Biology subjects

Elbracht, M.

Publications and source records attributed to Elbracht, M..

5 recordsLinked to original sources

bioRxiv2026

A relational architecture of the human genome

Complex systems exhibit organization not only through their components but also via their interdependencies. Local nucleotide dependencies underpin regulatory motifs and RNA structures, yet principles governing a higher-order organization across the human genome remain unknown. Here we systematically perturbed every position of the com- plete telomere-to-telomere human genome using Nucleotide Transformer v3 with an ex- ascale compute cluster, generating nucleotide-resolution dependency maps that reveal a directed, weighted network of position-to-position genomic relationships. This network recapitulates known biological features and uncovers recurrent, directional architec- tures termed 'relons', spanning coding, non-coding, and unannotated regions. These re- lational modules reflect evolutionary and functional constraints and associate with hu- man genetic variation and pathogenic variants. Our findings suggest a higher-order rela- tional layer of genome organization, providing a framework to interpret DNA as a complex system of dependencies beyond linear and 3D structure, with implications for under- standing genetic variation and disease.

genomics↗

bioRxiv2025

Genolator: A Multimodal Large Language Model Fusing Natural Language, Genomic, and Structural Tokens for Protein Function Interpretation

BackgroundDecoding the genetic code to unveil its genome functionality is a monumental task which would greatly advance the understanding of disease mechanisms and development of targeted treatment approaches. Although large language models (LLMs) have transformed natural language processing across diverse domains, translating the complex language of DNA into human-readable form remains challenging due to genomic data complexity and unexplored regions of the human genome. Current (genomic) language models either have a solid understanding of natural language or of the genomic code. Models fusing both aspects are largely lacking. ResultsHere we present Genolator, a multimodal large language model that integrates embeddings from DNA sequences, amino acid sequences, and protein structures with natural language queries. Fine-tuned on over 370,000 question-answer pairs generated using abstracted Gene-Ontology (GO) terms, Genolator effectively answers queries regarding protein subcellular localization, molecular function, and biological processes. Evaluation demonstrates high accuracy in confirming or denying protein function associations, outperforming baseline models such as openly available allrounder LLMs like GPT 4.1 as well as smaller domain specific models integrating knowledge from foundation models like Evo2 and ESM-2. Explorations of the Genolators hidden states unveil a biologically and linguistically plausible organization of its learned representations. ConclusionGenolator enhances accessibility to genomic information by enabling natural language interaction with protein data, facilitating biological discovery, and clinical research. It represents a step towards bridging genomic code and human language through the integration of a multimodal LLM.

bioinformatics↗

bioRxiv2024

Utilizing protein structure graph embeddings to predict the pathogenicity of missense variants

BackgroundGenetic variants can impact the structure of the corresponding protein, which can have detrimental effects on protein function. While the effect of protein truncating variants is often easier to evaluate, most genetic variants are missense variants. These variants are mostly single nucleotide variants which result in the exchange of a single amino acid. The effect on protein function of these variants can be challenging to deduce. To aid the interpretation of missense variants a variety of bioinformatic algorithms have been developed, yet current algorithms rarely directly use the protein structure as a feature to consider. ResultsWe developed a machine learning workflow that utilizes the protein-language-model ESMFold to predict the protein structure of missense variants, which is subsequently embedded using graph autoencoders. The generated embeddings are used in a classifier model which predicts pathogenicity. We provide evidence that the generated graph embeddings improve classification accuracy of a XGBoost pathogenicity predictor, which should lead to a wide applicability for human genetic diseases. Additionally, we explored different levels of abstraction of the graph embeddings and their influence on the classifier. Finally, we compare the utility of graph embeddings from different protein-folding models.

bioinformatics↗

bioRxiv2024

Mutational Constraint Analysis Workflow for Overlapping Short Open Reading Frames and Genomic Neighbours

Understanding the dark genome is a priority task following the complete sequencing of the human genome. Short open reading frames (sORFs) are a group of largely unexplored elements of the dark genome with the potential for being translated into microproteins. The definitive number of coding and regulatory sORFs is not known, however they could account for up to 1-2% of the human genome. This corresponds to an order of magnitude in the range of canonical coding genes. For a few sORFs a clinical relevance has already been demonstrated, but for the majority of potential sORFs the biological function remains unclear. A major limitation in predicting their disease relevance using large-scale genomic data is the fact that no population-level constraint metrics for genetic variants in sORFs are yet available. To overcome this, we used the recently released gno-mAD 4.0 dataset and analysed the constraint of a consensus set of sORFs and their genomic neighbours. We demonstrate that sORFs are mostly embedded into a moderately constraint genomic context, but within the gencode dataset we identified a subset of highly constrained sORFs comparable to highly constrained canonical genes.

bioinformatics↗

bioRxiv2018

Aberrant DEGS1 sphingolipid metabolism impairs central and peripheral nervous system function in humans

Sphingolipids including ceramides are important components of cellular membranes and functionally associated with fundamental processes such as cell differentiation, neuronal signaling and myelin sheath formation. Defects in the synthesis or degradation of sphingolipids are associated with various neurological pathologies, however, the entire spectrum of disorders affecting sphingolipid metabolism remains elusive. By whole-exome sequencing in a patient with a multisystem neurological disorder of both the central and peripheral nervous system, we identified a homozygous variant p.(Ala280Val) in DEGS1, encoding an enzyme of the ceramide synthesis pathway. The blood sphingolipid profile and patient-derived fibroblasts both showed a significant shift from the unsaturated to the dihydro-forms of sphingolipids. Moreover, an atypical and potentially toxic sphingolipid metabolite is formed as consequence of the altered synthesis pathway. The changes in the sphingolipid profile were recapitulated in a CRISPR/Cas-based DEGS1 knockout HAP1-cell model and by chemical inhibition of DEGS1, suggesting a loss of DEGS1 function in the disease. DEGS1 insufficiency is thus a novel cause for a multisystem neurological disorder. A sphingolipid-rich diet may correct the metabolic profile and improve the clinical outcome of affected individuals and suggests that this heritable condition might be treatable.\n\nAbbreviations

genetics↗