Biology subjects

Molnar, F.

Publications and source records attributed to Molnar, F..

5 recordsLinked to original sources

bioRxiv2025

Bridging Prediction and Reality: Comprehensive Analysis of Experimental and AlphaFold 2 Full-Length Nuclear Receptor Structures

AlphaFold 2 has revolutionized protein structure prediction, yet systematic evaluations of its performance against experimental structures for specific protein families remain limited. Here we present the first comprehensive analysis comparing AlphaFold 2-predicted and experimental nuclear receptor structures, examining root-mean-square deviations, secondary structure elements, domain organization, and ligand-binding pocket geometry. While AlphaFold2 achieves high accuracy in predicting stable conformations with proper stereochemistry, it shows limitations in capturing the full spectrum of biologically relevant states, particularly in flexible regions and ligand-binding pockets. Statistical analysis reveals significant domain-specific variations, with ligand-binding domains showing higher structural variability (CV = 29.3%) compared to DNA-binding domains (CV = 17.7%). Notably, Alphafold 2 systematically underestimates ligand-binding pocket volumes and captures only single conformational states in homodimeric receptors where experimental structures show functionally important asymmetry. These findings provide critical insights for structure-based drug design targeting nuclear receptors and establish a framework for evaluating Alphafold 2 predictions across other protein families.

bioinformatics↗

bioRxiv2025

CSMA: an ImageJ Plugin for the Analysis of Wound Healing Assays

Accurate quantification of wound closure in cell migration assays is crucial yet challenging. Still, existing methods often underperform due to omitting cell detection within the wound area, resulting in biased outcomes. To overcome this limitation, we developed the CSMA plugin for ImageJ. CSMA utilizes advanced image processing techniques, including contrast enhancement, edge detection, and morphological operations, to precisely identify and quantify cells in the wound region. The plugin offers user-friendly features and adjustable parameters to accommodate different imaging conditions, ensuring robust performance across diverse experimental setups. Validation against conventional tools confirms CSMAs superior ability to delineate wound boundaries and provide accurate estimations of area and width at every time point. As applied to SW480-ADH colon cancer cells treated with various compounds, CSMA proves valuable in biomedical research. It represents a significant advancement in wound healing assay analysis, providing researchers with a simple and reliable tool for studying cell migration dynamics with enhanced precision and reproducibility. CSMA is available as an ImageJ plugin and source code at https://github.com/AminaSagymbayeva/CSMA_WoundHealing.

bioinformatics↗

bioRxiv2024

LipiDetective - a deep learning model for the identification of molecular lipid species in tandem mass spectra

Lipids are involved in many vital processes within the cell, and alterations in lipid homeostasis have been associated with various diseases such as cancer or type 2 diabetes. Confidently identifying lipids in samples is a prerequisite for understanding the multiple functions lipids fulfill in health and disease. However, the accurate identification of molecular lipid species based on tandem mass spectrometry data is still a key challenge in lipidomics. Most current approaches rely on using a custom pipeline to process and match the measured spectra against an in-house spectra reference library, which hinders the comparability of results. To address this challenge, a transformer model called LipiDetective was developed and trained on a dataset composed of reference spectra measured from lipid standards, spectra from databases, and published experiments, utilizing both shotgun as well as liquid-chromatography mass spectrometry. LipiDetective demonstrates, for the first time, that artificial neural networks can learn the characteristic lipid fragmentation patterns to automatically and accurately annotate molecular lipids species in tandem mass spectra independently of the experimental setup. The model can even correctly predict lipid species for which it has never seen a spectrum before as it is able to generalize the learned lipid fragmentation patterns. Analysis of the integrated gradients reveals that LipiDetective focuses on relevant peaks that can be matched to known fragments and are thus humanly interpretable. Therefore, LipiDetective has the potential to be a valuable tool to aid in the lipid identification process and support the comparability of results from different sources. Aside from Lipidetective as a "ready-to-use" application, this work primarily offers a deeper understanding of how the model functions and how future deep learning models for lipid identification in mass spectra could be improved.

bioinformatics↗

bioRxiv2021

MoSBi: Automated signature mining for molecular stratification and subtyping

The improving access to increasing amounts of biomedical data provides completely new chances for advanced patient stratification and disease subtyping strategies. This requires computational tools that produce uniformly robust results across highly heterogeneous molecular data. Unsupervised machine learning methodologies are able to discover de-novo patterns in such data. Biclustering is especially suited by simultaneously identifying sample groups and corresponding feature sets across heterogeneous omics data. The performance of available biclustering algorithms heavily depends on individual parameterization and varies with their application. Here, we developed MoSBi (Molecular Signature identification using Biclustering), an automated multi-algorithm ensemble approach that integrates results utilizing an error model-supported similarity network. We evaluated the performance of MoSBi on transcriptomics, proteomics and metabolomics data, as well as synthetic datasets covering various data properties. Profiting from multi-algorithm integration, MoSBi identified robust group and disease specific signatures across all scenarios overcoming single algorithm specificities. Furthermore, we developed a scalable network-based visualization of bicluster communities that support biological hypothesis generation. MoSBi is available as an R package and web-service to make automated biclustering analysis accessible for application in molecular sample stratification.

bioinformatics↗

bioRxiv2020

Predictability of cortical-cortical connections in the mammalian brain

Despite a five-order magnitude range in size, the mammalian brain exhibits many shared anatomical and functional characteristics that should translate into cortical network commonalities. Here we develop a framework employing machine learning to quantify the degree of predictability of the weighted interareal cortical matrix. Data were obtained with retrograde tract-tracing experiments supplemented by projection length measurements. Using this framework with consistent and edge-complete empirical datasets in the macaque and mouse cortex, we show that there is significant amount of predictability embedded in the interareal cortical networks of both species. At the binary level, links are predictable with an Area Under the ROC curve of at least 0.8 for the macaque. At the weighted level, strengths of the medium and strong links are predictable with at least 85-90% accuracy in mouse and 70-80% in macaque, whereas weak links are not predictable in either species. These observations suggest that the formation and evolution of the cortical network at the mesoscale is to a large extent, rule-based, motivating further research on the architectural invariants of the cortical connectome.

neuroscience↗