Biology subjects

Probst, D.

Publications and source records attributed to Probst, D..

6 recordsLinked to original sources

bioRxiv2026

Acquiring Improved Protein Variants With Probabilistic Preferential Learning

Variant effect prediction (VEP) models can be used to select promising novel enzymes from a pool of candidates. Most supervised VEP models are framed as regression tasks, placing more emphasis on getting the predicted quantities correct than on the relative comparison of individual candidates. Preferential or contrastive models may better align with the goal of selection, or acquisition, especially when informed by predictive uncertainty. Here, we introduce a probabilistic preferential learning model based on the Kermut Gaussian process (PKermut) that we designed with the ambition to increase the hit rate among selected variants. We benchmark PKermut against established models, including the original Kermut, the RITA regressor, and an augmented Potts model, on 69 curated ProteinGym datasets across various assay categories. To evaluate acquisition performance, we propose a novel quantile cross-validation scheme that ensures the evaluation of a models ability to extrapolate by reserving high-performing variants exclusively for the test set. We assess models using Spearman correlation and evaluate their acquisition performance using five different acquisition functions, encompassing both uncertainty-aware and unaware strategies. Our experimental results indicate that uncertainty estimates improve the acquisition ability of our models, and that strategies that reward uncertainty generally result in better outcomes than those that do not on single-mutation variant datasets. We observe that PKermuts Spearman scores and ability to acquire improved variants are greatly affected by the number of variant comparisons sampled in the training set. Kermut achieves the highest Spearman correlation in 54/69 datasets (78%), compared to 12/69 (17%) for PKermut. For acquisition performance, Kermut leads in 44/69 datasets (64%), while PKermut leads in 15/69 (22%). While at this stage PKermut is not a recommended alternative to Kermut, its contrastive nature offers several conceptual opportunities. We share our findings to inspire further development aimed at improving the alignment between training objectives of VEP models and their downstream application in protein engineering.

bioinformatics↗

bioRxiv2025

Enhancing protein graph representations through static-dynamic fusion

Machine learning for protein modeling faces significant challenges due to proteins inherently dynamic nature, yet most graph-based machine learning methods rely solely on static structural information. Recently, the growing availability of molecular dynamics trajectories provides new opportunities for understanding the dynamic behavior of proteins; however, computational methods for utilizing this dynamic information remain limited. We propose a novel graph representation that integrates both static structural information and dynamic correlations from molecular dynamics trajectories, enabling more comprehensive modeling of proteins. By applying relational graph neural networks (RGNNs) to process this heterogeneous representation, we demonstrate significant improvements over structure-based approaches across three distinct tasks: atomic adaptability prediction, binding site detection, and binding affinity prediction. Our results validate that combining static and dynamic information provides complementary signals for understanding proteinligand interactions, offering new possibilities for drug design and structural biology applications.

bioinformatics↗

bioRxiv2024

Traction force generation in motile malaria parasites is modulated by the Plasmodium adhesin TLP

Plasmodium sporozoites are the highly polarized and motile forms of the malaria parasite that are transmitted by mosquitoes to the vertebrate hosts. Sporozoites use myosin molecular motors to generate retrograde flow of actin filaments. These are linked to plasma-membrane spanning adhesins, which in turn bind to the extracellular environment, resulting in forward directed gliding motility. The gliding motility machine of sporozoites leads to high speeds in the range of micrometer per second, which are essential for efficient migration in the skin. Yet, it is not clear how the individual parts of the machinery work together to generate force during migration. Sporozoites are elongated and curved cells and move on circular tracks in vitro. Sporozoites lacking the adhesin thrombospondin related anonymous protein (TRAP) like protein, TLP, can still migrate in the skin, but at a lower level. TLP lacking sporozoites generate a lower force on the dorsal (non-substrate facing) surface as measured by laser tweezers. Here we use traction force microscopy to investigate motile sporozoites and the forces they produce during migration on their ventral surface. Both wild type and tlp(-) sporozoites show distinct foci of force generation, but tlp(-) sporozoites generating overall lower forces. Our findings demonstrate that TLP is an important element of the force-generating machinery during sporozoite gliding motility.

microbiology↗

bioRxiv2023

Explainable prediction of catalysing enzymes from reactions using multilayer perceptrons

Assigning or proposing a catalysing enzyme given a chemical or biochemical reaction is of great interest to life sciences and chemistry alike. The exploration and design of metabolic pathways or the challenge of finding more sustainable enzyme-catalysed alternatives to traditional organic reactions are just two examples of tasks that require an association between reaction and enzyme. However, given the lack of large and balanced annotated data sets of enzyme-catalysed reactions, assigning an enzyme to a reaction still relies on expert-curated rules and databases. Here, we present a data-driven explainable human-in-the-loop machine learning approach to support and ultimately automate the association of a catalysing enzyme with a given biochemical reaction. In addition, the proposed method is capable of predicting enzymes as candidate catalysts for arbitrary organic reactions. Finally, the introduced explainability and visualisation methods can easily be generalised to support other machine-learning approaches involving chemical and biochemical reactions.

bioinformatics↗

bioRxiv2022

Cell size and actin architecture determine force generation in optogenetically activated adherent cells

Adherent cells use actomyosin contractility to generate mechanical force and to sense the physical properties of their environment, with dramatic consequences for migration, division, differentiation and fate. However, the organization of the actomyosin system within cells is highly variable, with its assembly and function being controlled by small GTPases from the Rho-family. How activation of these regulators translates into cell-scale force generation and the corresponding sensing capabilities in the context of different physical environments is not understood. Here we probe this relationship combining recent advances in non-neuronal optogenetics with micropatterning and traction force microscopy on soft elastic substrates. We find that after whole-cell RhoA-activation by the CRY2/CIBN optogenetic system with a short pulse of 100 milliseconds, single cells contract before returning to their original tension setpoint with near perfect precision on a time scale of several minutes. To decouple the biochemical and mechanical elements of this response, we introduce a mathematical model that is parametrized by fits to the dynamics of the substrate deformation energy. We find that the RhoA-response builds up quickly on a time scale of 20 seconds, but decays slowly on a time scale of 50 seconds. The larger the cells and the more polarized their actin cytoskeleton, the more substrate deformation energy is generated. RhoA-activation starts to saturate if optogenetic pulse length exceeds 50 milliseconds, revealing the intrinsic limits of biochemical activation. Together our results suggest that adherent cells establish tensional homeostasis by the RhoA-system, but that the setpoint and the dynamics around it are strongly determined by cell size and the architecture of the actin cytoskeleton, which both are controlled by the extracellular environment.

biophysics↗

bioRxiv2017

Harnessing optogenetics to probe sub-cellular mechanics

The mechanics of the actin cytoskeleton regulates cell morphogenesis during essential physiological processes. While cellular contractility is known to be largely RhoA-dependent, the process by which localized biochemical signals are translated into cell-level responses is not well understood. Here we combine optogenetic control of RhoA in space and time, live cell imaging and traction force microscopy to investigate the dynamics of actomyosin-based force generation in adherent fibroblasts. Local activation of RhoA not only stimulates local recruitment of actin and myosin, but also leads to increased traction forces outside the activation region that rapidly propagate across the cell via stress fibers and drive actin flow towards the region of heightened RhoA. Surprisingly, the flow reverses direction when local RhoA activation stops. We identify zyxin as a regulator of stress fiber mechanics, as stress fibers are fluid-like without flow reversal in its absence. These experimental data are used to constrain a physical model, which demonstrates that stress fibers are elastic-like, even at time scales exceeding turnover of constituent proteins. Such molecular control of actin mechanics likely plays critical roles in regulation of morphogenic events.

biophysics↗