Biology subjects

Uhl, M.

Publications and source records attributed to Uhl, M..

9 recordsLinked to original sources

bioRxiv2026

CRISCross: Integration of Epigenetic Information for Deep-Learning-Based Prediction of CRISPR-Cas9 Off-Targets Across Cell Types

CRISPR-Cas9 enables precise genome editing but remains limited by off-target effects, raising safety concerns and necessitating accurate prediction. Although epigenetic features influence Cas9 binding and cleavage, their predictive signals are often weak, highlyvariable, and context-dependent, limiting the effectiveness of existing machine learning approaches. Cross-cell-type settings are especially challenging: epigenetic landscapes differ substantially, and models trained on one cell type frequently fail to generalize. Here, we systematically investigate the role of epigenetic context in off-target prediction and demonstrate that its utility depends on modeling interactions with sequence features rather than treating it as an independent signal. Tabular and deep learning experiments demonstrate that epigenetic features provide a consistent but modestly complementary signal, and that late-stage fusion strategies are insufficient to effectively exploit this information. To address this, we introduce CRISCross, a cross-attention-based architecture that captures sequence-epigenetic interactions at the nucleotide level, alongside CRISPRAT, a synthetic pretraining strategy that learns transferable representations from genome-derived sgRNA-off-target pairs without relying on dataset-specific pretraining. CRISCross, incorporating AlphaGenome-derived chromatin accessibility signals, shows better predictive performance than sequence-only baselines across three independent test datasets spanning two cell types, demonstrating improved generalization across guide RNAs, datasets, and cellular contexts. This is particularly relevant for therapeutic applications involving cell types or patient-derived contexts that lack dedicated experimental datasets. In such settings, predicted chromatin accessibility may provide scalable contextual information when direct epigenomic profiling is unavailable or impractical. Together, these results show that accurate CRISPR off-target prediction benefits from interaction-aware integration of epigenetic context alongside generalizable pretraining strategies.

bioinformatics↗

bioRxiv2026

Biological Foundation Models Enable CRISPR Array Detection Without Metagenomic Assembly

Accurate identification of CRISPR arrays is essential for studying prokaryotic adaptive immunity, yet existing tools struggle with short-read sequencing data and arrays containing degenerate repeats. These limitations restrict CRISPR analysis in metagenomic and fragmented genomic datasets. We present a foundation model-based approach for CRISPR array detection that addresses both these challenges. We fine-tune a large genomic foundation model using the Parameter-Efficient Fine-Tuning (PEFT) method, Low-Rank Adaptation (LoRA) to perform per-nucleotide classification of DNA sequences into repeat, spacer, and non-array regions directly from raw input nucleotide sequences. We develop two model variants for different sequence context lengths. The long-context model supporting sequences of up to 8,192 nucleotides achieves 98.16% test accuracy and identifies degenerate repeat candidates missed by similarity-based CRISPR detection tools. The short-context model supports sequences of up to 150 nucleotides, optimized for Illumina reads, reaches 90.03% accuracy and enables direct analysis of individual reads without assembly. On simulated metagenomic data, it achieves a spacer recall of 49.12% and recovers 12.57% of spacers that are otherwise not detected by dedicated metagenomic CRISPR array detection methods which require metagenomic assembly. Together, these results demonstrate that genomic foundation models provide a robust and complementary paradigm for CRISPR array detection.

molecular biology↗

bioRxiv2025

CochleaNet: deep learning-based image analysis for cochlear connectomics and gene therapy

With the emergence of gene and optogenetic therapies targeting deafness, the comprehensive analysis of the molecular anatomy and physiology of the cochlea has become ever more important. Here, we introduce CochleaNet, a deep learning-based framework to analyze volumetric imaging data obtained by light-sheet microscopy of decalcified, cleared and fluorescently labeled cochleae. CochleaNet covers the workflow from reconstruction of the cochlea to segmentation of inner hair cells, spiral ganglion neurons and their afferent synapses, to analyzing the expression of gene therapy products. We validated CochleaNet by comparison to manual image analysis. Trained on high isotropic resolution mouse data, CochleaNet was also applicable to the cochlea of the gerbil, another relevant animal model, and lower-resolution mouse data from a commercially available microscope. We conclude that the combination of light-sheet microscopy and image analysis with CochleaNet paves the way for rapid and reliable quantification of cochlear molecular anatomy and preclinical gene therapy outcomes.

neuroscience↗

bioRxiv2025

Unravelling the role of epigenetic regulators during embryonic development of Rhipicephalus micropolus

Epigenetic modifications are long-lasting changes to the genome that influence a cells transcriptional potential, thereby altering its function. These modifications can trigger adaptive responses that impact protein expression and various cellular processes, including differentiation and growth. The primary epigenetic mechanisms identified to date include DNA and RNA methylation, histone modifications, and microRNA-mediated regulation of gene expression. The intricate crosstalk among these mechanisms makes epigenetics a compelling field for the development of novel control strategies, particularly through the use of epigenetic drugs targeting arthropod vectors such as ticks. In this study, we identified the Rhipicephalus microplus orthologs of canonical histone-modifying enzymes, along with components of the machinery responsible for m5C and 6mA-DNA, and m6A-RNA methylations. We further characterized their transcriptional profiles and enzymatic activities during embryonic development. To explore the functional consequences of epigenetic regulation in R. microplus, we evaluated the effects of various epigenetic inhibitors on the BME26 tick embryonic cell line. Molecular docking simulations were performed to predict the binding mode of these inhibitors to tick enzymes, followed by in vitro assessment of their effects on cell viability and morphology. Tick cells exposed to these inhibitors exhibited phenotypic and molecular alterations. Notably, we observed higher levels of DNA methylation in the mitochondrial genome compared to nuclear DNA. Inhibition of DNA methylation using 5-azacytidine (5-AZA) was associated with increased activity of the mitochondrial electron transport chain and ATP synthesis, but reduced cellular proliferation. Our findings highlight the importance of epigenetic regulation during tick embryogenesis and suggest that targeting these pathways may offer a novel and promising strategy for tick control. HighlightsO_LIR. microplus presents a complete set of epigenetic enzymes that modify histones, DNA and RNA C_LIO_LIEpigenetic regulation is highly dynamic, and functionally significant during R. microplus embryogenesis C_LIO_LIInhibition of DNA methylation leads to overactivation of the mitochondrial electron transport chain C_LI

molecular biology↗

bioRxiv2024

RNA-Protein Interaction Classification via Sequence Embeddings

RNA-protein interactions (RPI) are ubiquitous in cellular organisms and essential for gene regulation. In particular, protein interactions with non-coding RNAs (ncRNAs) play a critical role in these processes. Experimental analysis of RPIs is time-consuming and expensive, and existing computational methods rely on small and limited datasets. This work introduces RNAInterAct, a comprehensive RPI dataset, alongside RPIembeddor, a novel transformer-based model designed for classifying ncRNA-protein interactions. By leveraging two foundation models for sequence embedding, we incorporate essential structural and functional insights into our task. We demonstrate RPIembeddors strong performance and generalization capability compared to state-of-the-art methods across different datasets and analyze the impact of the proposed embedding strategy on the performance in an ablation study.

bioinformatics↗

bioRxiv2024

SelNeTime: a python package inferring effective population size and selection intensity from genomic time series data

Genomic samples collected from a single population over several generations provide direct access to the genetic diversity changes occurring within a specific time period. This provides information about both demographic and adaptive processes acting on the population during that period. A common approach to analyze such data is to model observed allele counts in finite samples using a Hidden Markov Model (HMM) where hidden states are true allele frequencies over time (i.e. a trajectory). The HMM framework allows one to compute the full likelihood of the data, while accounting both for the stochastic evolution of population allele frequencies along time and for the noise arising from sampling a limited number of individuals at possibly spread out generations. Several such HMM methods have been proposed so far, differing mainly in the way they model the transition probabilities of the Markov chain. Following Paris et al. (2019a), we consider here the Beta with Spikes approximation, which avoids the computational issues associated to the Wright-Fisher model while still including fixation probabilities, in contrast to other standard approximations of this model like the Gaussian or Beta distributions. To facilitate the analysis and exploitation of genomic time series data, we present an improved version of Paris et al. (2019a) s approach, denoted SelNeTime, whose computation time is drastically reduced and which accurately estimates effective population size (assuming no selection) or the selection intensity at each locus (given a previously estimated value of N). We also evaluate the performance of this method in realistic situations where selection is present and both demography and selection need to be inferred. SelNeTime is implemented in a user friendly python package, which can also easily simulate genomic time series data under a user-defined evolutionary model and sampling design.

evolutionary biology↗

bioRxiv2024

Microtubule networks in zebrafish hair cells facilitate presynapse transport and fusion during development

Sensory cells in the retina and inner ear rely on specialized ribbon synapses for neurotransmission. Disruption of these synapses is linked to visual and auditory dysfunction, but it is unclear how these unique synapses form. Ribbon synapses are defined by a presynaptic density called a ribbon. Using live imaging in zebrafish hair cells, we find that numerous small ribbon precursors are present throughout the cell early in development. As development progresses, fewer large ribbons remain, and localize at the presynaptic active zone (AZ). Using tracking analyses, we show that ribbon precursors exhibit directed motion along an organized microtubule network to reach the presynaptic AZ. In addition, we show that ribbon precursors can fuse together on microtubules. Using pharmacology, we find that microtubule disruption interferes with ribbon motion, fusion, and normal synapse formation. Overall, this work demonstrates a dynamic series of events that underlies the formation of a critical synapse required for sensory function.

developmental biology↗

bioRxiv2019

GraphProt2: A novel deep learning-based method for predicting binding sites of RNA-binding proteins

CLIP-seq is the state-of-the-art technique to experimentally determine transcriptome-wide binding sites of RNA-binding proteins (RBPs). However, it relies on gene expression which can be highly variable between conditions, and thus cannot provide a complete picture of the RBP binding landscape. This creates a demand for computational methods to predict missing binding sites. Here we present GraphProt2, a computational RBP binding site prediction framework based on graph convolutional neural networks (GCNs). In contrast to current CNN methods, GraphProt2 offers native support for the encoding of base pair information as well as variable length input, providing increased flexibility and the prediction of nucleotide-wise RBP binding profiles. We demonstrate its superior performance compared to GraphProt and two CNN-based methods on single as well as combined CLIP-seq datasets. Conceived as an end-to-end method, GraphProt2 includes all necessary functionalities, from dataset generation over model training to the evaluation of binding preferences and binding site prediction. Various input types and features are supported, accompanied by comprehensive statistics and visualizations to inform the user about datatset characteristics and learned model properties. All this makes GraphProt2 the most versatile and complete RBP binding site prediction method available so far.

bioinformatics↗

bioRxiv2018

MechRNA: prediction of lncRNA mechanisms from RNA-RNA and RNA-protein interactions

MotivationLong non-coding RNAs (lncRNAs) are defined as transcripts longer than 200 nucleotides that do not get translated into proteins. Often these transcripts are processed (spliced, capped, polyadenylated) and some are known to have important biological functions. However, most lncRNAs have unknown or poorly understood functions. Nevertheless, because of their potential role in cancer, lncRNAs are receiving a lot of attention, and the need for computational tools to predict their possible mechanisms of action is more than ever. Fundamentally, most of the known lncRNA mechanisms involve RNA-RNA and/or RNA-protein interactions. Through accurate predictions of each kind of interaction and integration of these predictions, it is possible to elucidate potential mechanisms for a given lncRNA.\n\nApproachHere we introduce MechRNA, a pipeline for corroborating RNA-RNA interaction prediction and protein binding prediction for identifying possible lncRNA mechanisms involving specific targets or on a transcriptome-wide scale. The first stage uses a version of IntaRNA2 with added functionality for efficient prediction of RNA-RNA interactions with very long input sequences, allowing for large-scale analysis of lncRNA interactions with little or no loss of optimality. The second stage integrates protein binding information pre-computed by GraphProt, for both the lncRNA and the target. The final stage involves inferring the most likely mechanism for each lncRNA/target pair. This is achieved by generating candidate mechanisms from the predicted interactions, the relative locations of these interactions and correlation data, followed by selection of the most likely mechanistic explanation using a combined p-value.\n\nResultsWe applied MechRNA on a number of recently identified cancer-related lncRNAs (PCAT1, PCAT29, ARLnc1) and also on two well-studied lncRNAs (PCA3 and 7SL). This led to the identification of hundreds of high confidence potential targets for each lncRNA and corresponding mechanisms. These predictions include the known competitive mechanism of 7SL with HuR for binding on the tumor suppressor TP53, as well as mechanisms expanding what is known about PCAT1 and ARLn1 and their targets BRCA2 and AR, respectively. For PCAT1-BRCA2, the mechanism involves competitive binding with HuR, which we confirmed using HuR immunoprecipitation assays.\n\nAvailabilityMechRNA is available for download at https://bitbucket.org/compbio/mechrna\n\nContactbackofen@informatik.uni-freiburg.de, cenksahi@indiana.edu\n\nSupplementary informationSupplementary data are available at Bioinformatics online.

bioinformatics↗