Biology subjects

Schug, A.

Publications and source records attributed to Schug, A..

16 recordsLinked to original sources

bioRxiv2026

Conditional Generation And Inpainting Of Non-coding RNA Sequences With Masked Discrete Diffusion

Designing functional non-coding RNA (ncRNA) is fundamental to synthetic biology and RNA therapeutics, yet generative modelling for ncRNA has received far less attention than protein design. We present RNA-MDLM, a framework that extends Masked Discrete Language Models (MDLM) to the conditional generation and inpainting of ncRNA. We make two additions: first, conditioning on RNA-type representations from a pretrained RNA language model, and second, a modified classifier-free guidance scheme (Mod-CFG) that interpolates among conditional, unconditional, and random-sequence probabilities for better control. We also introduce REPAINT GAMES, a benchmark of seven structured masking tasks to probe a model's performance on sequence patterns, structural motifs, and base-pairing per RNA type. Our model is trained on 4.6 million ncRNA sequences spanning six evaluable classes. It produces sequences whose composition and folding statistics closely match natural RNAs. Through extensive ablation studies, we find that the embedding-conditioned model achieves the best balance of structural fidelity, biological novelty, and inpainting accuracy, and its class label steers generation far more strongly than a plain label baseline. We further show that a model trained on a smaller, class-balanced subset can appear more realistic mainly by copying abundant natural sequences rather than learning their rules. We also benchmark against a masked-diffusion model and a family-specific VAE on ribozyme families, and find that a type-conditioned model like ours and a per-family model are solving different tasks, which must be accounted for in a fair comparison. We will release the code and the trained models.

biophysics↗

bioRxiv2026

ER-stress signaling and Alzheimer's proteins adjust the quality of human protein synthesis

Proteostasis is the balance of protein synthesis, protein maintenance and protein degradation. Proteostasis is disturbed in neurodegenerative disorders like Alzheimers disease (AD) of the aging human body. Protein synthesis by the ribosome is the most error-prone process in gene expression. If and how the error-rate of protein synthesis is regulated during human aging and contributes to AD is unknown. Here we show that ribosomal error-rate is adapted in cellular models of human aging, but not in mouse aging. This adaptation involves ER-stress signaling and the Alzheimers disease-related proteins amyloid-beta precursor protein and presenilin 1. Our results suggest that ribosomal error-rate is a relevant parameter in human aging and disease.

cell biology↗

bioRxiv2025

NucleicBERT: Deciphering the language of nucleic acids

The vast majority of the human genome comprises non-protein-coding regions whose structural and functional roles remain poorly understood. Many of these regions function through RNA, yet progress in deep learning for RNA has lagged behind proteins because most methods rely on abundant structural labels or evolutionary alignments, both sparse for RNA. To address these challenges, we developed NucleicBERT, a self-supervised masked-language model that learns contextual representations capturing local and distal dependencies without requiring alignments or evolutionary information. Explainable AI analysis reveals that the model clusters RNA types in latent space and attends to structural properties like secondary structure and tertiary contacts, effectively "rediscovering" RNA biology from sequence correlations alone. When fine-tuned for downstream structural and functional tasks, NucleicBERT requires only single sequences, yet surpasses current state-of-the-art RNA models. This alignment-free framework addresses the scarcity of annotated 3D RNA data while providing a rapid, computational complement to experimental techniques. By bridging abundant unlabeled primary sequence corpora with more scarce structural annotations, NucleicBERT advances RNA structure prediction and provides insights into the working of LLMs. NucleicBERT is available at https://github.com/KIT-MBS/NucleicBERT.

biophysics↗

bioRxiv2024

NucleoSeeker - Precision filtering of RNA databases to curate high-quality datasets

The structural prediction of biomolecules via computational methods complements the often involved wet-lab experiments. Un-like protein structure prediction, RNA structure prediction remains a significant challenge in bioinformatics, primarily due to the scarcity of annotated RNA structure data and its varying quality. Many methods have used this limited data to train deep learning models but redundancy, data leakage and bad data quality hampers their performance. In this work, we present NucleoSeeker, a tool designed to curate high-quality, tailored datasets from the Protein Data Bank (PDB) database. It is a unified framework that combines multiple tools and streamlines an otherwise complicated process of data curation. It offers multiple filters at structure, sequence and annotation levels, giving researchers full control over data curation. Further, we present several use cases. In particular, we demonstrate how NucleoSeeker allows the creation of a non-redundant RNA structure dataset to assess AlphaFold3s performance for RNA structure prediction. This demonstrates NucleoSeekers effectiveness in curating valuable non-redundant tailored datasets to both train novel and judge existing methods. NucleoSeeker is very easy to use, highly flexible and can significantly increase the quality of RNA structure datasets.

bioinformatics↗

bioRxiv2024

Leveraging Experimental Vasculature Data for High Resolution Brain Tumor Simulations

Cancer remains a leading cause of mortality. Multidisciplinary studies probe its complex pathology to increase treatment options. Computational modeling of tumor growth on high-performance computing resources offers microscopic insight into its progress and a valuable avenue for advancing our understanding. However, the effective initialization and parameterization of the underlying models require high-resolution data from real tissue structures. Here, we leveraged high performance computing resources and a massive dataset of a mouse brains entire vascular network. We processed these image stacks into detailed three-dimensional representations, identified brain regions of interest, and conducted a series of large-scale simulations to investigate how tumor growth is influenced by local vascular network characteristics. By simulating tumor growth with sub-cellular resolution, we can probe to which extent vessel density and vessel network length influence tumor growth. We determined that vessel density is the primary determinant of growth rate. Finally, our results allowed us to extrapolate tumor cell growth predictions for the entire mouse brain, highlighting the critical role of vascular topology in tumor progression. Such increasingly realistic simulations of cancer cells and their microenvironment enable researchers to increasingly bridge the gap between basic biology and clinical practice, ultimately supporting the development of more effective personalized cancer therapies.

cancer biology↗

bioRxiv2024

Influence of Contact Map Topology on RNA Structure Prediction

The available sequence data of RNA molecules have highly increased in the past years. Unfortunately, while computational power is still under exponential growth, the computer prediction quality from sequence to final structure is still inferior to the labour intensive experimental work. Although a reliable end-to-end procedure was already developed for proteins since Alphafold2, while its successor AlphaFold3 can also predict RNA, its confidence in particular for novel sequences and folds appear still limited. Another strategy entails two steps: (i) predicting potential contacts in the form of a contact maps from evolutionary data, and (ii) simulating the molecule with a physical force field while using the contact map as restraint. However, the quality of the structure prediction crucially depends on the quality of the contact map. Until now, only the proportion of true positive contacts was considered as a quality characteristic. We propose to also include the distribution of these contacts, and have done so in our recent studies. We observed that the clustering of contacts, as is common for many AI algorithms, has a negative impact on prediction quality. In contrast, a more distributed topology is beneficial. We have applied these findings from computer experiments to current algorithms and introduced a measure of distribution, the Gaussian score.

biophysics↗

bioRxiv2022

Development of a scoring function for comparing simulated and experimental tumor spheroids

Enormous progress continues in the field of cancer biology, yet much remains to be unveiled regarding the mechanisms of cancer invasion. In particular, complex biophysical mechanisms enable a tumor to remodel the surrounding extracellular matrix (ECM), thus allowing cells to escape and invade alone or as multicellular collectives. Tumor spheroids cultured in collagen represent a simplified, reproducible 3D model system, which is sufficiently complex to recapitulate the evolving internal organization of cells and external interaction with the ECM that occur during invasion. Recent experimental approaches enable high resolution imaging and quantification of the internal structure of invading tumor spheroids. Concurrently, computational modeling enables simulations of complex multicellular aggregates based on first principles. The comparison between real and simulated spheroids represents a way to fully exploit both data sources, but remains a challenging task. We hypothesize that comparing any two spheroids requires first the extraction of basic features from the raw data, and second the definition of key metrics to match such features. Here, we present a novel data-agnostic method to compare spatial features of spheroids in 3D. To do so, we define and extract features from spheroid point cloud data, which we simulated using Cells in Silico (CiS), a high-performance framework for large-scale tissue modeling previously developed by our group. We then define metrics to compare features between individual spheroids, and combine all metrics into an overall deviation score. Finally, we use our features to compare experimental data on invading spheroids in increasing collagen densities. We propose that our approach represents the basis for defining improved metrics to compare large 3D data sets. Moving forward, this approach will enable informing in silico spheroids based on their in vitro counterparts, and vice versa, thus enabling both basic and applied researchers to close the loop between modeling and experiments in cancer research. Author summaryCells within a tumor use various methods to escape and thereby invade into healthy parts of the body. These methods are studied experimentally by examining tumor spheroids, spherical aggregates of hundreds to thousands of individual cells. Such spheroids can also be simulated, and the comparison of simulation and experiment is desirable. Here, we present an analysis strategy for the comparison of tumor spheroids, a widely used workhorse of cancer research. Using this strategy, we aim to improve the collaborative potential between experimentalists and theorists.

cancer biology↗

bioRxiv2021

Mechanical influences on in silico tumor evolution

Experimental insight and conceptual understanding of tumor growth are steadily growing and leading to new therapeutic interventions. Experiments and clinical studies are able to link single-cell properties to macroscopic tumor attributes. The development of cellular subpopulations in heterogeneous tumors can be understood as an evolutionary system with different cell types competing over both space and nutrients. However, to predict the growth trajectory and development of a tumor, fitness and trade-offs of cell properties in the context of the surroundings are required and often inaccessible. The optimum of the evolutionary trajectory provides a target for intervention, but can mostly not be identified. We propose that the optimal value of cellular properties is influenced by the tumor surrounding. Computational multiscale-modeling of tissue enables the observation of the trajectory of each cell while modeling the tumor surrounding. We model a 3D spheroid tumor and the fitness of individual cells and the evolutionary behavior of the tumor are quantified and linked to global parameters. Cell-cell adhesion and cell motility are two important mechanical properties for cell development and used as free parameters. Mechanical properties alone are able to drive the tumor towards low adhesion.We implement a dynamically changing nutrient surrounding representing the fluctuating blood-supply through blood vessel collapse and angiogenesis. We find that the evolutionary speed depends on the frequency of the fluctuations. We identify a frequency domain in which the evolutionary speed is significantly increased over a tumor with constant nutrient supply. The findings suggest that mechanically-induced fluctuations can accelerate tumor evolution. Author summaryLimited space and nutrients together with competing cell types drive an evolutionary process inside tumors. This process selects for the fittest cell types and optimizes the growing behavior for its local surroundings. An expanding tumor exerts mechanical forces on its cells and its surroundings, leading to a fluctuating nutrient supply through collapsing blood vessels. Here, we observe the influence of a dynamically changing surrounding on the evolutionary behavior of heterogeneous tumors in a high-resolution computational model. We find that the evolutionary speed depends on the frequency of the fluctuations and a fitness advantage of low-adhesion cells.

cancer biology↗

bioRxiv2020

The effect of salt on the dynamics of CoV-2 RBD at ACE2

In this article, we investigate the effect of electrolytes on the stability of the complex between the coronavirus type 2 spike protein receptor domain (CoV-2 RBD) and ACE2, which plays an important role in the activation cascade at the viral entry of CoV-2 into human cells. At the cellular surface, electrolytes play an important role, especially in the interaction of proteins near the membrane surface. Additionally, the binding interface of the CoV-2 RBD - ACE2 complex is highly hydrophilic. We simulated the CoV-2 RBD - ACE2 complex at varying salt concentrations over the concentration range from 0.03 M to 0.3 M of calcium and sodium chloride over an individual simulation length of 750 ns in 9 independent simulations (6.75 {micro}s total). We observe that the CoV-2 RBD - ACE2 complex is stabilized independent of the salt concentration. We identify a strong negative electrostatic potential at the N-terminal part of CoV-2 RBD and we find that CoV-2 RBD binds even stronger at higher salt concentrations. We observe that the dynamics of the N-terminal part of CoV-2 RBD stabilize the protein complex leading to strong collective motions and a stable interface between CoV-2 RBD and ACE2. We state that the sequence of CoV-2 RBD might be optimized for a strong binding to ACE2 at varying salt concentrations at the cellular surface, which acts as a key component in the activation of CoV-2 for its viral entry. SIGNIFICANCEA novel coronavirus, coronavirus type 2 (CoV-2), was identified as primary cause for a worldwide pandemic of the severe acute respiratory syndrome (SARS CoV-2). The CoV-2 spike protein is a major target for the development of a vaccine and potential strategies to inhibit the viral entry into human cells. At the cellular surface, CoV-2 activation involves the direct interaction between ACE2 and CoV-2 RBD. At the cellular surface, electrolytes play an important role, especially in the interaction of proteins near the membrane surface. We thus investigate the effect of ion conditions on the interaction of the CoV-2 RBD - ACE2 complex and find stabilizing effects. We speculate that CoV-2 RBD is optimized for strong binding to ACE2 at varying salt concentrations.

biophysics↗

bioRxiv2020

Emerging Tumor Development by Simulating Single-cell Events

Despite decades of substantial research, cancer remains a ubiquitous scourge in the industrialized world. Effective treatments require a thorough understanding of macroscopic cancerous tumor growth out of individual cells. Clinical imaging methods, however, only detect late-stage macroscopic tumors, while many quantitative experiments focus on small clusters of cancerous cells in microscopic detail but struggle to grow full tumors in-vitro. Here, we introduce the critical scale-bridging link between both these scopes. We are able to simulate the growth of mm-sized tumors composed of 1.5 million m-resolved individual cells by employing highly parallelized code on a supercomputer. We observe the competition for resources and space, which can lead to hypoxic or necrotic tissue regions. Cellular mutations and tumor stem cells can lead to tissue heterogeneity and change tumor properties. We probe the effects of different chemotherapy and radiotherapy treatments and observe selective pressure. This improved theoretical understanding of cancer growth as emerging behavior from single-cells opens new avenues for various scientific fields, ranging from developing better early-stage cancer detection devices to testing treatment regimes in-silico for personalized medicine. Author summaryExperimental and microscopy techniques are rapidly advancing biology and the observability of tissue. The theoretical understanding of tissue either focuses on a few cells or continuous tissue. Here we introduce the scale-bridging theoretical link that is able to model single cells as well as tissue consisting of millions of those cells, harvesting the power of modern supercomputers. We close the gap between single-cells and tissue through access to the full time-resolved trajectories of each cell and the emerging behavior of the tissue. We apply our framework on a generalized model for tumor growth. Tumor heterogeneity, as well as tumor stem cells are introduced, and the changes of behavior in response to cancer treatments is observed and validated.

biophysics↗

bioRxiv2020

CoCoNet: Boosting RNA contact prediction by convolutional neural networks

Physics-based co-evolutionary models such as direct coupling analysis (DCA) in combination with machine learning (ML) techniques based on deep neural networks are able to predict protein contact maps with astonishing accuracy. Such contacts can be used as constraints in structure prediction and massively increase prediction accuracy. Unfortunately, the same ML methods cannot readily be applied to RNA as they rely on large structural datasets only available for proteins but not for RNAs. Here, we demonstrate how the small amount of data available for RNA can be used to significantly improve prediction of RNA contact maps. We introduce an algorithm called CoCoNet that is based on a combination of a Coevolutionary model and a shallow Convolutional Neural Network. Despite its simplicity and the small number of trained parameters, the method boosts the contact prediction accuracy by about 70% with respect to straightforward DCA as tested by cross-validation on a dataset of about sixty RNA structures. Both our extensive robustness tests and the limited number of parameters allow the generalization properties of our model. Finally, applications to other RNAs highlight the power of our approach. CoCoNet is freely available and can be found at https://github.com/KIT-MBS/coconet.

bioinformatics↗

bioRxiv2020

Inhibition of Corona virus spike protein binding to ACE2

In this paper, we investigate the molecular assembly processes of a Coronavirus Spike protein fragment, the hexapeptide YKYRYL on the ACE2 receptor and its inhibitory effect on the aggregation and activation of the CoV-2 spike receptor protein at the same receptor protein. In agreement with an experimental study, we find a high affinity of the hexapeptide to the binding interface between the spike receptor protein and ACE2, which we investigate using 20 independent equilibrium MD simulations over a total of 1 s and a 200 ns enhanced MD simulation. We then evaluate the effect of the hexapeptide on the aggregation process of the spike receptor protein to ACE2 in long-time enhanced MD simulations. In that set of simulations, we find that the spike receptor protein does not bind to ACE2 with the binding motif shown in experiments, but it rotates due to an electrostatic repulsion and forms a hydrophobic interface with ACE2. Surprisingly, we observe that the hexapeptide binds to the spike receptor domain, which has the effect that this protein only weakly attaches to ACE2, so that the activation of the spike protein receptor might be inhibited in this case. Our results indicate that the hexapeptide might be a possible treatment option which prevents the viral activation through the inhibition of the interaction between ACE2 and the spike receptor protein. SIGNIFICANCEA novel coronavirus, CoV-19 and a later phenotype CoV-2 were identified as primary cause for a severe acute respiratory syndrome (SARS CoV-2). The spike (S) protein of CoV-2 is one target for the development of a vaccine to prevent the viral entry into human cells. The inhibition of the direct interaction between ACE2 and the S-protein could provides a suitable strategy to prevent the membrane fusion of CoV-2 and the viral entry into human cells. Using MD simulations, we investigate the assembly process of a Coronavirus Spike protein fragment, the hexapeptide YKYRYL on the ACE2 receptor and its inhibitzory effect on the aggregation and activation of the CoV-2 spike receptor protein at the same receptor protein.

biophysics↗

bioRxiv2020

Refinement of protein structure via contact based potentials in replica exchange simulations

Proteins are complex biomolecules which perform critical tasks in living organisms. Knowledge of a proteins structure is essential for understanding its physiological function in detail. Despite the incredible progress in experimental techniques, protein structure determination is still expensive, time-consuming, and arduous. That is why computer simulations are often used to complement or interpret experimental data. Here, we explore how in silico protein structure determination based on replica exchange molecular dynamics can benefit from including contact information derived from theoretical and experimental sources, such as direct coupling analysis or NMR spectroscopy. To reflect the influence from erroneous and noisy data we probe how false-positive contacts influence the simulated ensemble. Specifically, we integrate varying numbers of randomly selected native and non-native contacts and explore how such a bias can guide simulations towards the native state. We investigate the number of contacts needed for a significant enrichment of native-like conformations and show the capabilities and limitations of this method. Adhering to a threshold of approximately 75% true-positive contacts within a simulation, we obtain an ensemble with native-like conformations of high quality. We find that contact-guided REX MD is capable of delivering physically reasonable models of a proteins structure. Author summaryProtein structure prediction, that is obtaining a protein structure starting from a sequence using any computational method, is a great challenge. Over the past years a broad variety of methods evolved, ranging from algorithms for "blind" or de novo predictions using Monte-Carlo or physics-based biomolecular simulation methods to algorithms transferring structure information obtained from known homologous proteins. Recently, purely data-driven approaches using neural networks have shown to be capable of predicting high-quality structures. However, some local structural motifs are only poorly resolved and need further refinement. Here, we explore to what extent contact information helps guiding replica exchange molecular dynamics towards the native fold. By adding a contact pair bias potential to the energy function, we effectively guide the search towards the target structure by narrowing the conformational space to be sampled. We find that such an energetic bias, even if containing false-positive contacts to a certain extent, greatly enhances the refinement process and improves the chance of finding the native state in a single run.

biophysics↗

bioRxiv2019

Evaluating DCA-based method performances for RNA contact prediction by a well-curated dataset

RNA molecules play many pivotal roles in the cellular functioning that are still not fully understood. Any detailed understanding of RNA function requires knowledge of its three-dimensional structure, yet experimental RNA structure resolution remains demanding. Recent advances in sequencing provide unprecedented amounts of sequence data that can be statistically analysed by methods such as Direct Coupling Analysis (DCA) to determine spatial proximity or contacts of specific nucleic acid pairs, which improve the quality of structure prediction. To quantify this structure prediction improvement, we here present a well curated dataset of about seventy RNA structures with high resolution and compare different nucleotide-nucleotide contact prediction methods available in the literature. We observe only minor difference between the performances of the different methods. Moreover, we discuss how these predictions are robust for different contact definitions and how strongly depend on procedures used to curate and align the families of homologous RNA sequences.

bioinformatics↗

bioRxiv2019

pydca v1.0: a comprehensive software for Direct Coupling Analysis of RNA and Protein Sequences

The ongoing advances in sequencing technologies have provided a massive increase in the availability of sequence data. This made it possible to study the patterns of correlated substitution between residues in families of homologous proteins or RNAs and to retrieve structural and stability information. Direct coupling Analysis (DCA) infers coevolutionary couplings between pairs of residues indicating their spatial proximity, making such information a valuable input for subsequent structure prediction. Here we present pydca, a standalone Python-based software package for the DCA of protein- and RNA-homologous families. It is based on two popular inverse statistical approaches, namely, the mean-field and the pseudo-likelihood maximization and is equipped with a series of functionalities that range from multiple sequence alignment trimming to contact map visualization. Thanks to its efficient implementation, features and user-friendly command line interface, pydca is a modular and easy-to-use tool that can be used by researchers with a wide range of backgrounds.\n\nAvailabilityhttps://github.com/KIT-MBS/pydca

bioinformatics↗

bioRxiv2018

Pcdh18a-positive tip cells instruct notochord formation in zebrafish

The notochord defines the axial structure of all vertebrates during development. Notogenesis is a result of major cell reorganization in the mesoderm, the convergence and the extension of the axial cells. However, it is currently not known how these processes act together in a coordinated way during notochord formation. Analysing the tissue flow, we determined the displacement of the axial mesoderm and identified, relative to the ectoderm, an actively-migrating notochord tip cell population and a group of trailing notochordal plate cells. Molecularly, these tip cells express Protocadherin18a, a member of the cadherin superfamily. We show that Pcdh18a-mediated recycling of E-cadherin adhesion complexes transforms these tip cells into a cohesive and fast migrating cell group. In turn, these tip cells subsequently instruct the trailing mesoderm. We simulated cell migration during early mesoderm formation using a lattice-based mathematical framework, and predicted that the requirement for an anterior, local motile cell cluster could guide the intercalation of the posterior, axial cells. Indeed, grafting experiments validated the predictions and induced ectopic notochord-like rods. Our findings indicate that the tip cells influence the trailing mesodermal cell sheet by inducing the formation of the notochord.

developmental biology↗