotal and challenging step in genomic and metagenomic data analysis. Current automatic annotation tools typically rely on the single most similar sequence from the query database and struggle to robustly set hit thresholds for annotation. The sparsity of proteins per annotation makes it challenging to confidently assign gene function for underrepresented families. Here, we present a contrastive learning framework for functional annotation. FAMUS (Functional Annotation Method Using Supervised contrastive learning) compares query sequences to a full array of profile Hidden Markov Models and transforms the similarity scores into a condensed vector space that minimizes the distance of proteins from the same family. The similarity scores of a query to all profiles are used for its representation instead of considering only the top-ranking hit. Unannotated sequences are incorporated as negative examples during training, enabling robust detection of proteins that fall outside the scope of the reference database without requiring a user-defined threshold. Using this approach, FAMUS outperformed KEGGs native KofamScan for KEGG Orthology annotation and InterPros InterProScan for PANTHER family annotation. We thus created four protein annotation models using protein families from the KEGG Orthology, InterPro family, OrthoDB, and EggNOG databases. All four models are available as a conda package and via our user-friendly web server, allowing users to annotate large-scale datasets. FAMUS is the first comprehensive and modular annotation framework based on contrastive learning. It supports both pre-defined and user-specific databases for tailored annotation, and can be easily integrated into any genomic and metagenomic analysis pipeline to facilitate accurate, large-scale functional annotation.

bioinformatics↗

bioRxiv2026

Optimizing Protein Tokenization: Reduced Amino Acid Alphabets for Efficient and Accurate Protein Language Models

Protein language models (pLMs) typically tokenize sequences at the single-amino-acid level using a 20-residue alphabet, resulting in long input sequences and high computational cost. Sub-word tokenization methods such as Byte Pair Encoding (BPE) can reduce sequence length but are limited by the sparsity of long patterns in proteins encoded by the standard amino acid alphabet. Reduced amino acid alphabets, which group residues by physicochemical properties, offer a potential solution but their performances with sub-word tokenization have not been systematically studied. In this work, we investigate the combined use of reduced amino acid alphabets and BPE tokenization in protein language models. We pre-train RoBERTa-based pLMs de novo using multiple reduced alphabets and evaluate them across diverse downstream tasks. Our results show that reduced alphabets enable substantially shorter input sequences and faster training and inference. These findings suggest that alphabet reduction may facilitate more effective sub-word tokenization, enabling increased efficiency with marginal impact on predictive performance, and for specific tasks even improving accuracy.

bioinformatics↗

bioRxiv2025

B-PPI: A Cross-Attention Model for Large-Scale Bacterial Protein-Protein Interaction Prediction

Protein-protein interactions (PPIs) are essential for the study of cellular function, yet computational prediction of bacterial PPIs remains limited. Most existing methods are trained on human data, reducing their applicability to bacterial systems. Here, we present B-PPI, a computational tool specifically designed for bacterial PPI prediction. B-PPI leverages embeddings from ProstT5, a structure-aware protein language model, and a cross-attention mechanism to capture residue-level inter-protein relationships. To facilitate training, we constructed B-PPI-DB, a large-scale bacterial PPI dataset derived from STRING, comprising 202,829 positive and negative interactions across 2,646 taxa with a 1:10 positive-to-negative ratio. We benchmarked B-PPI against TT3D, a state-of-the-art model trained on human PPI, which was previously evaluated on bacterial PPIs. B-PPI achieved substantially higher performance on bacterial data (AUPRC 0.926{+/-}0.006 vs. 0.230{+/-}0.005 and F1 0.866{+/-}0.007 vs. 0.299{+/-}0.005) with faster runtime. We further demonstrate that the model adapts to unseen bacterial interactions with minimal fine-tuning. Together, B-PPI and B-PPI-DB address a critical gap in computational microbiology, offering a framework for bacterial PPI prediction and a data resource for benchmarking and developing new tools in the field.

bioinformatics↗

bioRxiv2025

CAR T-cells dysfunction in the central nervous system is mediated by BBB-induced activation-induced cell death

Chimeric Antigen Receptor (CAR) T-cells can eliminate leukemia and lymphoma within the central nervous system (CNS). Following a case with CD19+ relapse in the CNS in the presence of persisting Tisagenlecleucel, we generated several models to study CAR T-cells in the CNS. In both immunocompetent and humanized leukemia-bearing mice, CNS-residing CAR T-cells were shown to have inferior ex-vivo killing capacities compared to CAR T-cells harvested from the spleen or bone marrow of receptive mice. In a mechanistic work-up, we identified two contributing factors: First, CAR T-cells cultured in cerebrospinal fluid showed impaired cytotoxic and proliferative function compared to those in serum or other media. Second, using a unique in vitro human blood-brain barrier (BBB) model composed of endothelial cells and brain pericytes, we demonstrated that CAR T-cells that crossed the BBB exhibited antigen-independent activation and diminished cytolytic capacity. This led to increased activation-induced cell death of the BBB-migrating CAR T-cells and was confirmed to be independent of the CAR target and the costimulatory domain. Our findings add possible mechanisms of CAR T-cell resistance in the CNS, relevant for CNS hematologic and non-hematologic tumors.

cancer biology↗

bioRxiv2025

Phage-encoded homing endonucleases attenuate bacterial immunity

The arms race between bacteria and bacteriophages (phages) gave rise to multiple layers of antagonistic mechanisms, many of which remain unexplored. Here, we investigated the anti-phage defense system GAPS4 and showed that it is a non-selective DNase triggered by sensing DNA breaks. We further demonstrated that this activation mechanism renders GAPS4 a double-edged sword, sensitizing bacteria to various forms of antibacterial antagonism. Using comparative genomics, we found that phage-encoded homing endonucleases, long considered selfish mobile genetic elements, enhance phage fitness by attenuating GAPS4-mediated immunity. Our findings shed light on the evolutionary advantage provided by these ubiquitous mobile elements to their host phages, and on the intricate evolutionary cross-talk between bacteria and their predators.

microbiology↗

bioRxiv2023

Using big sequencing data to identify chronic SARS-Coronavirus-2 infections

The evolution of SARS-Coronavirus-2 (SARS-CoV-2) has been characterized by the periodic emergence of highly divergent variants, many of which may have arisen during chronic infections of immunocompromised individuals. Here, we harness a global phylogeny of [~]11.7 million SARS-CoV-2 genomes and search for clades composed of sequences with identical metadata (location, age, and sex) spanning more than 21 days. We postulate that such clades represent repeated sampling from the same chronically infected individual. A set of 271 such chronic-like clades was inferred, and displayed signatures of an elevated rate of adaptive evolution, in line with validated chronic infections. More than 70% of adaptive mutations present in currently circulating variants are found in BA.1 chronic-like clades that predate the circulating variants by months, demonstrating the predictive nature of such clades. We find that in chronic-like clades the probability of observing adaptive mutations is approximately 10-20 higher than that in global transmission chains. We next employ language models to find mutations most predictive of chronic infections and use them to infer hundreds of additional chronic-like clades in the absence of metadata and phylogenetic information. Our proposed approach presents an innovative method for mining extensive sequencing data and providing valuable insights into future evolutionary patterns.

evolutionary biology↗

bioRxiv2023

A diverse repertoire of anti-defense systems is encoded in the leading region of plasmids

Plasmids are an important source of antibiotic-resistance genes that mobilize horizontally between bacteria, including many human pathogens. Bacteria express various defense mechanisms, such as CRISPR-Cas, restriction-modification systems, and SOS-response genes, to prevent the invasion of mobile elements. Yet, plasmids efficiently and robustly overcome these defenses during conjugation. Here, we show that the leading region of plasmids, which is the first to enter recipient cells, is a hotspot for an extensive repertoire of anti-defense systems, encoding anti-CRISPR, anti-restriction, anti-SOS, and other counter-defense proteins. We further demonstrate that focusing on these specific functional regions can lead to the discovery of diverse anti-defense genes. Promoters known to allow expression from ssDNA were prevalent in the leading regions, potentially facilitating rapid protection against bacterial immunity in the early stages of plasmid invasion. These findings reveal a new facet of plasmid dissemination and provide theoretical foundations for developing conjugative delivery systems for natural microbial communities.

microbiology↗

bioRxiv2022

Rapid and sensitive on-site genetic diagnostics of pest fruit flies using CRISPR-Cas12a

Bactrocera zonata, a major fruit pest species, is gradually spreading west from its native habitat in East Asia. In recent years it has become a major threat to the Mediterranean area, with the potential of invading Europe, the Americas, and Australia. To prevent its spreading, monitoring efforts in plantation sites and border controls are carried out. Despite these efforts, and due to morphological similarities between B. zonata and other pests in relevant developmental stages, the monitoring process is challenging, time-consuming, and requires external assistance from professional labs. CRISPR-Cas12a genetic diagnostics has been rapidly developing in recent years and provides an efficient tool for the genetic identification of pathogens, viruses, and other genetic targets. Here we design a CRISPR-Cas12a detection assay that differentially detects two major pest species, B. zonata and Ceratitis capitata. Our easy-to-use and affordable assay employs a simple DNA extraction technique together with isothermal amplification, and Cas12a-based detection. We demonstrate the specificity and high sensitivity of this method, and its relevance for on-site applications. This method is highly modular, and the presented target design method can be applied to a wide array of pests. Key MassageO_LIDistinguishing different pest fruit flies on-site is crucial for prevention of global spreading but can be difficult C_LIO_LIWe present a genetic identification assay for rapid, on-site detection of pest using CRISPR-Cas12a C_LIO_LIThe method is affordable, quick and easy-to-use, and can be applied in border controls or on-site C_LIO_LIThe design process can be easily tailored for any pest, and can greatly benefit developing countries C_LI

synthetic biology↗

bioRxiv2020

PCQC: Selecting optimal principal components for identifying clusters with highly imbalanced class sizes in single-cell RNA-seq data

SummaryPrior to identifying clusters in single cell gene expression experiments, selecting the top principal components is a critical step for filtering out noise in the data set. Identifying these top principal components typically focuses on the total variance explained, and principal components that explain small clusters from rare populations will not necessarily capture a large percentage of variance in the data. We present a computationally efficient alternative for identifying the optimal principal components based on the tails of the distribution of variance explained for each observation. We then evaluate the efficacy of our approach in three different single cell RNA-sequencing data sets and find that our method matches, or outperforms, other selection criteria that are typically employed in the literature. Availability and implementationpcqc is written in Python and available at github.com/RoussosLab/pcqc

bioinformatics↗

bioRxiv2020

B cells engineered to express an anti-HIV antibody allow memory retention, class switch recombination and clonal selection in mice

HIV viremia can be controlled by chronic antiretroviral therapy. As a potentially single-shot alternative, B cells engineered by CRISPR/Cas9 to express anti-HIV broadly neutralizing antibodies (bNAbs) were shown capable of secreting high antibody titers. Here, we demonstrate that, upon immunization of mice, adoptively transferred engineered B cells home to germinal centers (GC) where they predominate over the endogenous response and differentiate into memory and plasma cells while undergoing class switch recombination (CSR). Immunization with a higher affinity antigen increases accumulation in GCs and CSR rates. Boost immunization increases rates of engineered B cells in GCs and antibody secretion, indicating memory retention. Finally, antibody sequences of engineered B cells in the spleen show patterns of clonal selection. B cells may thus be engineered as a living and evolving drug.

immunology↗

bioRxiv2018

Processing of grassland soil C-N compounds into soluble and volatile molecules is depth stratified and mediated by genomically novel bacteria and archaea

Soil microbial activity drives the carbon and nitrogen cycles and is an important determinant of atmospheric trace gas turnover, yet most soils are dominated by organisms with unknown metabolic capacities. Even Acidobacteria, among the most abundant bacteria in soil, remain poorly characterized, and functions across groups such as Verrucomicrobia, Gemmatimonadetes, Chloroflexi, Rokubacteria are understudied. Here, we resolved sixty metagenomic, and twenty proteomic datasets from a grassland soil ecosystem and recovered 793 near-complete microbial genomes from 18 phyla, representing around one third of all organisms detected. Importantly, this enabled extensive genomics-based metabolic predictions for these understudied communities. Acidobacteria from multiple previously unstudied classes have genomes that encode large enzyme complements for complex carbohydrate degradation. Alternatively, most organisms we detected encode carbohydrate esterases that strip readily accessible methyl and acetyl groups from polymers like pectin and xylan, forming methanol and acetate, the availability of which could explain high prevalences of C1 metabolism and acetate utilization in genomes. Organism abundances among samples collected at three soil depths and under natural and amended rainfall regimes indicate statistically higher associations of inorganic nitrogen metabolism and carbon degradation in deep and shallow soils, respectively. This partitioning decreased in samples under extended spring rainfall indicating long term climate alteration can affect both carbon and nitrogen cycling. Overall, by leveraging natural and experimental gradients with genome-resolved metabolic profiles, we link organisms lacking prior genomic characterization to specific roles in complex carbon, C1, nitrate, and ammonia transformations and constrain factors that impact their distributions in soil.

microbiology↗

bioRxiv2018

Biological capacities clearly define a major subdivision in Domain Bacteria

ResumePhylogenetic analyses separate candidate phyla radiation (CPR) bacteria from other bacteria, but the degree to which their proteomes are distinct remains unclear. Here, we leveraged a proteome database that includes sequences from thousands of uncultivated organisms to identify protein families and examine their organismal distributions. We focused on widely distributed protein families that co-occur in genomes, as they likely foundational for metabolism. Clustering of genomes using the protein family presence/absence patterns broadly recapitulates the phylogenetic structure of the tree, suggesting persistence of core sets of protein families after lineage divergence. CPR bacteria group together and away from all other bacteria and archaea, in part due to novel proteins, some of which may be involved in cell-cell interactions. The diversity of combinations of protein families in CPR may exceed that of all other bacteria. Overall, the results extend the phylogeny-based suggestion that the CPR represent a major subdivision within Bacteria.

microbiology↗

bioRxiv2018

Hydrogen-based metabolism - An ancestral trait in lineages sibling to the Cyanobacteria

The metabolic machinery from which microbial aerobic respiration evolved is tightly linked to the origins of oxygenic Cyanobacteria (Oxyphotobacteria). Even though the majority of Oxyphotobacteria are photoautotrophs and can use carbohydrates with oxygen (O2) as the electron acceptor, all are fermenters under dark anoxic conditions. Studies suggest that the ancestor of Oxyphotobacteria may have used hydrogen (H2) as an electron donor and that two types of NiFe hydrogenases are essential for its oxidation. Melainabacteria and Sericytochromatia, close phylogenetic neighbors to Oxyphotobacteria comprise fermentative and aerobic representatives, or organisms capable of both. Margulisbacteria (candidate divisions RBX-1 and ZB3) and Saganbacteria (candidate division WOR-1), a novel cluster of bacteria phylogenetically related to Melainabacteria, Sericytochromatia and Oxyphotobacteria may further constrain the metabolic platform in which oxygenic photosynthesis and aerobic respiration arose. Here, we predict the metabolisms of Margulisbacteria and Saganbacteria from new and published metagenome-assembled genomes (MAGs) and single amplified genomes (SAGs), and compare them to their phylogenetic neighbors. Sediment-associated Margulisbacteria are predicted to have a fermentation-based metabolism featuring a variety of hydrogenases, a nitrogenase for nitrogen (N2) fixation, and electron bifurcating complexes involved in cycling of ferredoxin and NAD(P)H. Overall, the genomic features suggest the capacity for metabolic fine-tuning under strictly anoxic conditions. In contrast, the genomes of Margulisbacteria from the ocean ecosystem encode an electron transport chain that supports aerobic growth. Similarly, some Saganbacteria genomes encode various hydrogenases, and others may have the ability to use O2 under certain conditions via a putative novel type of heme copper O2 reductase. Like Melainabacteria and Sericytochromatia, Margulisbacteria and Saganbacteria have diverse energy metabolisms capable of fermentation, and aerobic or anaerobic respiration. In summary, our findings support the hypothesis that the ancestor of these groups was an anaerobe in which fermentation and H2 metabolism were central metabolic features. Our genomic data also suggests that contemporary lineages sibling to the Oxyphotobacteria may have acquired the ability to use O2 as a terminal electron acceptor under certain environmental conditions.

microbiology↗