Biology subjects

Borgwardt, K.

Publications and source records attributed to Borgwardt, K..

13 recordsLinked to original sources

bioRxiv2026

FastaLake selects compact metaproteomic databases from large sequence reservoirs by de novo peptide lookup

Metaproteomics analyses require protein sequence databases, yet the microbial proteins present in complex communities are not known in advance. Searching reference catalogs containing hundreds of millions of candidate sequences reduces identification sensitivity, while generating sample-specific databases through metagenomic sequencing adds time and cost. We introduce FastaLake, a computational workflow that uses full-length de novo peptide predictions by AlphaNovo to retrieve complete proteins from broad sequence resources and selects compact databases for conventional searching. In a defined eight-member community, FastaLake approached reference-search peptide yields without knowing its members. Using public sequence catalogs, FastaLake identified about as many peptides as matched metagenomic databases in two stool cohorts and 20-35% more in intestinal fluid at the same nominal 1% peptide false discovery rate. The resulting proteomes reveal personal functional profiles that persist across intestinal regions despite differences in leading species between intestinal fluid and stool. DNA, RNA and protein profiles agreed on broad differences between patients but only weakly on individual functions. All 726 acquisition-specific databases were smaller than the human reference proteome. Compact databases allowed non-specific searches of all 347 CapScan acquisitions, revealing histidine enrichment at peptide C termini in intestinal fluid. FastaLake enables metaproteomics analysis without matched metagenomic sequencing by constructing search databases from sequence resources and de novo peptide sequencing.

bioinformatics↗

bioRxiv2026

Joint protein, mRNA and DNA sequence design and optimization with nucleotide-level Potts models

Protein design typically generates amino acid sequences first and then translates them into DNA post-hoc limiting control over the nucleotide sequence via codon choice at design time. We introduce NuCaliby, a structure-conditioned model that jointly designs amino acid, mRNA, and DNA sequences of a protein at nucleotide resolution. NuCaliby predicts a Potts model over a nucleotide graph, trained using amino acid supervision by marginalizing over synonymous codons. It preserves residue-level designability on par with the original Caliby model. Its energy-based formulation supports composable inference-time guidance from sequence-level objectives through proposal construction and candidate acceptance, allowing the incorporation of biological proxies into the coding nucleotide sequence. Under organism-specific tRNA adaptation guidance, NuCaliby improves adaptation to host tRNA pools beyond synonymous sequence space, trading-off self-consistency. NuCaliby also embeds specified RNA motif sequences directly into coding sequences, avoiding the combinatorial growth required for the same task at the amino-acid level. In wet lab experiments, NuCaliby produces expressible and soluble proteins on par with state of the art methods, while enabling more complex nucleotide-level control and optimization. Together, these results extend protein design beyond the amino acid level, enabling joint optimization across the central dogma of molecular biology. The code and model weights are available here: https://github.com/blazejba/NuCaliby

biophysics↗

bioRxiv2026

Machine learning-guided discovery of a conserved plasmid proteomic signature enables MALDI-TOF MS detection of pOXA-48-carrying Enterobacterales

OXA-48 carbapenemases are among the most widespread and important resistance mechanisms in Enterobacterales. Yet detecting carbapenemases by conventional workflows necessitates additional testing, thus delaying optimization of therapy and implementation of infection control measures. Here, we present a machine learning approach that identifies the conserved pOXA-48 plasmid directly from routine MALDI-TOF spectra acquired for species identification. The model detects pOXA-48 carriers with an AUROC of 0.96-0.98 across two independent hospital cohorts and instrument platforms, indicating near-perfect discrimination. Using bottom-up proteomics, plasmid conjugation, and plasmid curing, we link the discriminative MALDI-TOF spectral features to proteins encoded on pOXA-48, with DUF1496 domain-containing protein producing the most discriminative spectral feature. Our approach reframes the resistance prediction task from inferring a resistance phenotype to detecting a conserved plasmid through its expressed proteomic signature and has the potential to enable rapid MALDI-TOF MS-based diagnostics for a wide range of plasmid-based resistance determinants.

microbiology↗

bioRxiv2025

Early detection of ampicillin susceptibility in Enterococcus faecium with MALDI-TOF MS and machine learning

BackgroundEnterococcus faecium can cause severe infections and is often resistant to the first-line antibiotic ampicillin. Consequently, clinicians usually prescribe broad-spectrum antibiotics, promoting the selection of multidrug-resistant bacteria. In this study, we investigate the application of machine learning techniques to detect ampicillin susceptibility directly from MALDI-TOF mass spectrometry. This technique could enable an earlier optimised treatment in infections with ampicillin-susceptible E. faecium. MethodsTwo datasets of clinical E. faecium MALDI-TOF spectra and their resistance phenotype were analysed: our own Technical University of Munich (TUM) dataset and the publicly available MS-UMG dataset. We tested logistic regression (LR) and LightGBM models on each dataset via nested cross-validation and explored transferability on the respective other dataset. ResultsLightGBM demonstrated slightly better performance than LR in identifying susceptible isolates in the TUM dataset (area under the precision-recall curve (AUPRC) 0.907 {+/-} 0.016 vs 0.902 {+/-} 0.030) as well as in the MS-UMG dataset (AUPRC 0.902 {+/-} 0.029 vs 0.899 {+/-} 0.054). External validation demonstrated good model transferability (AUPRC of 0.784 {+/-} 0.039 when trained on MS-UMG; 0.804 {+/-} 0.013 when trained on TUM). SHAP analysis consistently identified a top-ranked spectral feature corresponding to a peak at an m/z of 5091 in resistant isolate spectra. ConclusionThis study demonstrates that LR and LightGBM models can identify ampicillin-susceptible E. faecium isolates from MALDI-TOF spectra and generalise well to unseen datasets. While clinical implementation currently still requires confirmatory testing, the addition of larger datasets in the future will support the development of more robust machine learning models.

microbiology↗

bioRxiv2024

Bacterial protein function prediction via multimodal deep learning

Bacterial proteins are specialized with extensive functional diversity for survival in diverse and stressful environments. A significant portion of these proteins remains functionally uncharacterized, limiting our understanding of bacterial survival mechanisms. Hence, we developed Deep Expression STructure (DeepEST), a multimodal deep learning framework designed to accurately predict protein function in bacteria by assigning Gene Ontology (GO) terms. DeepEST comprises two modules: a multi-layer perceptron that takes gene expression and gene location as input features, and a protein structure-based predictor. Within DeepEST, we integrated these modules through a learnable weighted linear combination and introduced a novel masked loss function to fine-tune the structure-based predictor for bacterial species. These modeling choices are particularly well suited for bacteria due to the spatial organization of their circular genomes. Functionally related genes frequently co-localize and are co-transcribed within operons, allowing transcription dynamics to serve as crucial, condition-dependent regulatory signals. We show that DeepEST outperforms existing protein function prediction methods on a 25-species benchmark, relying solely on amino acid sequence or protein structure. Moreover, DeepEST predicts GO terms for unclassified hypothetical proteins across 25 human bacterial pathogens, facilitating the design of experimental setups for characterization studies. By combining expression, localization, and structure information in a unified deep learning framework, DeepEST bridges organism-specific data integration and structure-based transfer learning, providing a method tailored for bacterial protein function prediction in settings with structural and multi-condition expression data. AvailabilityAccompanying code and data can be found under https://github.com/BorgwardtLab/DeepEST.

bioinformatics↗

bioRxiv2024

Deep learning uncovers sequence-specific amplification bias in multi-template PCR

Multi-template polymerase chain reaction is a key step in many amplicon sequencing protocols enabling parallel amplification of diverse DNA molecules sharing common adapters in applications, ranging as wide as quantitative molecular biology and DNA data storage. However, this process results in a skewed amplicon abundance, due to sequence-specific amplification biases. In this study, one-dimensional convolutional neural networks (1D-CNNs) were trained on synthetic DNA pools to learn the PCR amplification efficiency of individual templates. These 1D-CNN models can predict poorly amplifying templates based solely on sequence information, achieving an AUROC/AUPRC of up to 0.88/0.44 with very imbalanced prevalence of 2%, thereby greatly outperforming baseline models relying only on GC content and nucleotide frequency as predictors. A new, general-purpose framework for interpreting deep learning models, termed CluMo provides mechanistic insights into the amplification biases. Most strikingly, specific amplification reactions were identified as suffering from adaptor-template self-priming a mechanism previously disregarded in PCR.

molecular biology↗

bioRxiv2023

Scalable covariance-based connectivity inference for synchronous neuronal networks

We present a novel method for inferring connectivity from large-scale neuronal networks with synchronous activity. Our approach leverages Dynamic Differential Covariance to address the associated computational challenges. First, we analyze spike trains generated from Leaky Integrate-and-Fire network simulations and evaluate the performance of several off-the-shelf multivariate connectivity inference methods. Next, we introduce a new approach, Fractional Dynamic Differential Covariance (FDDC), and demonstrate that it consistently outperforms the other methods. Finally, we apply FDDC to experimental data to assess the topological organization of inferred graphs of in vitro neural network recordings obtained using high-density microelectrode arrays (HD-MEAs). Our results indicate that FDDC-derived graphs exhibit a significant negative correlation between small-worldness and measures of network synchrony. In contrast, graphs inferred through the well-established pairwise correlation method do not show such a correlation. This finding implies that the graphs obtained through FDDC provide stronger evidence in support of the theoretical notion that networks with clustered connections tend to exhibit higher levels of synchronizability. We hypothesize that our findings have implications for the development of scalable connectivity inference methods for large-scale neural network data.

neuroscience↗

bioRxiv2022

SIMBSIG: Similarity search and clustering for biobank-scale data

SummaryIn many modern bioinformatics applications, such as statistical genetics, or single-cell analysis, one frequently encounters datasets which are orders of magnitude too large for conventional in-memory analysis. To tackle this challenge, we introduce SIMBSIG, a highly scalable Python package which provides a scikit-learn-like interface for out-of-core, GPU-enabled similarity searches, principal component analysis, and clustering. Due to the PyTorch backend it is highly modular and particularly tailored to many data types with a particular focus on biobank data analysis. AvailabilitySIMBSIG is freely available from PyPI and its source code and documentation can be found on GitHub (https://github.com/BorgwardtLab/simbsig) under a BSD-3 license. Contactmichael.adamer@bsse.ethz.ch

bioinformatics↗

bioRxiv2022

Improved MALDI-TOF MS based antimicrobial resistance prediction through hierarchical stratification

MotivationPredicting antimicrobial resistance using MALDI-TOF mass spectrometry based machine learning is a fast-growing field of research. Recent advances in machine learning methods specifically designed for MALDI-TOF mass spectra have outperformed established classification approaches. However, classification performance was observed to have a large standard deviation between different train-test splits. We hypothesise that this variance is caused by the underlying phylogenetic structure between microbial samples, which is implicitly reflected in their MALDI-TOF MS profiles, but not taken into account during the training of a model. ResultsIn this paper, we propose to infer this structure from the dataset--using agglomerative hierarchical clustering--and consider it during the dataset splitting between train and test. We show that incorporating such phylogenetic structure into the antimicrobial resistance prediction scenario leads to an improved classification performance. Average precision was increased from 42.3 to 47.1 for ciprofloxacin resistance prediction in Escherichia coli and from 44.6 to 50.8 for amoxicillin-clavulanic acid resistance prediction in Staphylococcus aureus using a Gaussian process classifier with a MALDI-TOF MS specific kernel. We envision that these results will support the quick and reliable identification of antimicrobial resistances, thus increasing patient well-being and reducing healthcare costs. AvailabilityAll data is available for download and code available as an easy-to-use Python package under https://github.com/BorgwardtLab/maldi_PIKE at branch maldi_stratification. Contactcaroline.weis@bsse.ethz.ch, karsten.borgwardt@bsse.ethz.ch Supplementary informationSupplementary information at the end of document.

bioinformatics↗

bioRxiv2022

Downregulating α-synuclein in iPSC-derived dopaminergic neurons mimics electrophysiological phenotype of the A53T mutation

Parkinsons disease (PD) is a common debilitating neurodegenerative disorder, characterized by a progressive loss of dopaminergic (DA) neurons. Mutations, gene dosage increase, and single nucleotide polymorphisms in the -synuclein-encoding gene SNCA either cause or increase the risk for PD. However, neither the function of -synuclein in health and disease, nor its role throughout development is fully understood. Here, we introduce DeePhys, a new tool that allows for data-driven functional phenotyping of neuronal cell lines by combining electrophysiological features inferred from high-density microelectrode array (HD-MEA) recordings with a robust machine learning workflow. We apply DeePhys to human induced pluripotent stem cell (iPSC)-derived DA neuron-astrocyte co-cultures harboring the prominent SNCA mutation A53T and an isogenic control line. Moreover, we demonstrate how DeePhys can facilitate the assessment of cellular and network-level electrophysiological features to build functional phenotypes and to evaluate potential treatment interventions. We find that electrophysiological features across all scales proved to be highly specific for the A53T phenotype, enabled to predict the genotype and age of individual cultures with high accuracy, and revealed a mutant-like phenotype after downregulation of -synuclein.

neuroscience↗

bioRxiv2021

networkGWAS: A network-based approach for genome-wide association studies in structured populations

While the search for associations between genetic markers and complex traits has led to the discovery of tens of thousands of trait-related genetic variants, the vast majority of these only explain a small fraction of observed phenotypic variation. One possible strategy to detect stronger associations is to aggregate the effects of several genetic markers and to test entire genes, pathways or (sub)networks of genes for association to a phenotype. The latter, network-based genome-wide association studies, in particular suffers from a vast search space and an inherent multiple testing problem. As a consequence, current approaches are either based on greedy feature selection, thereby risking that they miss relevant associations, or neglect doing a multiple testing correction, which can lead to an abundance of false positive findings. To address the shortcomings of current approaches of network-based genome-wide association studies, we propose networkGWAS, a computationally efficient and statistically sound approach to network-based genome-wide association studies using mixed models and neighborhood aggregation. It allows for population structure correction and for well-calibrated p-values, which are obtained through circular and degree-preserving network permutation schemes. networkGWAS successfully detects known associations on semi-simulated common variants from A. thaliana and on simulated rare variants from H. sapiens, as well as neighborhoods of genes involved in stress-related biological processes on a stress-induced phenotype from S. cerevisiae. It thereby enables the systematic combination of gene-based genome-wide association studies with biological network information. Availabilityhttps://github.com/BorgwardtLab/networkGWAS.git Contactgiulia.muzio@bsse.ethz.ch, karsten.borgwardt@bsse.ethz.ch

genomics↗

bioRxiv2020

Direct Antimicrobial Resistance Prediction from MALDI-TOF mass spectra profile in clinical isolates through Machine Learning

Early administration of effective antimicrobial treatments is critical for the outcome of infections. Antimicrobial resistance testing enables the selection of optimal antibiotic treatments, but current culture-based techniques take up to 72 hours. We have developed a novel machine learning approach to predict antimicrobial resistance directly from MALDI-TOF mass spectra profiles of clinical samples. We trained calibrated classifiers on a newly-created publicly available database of mass spectra profiles from clinically most relevant isolates with linked antimicrobial susceptibility phenotypes. The dataset combines more than 300,000 mass spectra with more than 750,000 antimicrobial resistance phenotypes from four medical institutions. Validation against a panel of clinically important pathogens, including Staphylococcus aureus, Escherichia coli, and Klebsiella pneumoniae, resulting in AUROC values of 0.8, 0.74, and 0.74 respectively, demonstrated the potential of using machine learning to substantially accelerate antimicrobial resistance determination and change of clinical management. Furthermore, a retrospective clinical case study found that implementation of this approach would have resulted in a beneficial change in the clinical treatment in 88% (8/9) of cases. MALDI-TOF mass spectra based machine learning may thus be an important new tool for antibiotic stewardship.

microbiology↗

bioRxiv2020

Large-scale DNA-based phenotypic recording and deep learning enable highly accurate sequence-function mapping

Predicting quantitative effects of gene regulatory elements (GREs) on gene expression is a longstanding challenge in biology. Machine learning models for gene expression prediction may be able to address this challenge, but they require experimental datasets that link large numbers of GREs to their quantitative effect. However, current methods to generate such datasets experimentally are either restricted to specific applications or limited by their technical complexity and error-proneness. Here we introduce DNA-based phenotypic recording as a widely applicable and practical approach to generate very large datasets linking GREs to quantitative functional readouts of high precision, temporal resolution, and dynamic range, solely relying on sequencing. This is enabled by a novel DNA architecture comprising a site-specific recombinase, a GRE that controls recombinase expression, and a DNA substrate modifiable by the recombinase. Both GRE sequence and substrate state can be determined in a single sequencing read, and the frequency of modified substrates amongst constructs harbouring the same GRE is a quantitative, internally normalized readout of this GREs effect on recombinase expression. Using next-generation sequencing, the quantitative expression effect of extremely large GRE sets can be assessed in parallel. As a proof of principle, we apply this approach to record translation kinetics of more than 300,000 bacterial ribosome binding sites (RBSs), collecting over 2.7 million sequence-function pairs in a single experiment. Further, we generalize from these large-scale datasets by a novel deep learning approach that combines ensembling and uncertainty modelling to predict the function of untested RBSs with high accuracy, substantially outperforming state-of-the-art methods. The combination of DNA-based phenotypic recording and deep learning represents a major advance in our ability to predict quantitative function from genetic sequence.

synthetic biology↗