Biology subjects

Abeln, S.

Publications and source records attributed to Abeln, S..

10 recordsLinked to original sources

bioRxiv2026

SST-MAE: Learning Spectral-Spatio-Temporal Representations from Plant Hyperspectral Time Series to Discover Complex Genotype-Phenotype Relations

Understanding the link between genetic variation and observable traits is key to crop breeding. Hyperspectral imaging captures physiological and biochemical profiles, but current supervised methods require costly trait annotations and treat each observation as a static snapshot, ignoring the temporal dynamics of plant development. We introduce SST-MAE, a self-supervised framework that learns genotype-discriminative representations from plant hyperspectral developmental trajectories, without requiring phenotypic labels. The model learns to reconstruct masked information, capturing multiple growth trajectories. Validated on 194 field-grown lettuce genotypes across eight time points, the frozen encoder serves as a feature extractor for downstream genotype classification. SST-MAE outperforms raw spectral and linear baselines, achieving AUROC > 0.89 for anthocyanin pigmentation SNPs and 0.77 for leaf serration. The learned features are highly label-efficient, attaining near-full performance with only 30-50% of labeled data, offering a scalable pathway toward high-throughput genetic screening from image-based phenotypes.

bioinformatics↗

bioRxiv2026

DeepPheno: A Deep Learning Framework for Linking Hyperspectral Imaging and SNP Genotypes in Lettuce

While whole-genome sequencing captures millions of single nucleotide polymorphisms (SNPs) and hyperspectral imaging (HSI) enables non-destructive plant phenotyping, integrating these modalities to link genotype to phenotype remains challenging due to their high dimensionality and non-linearity. This study presents DeepPheno a deep learning framework that predicts SNP genotypes from HSI data, using model predictability as a proxy for genotype-phenotype association. HSI data were acquired from 194 lettuce genotypes under field conditions. HSI data patches (20x20 pixels x 224 spectral bands) were used to train a hybrid CNN to predict the variant of a specific SNP. The framework was validated on SNPs with known phenotypic effects (anthocyanin, leaf serration, pale pigmentation), achieving high predictive performance (AUC ranging from 0.806 to 0.935), whereas models trained on randomly shuffled labels performed at chance (mean AUC {approx} 0.51). Extending the workflow to 50 randomly selected putatively neutral SNPs, most yielded low predictability, but two showed high performance (AUC > 0.76), suggesting uncharacterized genotype-phenotype links. Explainable AI, including SHAP and Grad-CAM, identified relevant spectral and spatial features driving these predictions, particularly the green and red-edge wavelengths associated with pigment dynamics and leaf structure. These results establish a framework for understanding complex genotype-phenotype interactions in plants and extracting these links from HSI data without predefining the exact trait values. It provides an avenue for high-throughput trait discovery and description and extends the integration of image-based phenomics with plant genetics.

plant biology↗

bioRxiv2026

Beyond the annotated: protein foundation models enable robust prediction of microbial root competence

BackgroundRoot competence, the ability of soil bacteria to establish and grow on plant roots, is a key ecological trait influencing plant nutrition, growth, and health. However, identifying genomic determinants of root competence across bacteria remains challenging, in part because model generalisability depends strongly on how genomes are represented. Traditional approaches based on curated annotations are incomplete and biased toward well-characterised organisms and functions, limiting generalisation. Sequence-similarity clustering improves coverage but yields high-dimensional features relative to dataset size, hindering training. Foundation models offer an alternative by learning com-pact representations without relying on prior annotation. ResultsHere, we compared pretrained genome representations from protein and DNA foundation models (ESM-2, Bacformer, DNABERT-S) with annotation- and clustering-based features (KEGG orthology, OrthoFinder protein families) for predicting root competence using synthetic microbial community data from Arabidop-sis thaliana and assessed generalisability across bacteria. When training and test sets contained taxonomically related bacteria, most approaches performed similarly. However, when test bacteria belonged to phyla entirely absent from training, reflecting high evolutionary separation across all levels of bacterial classification, only pretrained protein representations retained predictive performance. Bacformer-derived representations, which incorporate genomic context, supported the strongest generalisation, suggesting that conserved genomic organisation contributes to predicting root competence. Feature attribution quantifying protein contributions to model decisions linked root competence to TonB/SusD-dependent receptors, small-molecule transporters, and unannotated proteins with conserved regulatory motifs and homology to carbon starvation-response loci. ConclusionsProtein foundation models support generalisation across evolutionarily distant bacteria and identify genomic determinants of root competence, including unannotated proteins.

genomics↗

bioRxiv2025

Condition-dependent, amorphous protein agglomerates control cytoplasmic rheology

Molecular crowding in the bacterial cytoplasm restricts the diffusion of large molecules, impacting cellular processes. However, how nutrient availability influences cytoplasmic rheology is not well understood. With single-particle tracking in Escherichia coli, we observed a threefold variation in the diffusion of a 40-nm particle across exponential growth conditions. Previously suggested determinants of rheology did not account for this variation; instead, we found a strong anticorrelation between the diffusion coefficient and the abundance of amino acid metabolism proteins, persisting upon genetic perturbations and showing that lower diffusion is associated with increased viscoelasticity. Photoactivated light microscopy revealed that some amino acid metabolism proteins form clusters. Electron microscopy showed that these proteins could form amorphous agglomerates at physiological concentrations in vitro, likely driven by their low intrinsic disorder, high compactness and hydropathy score. These findings show that protein agglomerates regulate cytoplasmic rheology in a condition-dependent manner, suggesting an underappreciated level of cytoplasmic organization. HighlightsO_LIDiffusion of 40-nm particles varies threefold across growth conditions in E. coli C_LIO_LICytoplasmic diffusion inversely correlates with COG-E protein abundance COG-E proteins form agglomerates that increase cytoplasmic viscoelasticity C_LIO_LIProtein compactness and hydrophobicity predict condition-dependent crowding effects C_LI

biophysics↗

bioRxiv2024

Tumor break load quantitates structural variant-associated genomic instability with biological and clinical relevance across cancers

While structural variants (SVs) are a clear sign of genomic instability, they have not been systematically quantified per patient. Therefore, the biological and clinical impact of high numbers of SVs in patients is unknown. We introduce tumor break load (TBL), defined as the sum of unbalanced SVs, as a measure for SV-associated genomic instability. Using pan-cancer data from TCGA, PCAWG, and CCLE, we show that a high TBL is associated with significant changes in gene expression in 26/31 cancer types that consistently involve upregulation of DNA damage repair and downregulation of immune response pathways. Patients with a high TBL show a higher risk of recurrence and shorter median survival times for 5/15 cancer types. Our data demonstrate that TBL is a biologically and clinically relevant feature of genomic instability that may aid patient prognostication and treatment stratification. For the datasets analyzed in this study, TBL has been made available in cBioPortal.

cancer biology↗

bioRxiv2024

Data-driven evaluation of suitable immunogens for improved antibody selection

Antibodies are indispensable in laboratory and clinical applications due to their high specificity and affinity for protein antigens. However, selecting the right protein fragments as immunogens for antibody production remains challenging. Leveraging the Human Protein Atlas, this study systematically evaluates immunogen properties aiming to identify key factors that influence their suitability. Antibodies were classified as successful or unsuccessful based on standardized validation experiments, and the structural and functional properties of their immunogens were analyzed. Results indicated that longer immunogens often resulted in more successful but less specific antibodies. Shorter immunogens (50 residues or fewer) with disordered or unfolded regions at the N- or C-terminus and long coil stretches, were more likely to generate successful antibodies. Conversely, immunogens with high beta sheet content, multiple secondary structure elements, transmembrane regions, or disulfide bridges were associated with poorer antibody performance. Post-translational modification sites within immunogens appeared to mark beneficial regions for antibody generation. To support antibody selection, a novel R package, immunogenViewer, was developed, enabling researchers to easily apply these insights when immunogen sequences are disclosed. By providing a deeper understanding of immunogen suitability, this study promotes the development of more effective antibodies, ultimately addressing issues of reproducibility and reliability in antibody-based research. The findings are highly relevant to the research community, as end-users often lack control over the immunogen selection process in antibody production.

bioinformatics↗

bioRxiv2024

Comparison of sequence- and structure-based antibody clustering approaches on simulated repertoire sequencing data

Repertoire sequencing allows us to investigate the antibody-mediated immune response. The clustering of sequences is a crucial step in the data analysis pipeline, aiding in the identification of functionally related antibodies. The conventional clustering approach of clonotyping relies on sequence information, particularly CDRH3 sequence identity and V/J gene usage, to group sequences into clonotypes. It has been suggested that the limitations of sequence-based approaches to identify sequence-dissimilar but functionally converged antibodies can be overcome by using structure information to group antibodies. Recent advances have made structure-based methods feasible on a repertoire level. However, so far, their performance has only been evaluated on single-antigen sets of antibodies. A comprehensive comparison of the benefits and limitations of structure-based tools on realistic and diverse repertoire data is missing. Here, we aim to explore the promise of structure-based clustering algorithms to replace or augment the standard sequence-based approach, specifically by identifying low-sequence identity groups. Two methods, SAAB+ and SPACE2, are evaluated against clonotyping. We curated a dataset of well-annotated pairs of antibodies that show high overlap in epitope residues and thus bind the same region within their respective antigen. This set of antibodies was introduced into a simulated repertoire to compare the performance of clustering approaches on a diverse antibody set. Our analysis reveals that structure-based methods do produce more multiple-occupancy clusters compared to clonotyping. However, it also highlights the limitations associated with the need for same-length CDR regions by SPACE2. This work thoroughly compares the utility of different clustering methods and provides insights into what further steps are required to effectively use antibody structural information to group immune repertoire data.

bioinformatics↗

bioRxiv2023

Proteome encoded determinants of protein sorting into extracellular vesicles

Extracellular vesicles (EVs) are membranous structures released by cells into the extracellular space and are thought to be involved in cell-to-cell communication. While EVs and their cargo are promising biomarker candidates, protein sorting mechanisms of proteins to EVs remain unclear. In this study, we ask if it is possible to determine EV association based on the protein sequence. Additionally, we ask what the most important determinants are for EV association. We answer these questions with explainable AI models, using human proteome data from EV databases to train and validate the model. It is essential to correct the datasets for contaminants introduced by coarse EV isolation workflows and for experimental bias caused by mass spectrometry. In this study, we show that it is indeed possible to predict EV association from the protein sequence: a simple sequence-based model for predicting EV proteins achieved an area under the curve of 0.77{+/-}0.01, which increased further to 0.84{+/-}0.00 when incorporating curated post-translational modification (PTM) annotations. Feature analysis shows that EV associated proteins are stable, polar, and structured with low isoelectric point compared to non-EV proteins. PTM annotations emerged as the most important features for correct classification; specifically palmitoylation is one of the most prevalent EV sorting mechanisms for unique proteins. Palmitoylation and nitrosylation sites are especially prevalent in EV proteins that are determined by very strict isolation protocols, indicating they could potentially serve as quality control criteria for future studies. This computational study offers an effective sequence-based predictor of EV associated proteins with extensive characterisation of the human EV proteome that can explain for individual proteins which factors contribute to their EV association.

bioinformatics↗

bioRxiv2021

PIPENN: Protein Interface Prediction with an Ensemble of Neural Nets

MotivationProtein interactions play an essential role in many biological and cellular processes, such as protein-protein interaction (PPI) in signaling pathways, binding to DNA in transcription, and binding to small molecules in receptor activation or enzymatic activity. Experimental identification of protein binding interface residues is a time-consuming, costly, and challenging task. Several machine learning and other computational approaches exist which predict such interface residues. Here we explore if Deep Learning (DL) can be used effectively for this prediction task, and which learning strategies and architectures may be most efficient. We introduce seven DL architectures that are applied to eleven independent test sets, focused on the residues involved in PPI interfaces and in binding RNA/DNA and small molecule ligands. ResultsWe constructed a large data set dubbed BioDL, comprising protein-protein interaction data from the PDB and protein-ligand interactions (DNA, RNA and small molecules) from the BioLip database. Additionally, we reused our existing curated homo- and heteromeric PPI data sets. We performed several experiments to assess the impact of different data features, spatial forms, encoding schemes, network initializations, loss functions, regularization mechanisms, and activation functions on the performance of the predictors. Benchmarking the resulting DL models with an independent test set (ZK448) shows no single DL architecture performs best on all instances, but that an ensemble of DL architectures consistently achieves peak prediction performance. Our PIPENNs ensemble predictor outperforms current state-of-the-art sequence-based protein interface predictors on all interaction types, achieving AUCs of 0.718 (protein-protein), 0.823 (protein-nucleotide) and 0.842 (protein- small molecule) respectively. AvailabilitySource code and data sets at https://github.com/ibivu/pipenn/ Contactr.haydarlou@vu.nl

bioinformatics↗

bioRxiv2020

SeRenDIP-CE: Sequence-based Interface Prediction for Conformational Epitopes

MotivationAntibodies play an important role in clinical research and biotechnology, with their specificity determined by the interaction with the antigens epitope region, as a special type of protein-protein interaction (PPI) interface. The ubiquitous availability of sequence data, allows us to predicting epitopes from sequence in order to focus time-consuming wet-lab experiments onto the most promising epitope regions. Here, we extend our previously developed sequence-based predictors for homodimer and heterodimer PPI interfaces to predict epitope residues that have the potential to bind an antibody. ResultsWe collected and curated a high quality epitope dataset from the SAbDaB database. Our generic PPI heterodimer predictor obtained an AUC-ROC of 0.666 when evaluated on the epitope test set. We then trained a random forest model specifically on the epitope dataset, reaching AUC 0.694. Further training on the combined heterodimer and epitope datasets, improves our final predictor to AUC 0.703 on the epitope test set. This is better than the best state-of-the-art sequence-based epitope predictor BepiPred-2.0. On one solved antibody-antigen structure of the COVID19 virus spike RNA binding domain, our predictor reaches AUC 0.778. We added the SeRenDIP-CE Conformational Epitope predictors to our webserver, which is simple to use and only requires a single antigen sequence as input, which will help make the method immediately applicable in a wide range of biomedical and biomolecular research. AvailabilityWebserver, source code and datasets are available at www.ibi.vu.nl/programs/serendipwww/ Contactk.a.feenstra@vu.nl

bioinformatics↗