@phdthesis{22744, author = {Javoor, Manjunath}, isbn = {978-3-99078-090-9 }, issn = {2663-337X}, keywords = {Actin cytoskeleton, Cell migration, cryo-electron tomography}, pages = {121}, publisher = {Institute of Science and Technology Austria }, title = {{Large-scale imaging of cellular actin networks at single filament resolution using montage cryo-electron tomography}}, doi = {10.15479/AT-ISTA-22744}, year = {2026}, } @article{22750, abstract = {Robust oxygenic photosynthesis requires the efficient assembly and repair of the multi-subunit oxygen-evolving photosystem II (PSII) complex. Previous cryogenic electron microscopy (cryo-EM) structures of PSII assembly/disassembly intermediates have relied on the analysis of deletion mutants or removal of PSII subunits in vitro. Here we report the cryo-EM structures of naturally occurring dimeric PSII intermediates from the cyanobacterium Thermosynechococcus vestitus at a resolution of about 2.2 Å. These intermediates contain inactive dimers lacking the oxygen-evolving complex (OEC) and semi-active dimers with the OEC present in one of the two monomers. Our structural data provide a mechanism for how assembly and disassembly of the Mn4CaO5 cluster is coordinated with the binding and release of the extrinsic proteins: restructuring of the C-terminal tail of D1 subunit during assembly or disassembly of the Mn cluster triggers conformational changes in D2, CP47 and CP43 to drive the binding/release of the extrinsic proteins. A combination of structural and mass spectrometry data also suggests that the inactive PSII complexes may include damaged complexes containing oxidized D1-His332, a monodentate ligand to one of the Mn ions of the OEC.}, author = {Zhao, Ziyu and Vercellino, Irene and Whitelegge, Julian P. and Maghlaoui, Karim and Białek, Wojciech and Nixon, Peter J. and Sazanov, Leonid A}, issn = {2041-1723}, journal = {Nature Communications}, publisher = {Springer Nature}, title = {{Cryo-EM structures of naturally occurring dimeric photosystem II complexes lacking the Mn4CaO5 cluster}}, doi = {10.1038/s41467-026-75324-x}, volume = {17}, year = {2026}, } @misc{22852, abstract = {This Research Data contains supplemental videos for Chapter 4 "Designing bistable nanostructures for target behavior" of my PhD Thesis "Biological functionality without biochemistry: designing nanomachines for target behavior". Supplemental video 1: Video showing the transition pathway of a bistable nanostructure with sphere-based arms, corresponding to the Machine in Scenario 4. Supplemental video 2: Video showing the transition pathway of the Source in Scenario 1. The arm tips change sides during the transition, demonstrating that the arms pass through each other. Supplemental video 3: Video showing the transition pathway of a fully polyhedral hinge structure with unconstrained arms. Note that we only show the ends of the arms. Supplemental video 4: Video showing the transition pathway of the coupled energy-delivery reaction of a Machine (gray) and a Source (blue) nanostructure for the optimized parameters in Scenario 3. Note that we only show the ends of the arms.}, author = {Ehrmann, Andreas}, keywords = {functional nanostructures, bistability, target behavior, transition pathway}, publisher = {Institute of Science and Technology Austria}, title = {{Supplemental videos for Designing bistable nanostructures for target behavior}}, doi = {10.15479/AT-ISTA-22852}, year = {2026}, } @phdthesis{22873, author = {Ehrmann, Andreas}, isbn = {978-3-99078-092-3}, issn = {2663-337X}, keywords = {PhD Thesis, functional nanomachines, biological functionality, nanotechnology, energy delivery, target behavior, dynamics, design principles, optimization, differentiable statistical physics, machine learning}, pages = {168}, publisher = {Institute of Science and Technology Austria}, title = {{Biological functionality without biochemistry: Designing nanomachines for target behavior}}, doi = {10.15479/AT-ISTA-22873}, year = {2026}, } @phdthesis{21198, abstract = {In recent years there has been a massive increase in the amount of data generated in a decentralized manner. Ever more powerful edge devices, such as smartphones, have become ubiquitous in most societies on earth. Through text typed, photos taken and apps used, these devices, which we refer to as clients, generate enormous amounts of high quality and complex data. Moreover, the nature of these devices means the data they generate is often sensitive and privacy concerns prevent it being gathered and stored in a central location. This presents a challenge to the modern machine learning paradigm that requires central access to large amounts of data. Federated learning (FL) has emerged as one of the answers to this problem. Rather than bringing the data to the model, FL sends the model to the data. Model training takes place on device, with periodically synchronized updates, allowing data to remain locally stored. While this approach offers significant privacy advantages it comes with its own set of unique challenges. These include: data heterogeneity, the notion that different devices generate data in distinct ways which can negatively impact training dynamics; systems heterogeneity, meaning that different devices may have differing hardware specifications; high communication costs, which are induced by the repeated transferring of models over the network and low device computational power, which limits the use of larger models on device. In this thesis we present a range of methods for federated learning. We focus primarily on the challenge of data heterogeneity, though the methods presented are designed to be well adapted to the other challenges of a federated setting, such as the constraints of limited compute and communication overhead. We first present a method for explicitly modeling client data heterogeneity. The approach formulates clients as samples from a certain probability distribution and infers the parameters of this distribution from the available training clients. This learned distribution then represents the heterogeneity present among the clients and can be sampled from in order to create new simulated clients that are similar to the real clients we have observed so far. Following this we present two methods for directly dealing with data heterogeneity through personalization. Highly heterogeneous client data distributions can mean that learning a single global model becomes suboptimal, and some form of personalization of models to each individual client is required. Our approaches are based around hypernetworks, which we use to generate personalized model parameters without the need for additional training or finetuning. In the first approach we focus on generating full parameterizations of client models using learned embeddings of client data and labels, with a hypernetwork located on the central server. In the second approach we address the more challenging scenario where we want to generate a personalized model for a client without any label information. The hypernetwork is trained to generate a low dimensional representation of a client’s personalized model parameters, allowing it to be transferred to and run on the client devices. In our final presented method, we change our focus and rather than aim to directly address the challenge of data heterogeneity, we instead ensure we are unaffected by it. This is done in the context of k-means clustering and we present a method for federated clustering with a focus on added privacy guarantees.}, author = {Scott, Jonathan A}, issn = {2663-337X}, pages = {158}, publisher = {Institute of Science and Technology Austria}, title = {{Data heterogeneity and personalization in federated learning}}, doi = {10.15479/AT-ISTA-21198}, year = {2026}, } @phdthesis{21854, abstract = {As neural-network-based models grow both in size and popularity, interest has grown in making the models smaller and more efficient to train. To that end, many methods have been proposed to prune models by reducing their number of nonzero parameters. Additionally, parameter-efficient fine-tuning, in which a much smaller number of parameters than the total contained in the model is updated during training, has become very popular, especially in the space of Large Language Models. At the same time, the increasingly routine deployment of machine learning in real-world applications has spurred a drive to make them more trustworthy - in the sense of, among other things, being unbiased, interpretable, and editable. In this thesis, we examine the interplay between efficiency and trustworthiness. First, we analyze the effects of model pruning on bias in computer vision models, demonstrating that increased sparsity leads to greater bias, largely as a function of increased model uncertainty in marginal cases. Based on this observation, we propose several bias mitigation techniques. Then, we demonstrate that example-specific model pruning can improve model interpretation methods while improving pruning efficiency to make example-specific model pruning feasible in real time. Then, we investigate the effectiveness of parameter-efficient and data-efficient model personalization via fine-tuning, demonstrating that it is highly feasible with very small computational and data resources. Finally, we consider efficiency in editing model knowledge using a custom synthetic data framework, demonstrating that parameter-efficient, low-rank fine-tuning frequently outperforms full-rank fine-tuning, and, additionally, that restricting which model blocks are fine-tuned frequently improves results. Together, the results in this thesis provide new insights and techniques for combining trustworthiness and efficiency during neural network inference and training. }, author = {Iofinova, Eugenia B}, issn = {2663-337X}, pages = {237}, publisher = {Institute of Science and Technology Austria}, title = {{On the utility and effects of efficiency in artificial neural networks}}, doi = {10.15479/AT-ISTA-21854}, year = {2026}, } @article{21982, abstract = {A floating Leidenfrost droplet exhibits curvature inversion of its underside, due to the balance of vapor pressure and surface tension. Using interferometric imaging, we find different behavior for a levitated hydrogel sphere. Curvature inversion is observed briefly just after deposition, but quickly gives way to a steady state with no inversion. We show the essential role of vaporization in shaping the underbelly of the hydrogel, where changes due to direct mass loss are more significant than the balance of vapor pressure and elastic forces.}, author = {Diaz Melian, Vicente L and Lenton, Isaac C and Binysh, Jack and Souslov, Anton and Waitukaitis, Scott R}, issn = {2470-0053}, journal = {Physical Review E}, number = {5}, publisher = {American Physical Society}, title = {{Geometry of the vapor layer under a Leidenfrost hydrogel sphere}}, doi = {10.1103/m7gr-2t6j}, volume = {113}, year = {2026}, } @article{22972, abstract = {Genomic measurements of DNA methylation, gene expression or protein levels are becoming more prevalent and are increasingly used to study health outcomes. However, most proposed association testing methods consider only marginal effects of each feature on a single outcome variable and are not set up to handle highly correlated, continuous data. Here, we introduce MAJA, a method to learn shared and outcome-specific effects for multiple traits in multi-omics data. MAJA determines the unique contribution of individual loci, genes, or molecular pathways to variation in one or more traits, conditional on all other measured “omics” data genome-wide. Simulations show MAJA accurately finds shared and distinct associations between omics-data and multiple traits and estimates omics-specific (co)variances, allowing for sparsity and correlations within the data. Applying MAJA to 12 outcome traits in Generation Scotland methylation data (n = 18 264), we find novel shared epigenetic probes among cholesterol metabolism, osteoarthritis, blood pressure and asthma. In contrast to marginal testing, we find only 10 CpG probes with significant effects above the genome-wide background. This highlights the need for joint association testing in highly correlated methylation data from whole blood and for studies of increased sample size in order to refine epigenomic associations in observational data.}, author = {Krätschmer, Ilse and Smith, Hannah M and McCartney, Daniel L and Bernabeu, Elena and Mahmoudi, Mahdi and Campbell, Archie and Corley, Janie and Harris, Sarah E and Cox, Simon R and Marioni, Riccardo E and Robinson, Matthew Richard}, issn = {2635-0041}, journal = {Bioinformatics Advances}, number = {1}, publisher = {Oxford University Press}, title = {{MAJA: Multivariate Bayesian model for discovery of shared epigenetic pathways across human phenotypes}}, doi = {10.1093/bioadv/vbag231}, volume = {6}, year = {2026}, } @article{23089, abstract = {The location of the hydrogen-helium (H/He) immiscibility boundary controls whether and where helium rain occurs in giant planets, yet it remains uncertain because high-pressure experiments are challenging and ab initio simulations are limited in system size and simulation time. We map this boundary by computing composition-dependent chemical potentials from large-scale molecular dynamics driven by machine learning potentials trained on three density functional approximations (PBE, vdW-DF, and the hybrid HSE). Across the pressure range of 100-1000 GPa, the demixing temperatures are typically ~ 2000 K lower than previous ab initio simulations using small system sizes, and this discrepancy is larger than the estimated error from the chosen density functionals, machine learning potentials, neglect of nuclear quantum effects, and statistical uncertainties. Fitting the H/He mixing free energy to a Redlich-Kister regular solution model rationalizes the thermodynamic driving force for phase separation, and provides a predictive representation of the boundary. Comparing with current planetary interior profiles indicates that helium rain is plausible in Saturn but unlikely in the warmer interior of Jupiter. Our results narrow the uncertainty in the H/He immiscibility boundary and provide inputs for planetary models that couple demixing, heat transport, and composition gradients in gas giants.}, author = {Wang, Xiaoyu and Hamel, Sebastien and Cheng, Bingqing}, issn = {2041-1723}, journal = {Nature Communications}, publisher = {Springer Nature}, title = {{Hydrogen-helium immiscibility boundary in gas-giant planetary interiors from machine-learning molecular dynamics}}, doi = {10.1038/s41467-026-77117-8}, volume = {17}, year = {2026}, } @phdthesis{23072, abstract = {Inferring signal from large quantities of data is one of the most important problems in today’s world, with neural networks emerging as the most powerful solution. While there have been massive breakthroughs in the ML space, there are two main drawbacks. The first is that most of these models are “black boxes”, with very little understanding of why they work, which makes it hard to find the correct architecture for a given problem. The architecture is crucial, as the widely used training algorithms are “local” and thus rely on the loss landscape being benign enough to find a global optimum. This brings us to the second problem: while there is a lot of evidence that this can be overcome with large enough data, this is less clear for datasets which are noisy, do not have enough samples, or have a non-standard structure. Historically, on a smaller scale, the alternative has been simple (Bayesian) statistical methods, which also have the advantage of being interpretable and adaptable. However, many of the classical results cannot be easily extended to the high-dimensional regime in which modern data lives. In this work we want to consider two different settings relating to the above. The first is approximate message passing (AMP), an iterative algorithm for high-dimensional (generalized) linear models that easily lends itself to Bayesian setups, and which in many cases is provably Bayes-optimal. The second is an autoencoder with a shallow encoder, for which we can analyze the weights along the training trajectory of gradient descent. Surprisingly, we will find a connection between the two, as we recover certain structures of AMP. We will first introduce the AMP framework by studying an extension that can be applied to a wider variety of data: rotationally invariant generalized approximate message passing (RI-GAMP). This shows that one can extend the AMP framework to a much wider class of algorithms, relevant to more practical settings. The results are then also confirmed on real data that go beyond the theoretical setting. We then move to the setup of an autoencoder with a shallow encoder and a Bayesian prior in the high-dimensional regime, in which the input and compression dimensions are large and proportional; this is also a first step towards opening one of the black boxes above, as we can say precisely when and why the network works. While autoencoders are widely used in modern ML architectures, most fundamental research in this direction has focused on extreme asymptotic regimes, where the autoencoder is either severely over- or under-parameterized, and which do not extend to the proportional high-dimensional regime. We show that for Gaussian priors the network finds the global optimum, and these theoretical findings extrapolate strikingly well to real image data. Next we extend this analysis to more structured priors and a deeper decoder. For sparse Gaussian inputs in the training regime studied here, the shallow decoder does not exploit sparsity and yields the same dynamics as in the Gaussian setting. We then show how a deeper decoder can break this behavior, and interestingly, if we treat the problem as an AMP problem, the network eventually learns the same decoding as AMP. This too extrapolates to real image data. Lastly, we will show that the AMP setup has real practical use, by using a practical variant, genetic vector approximate message passing (gVAMP), for predicting plasma protein levels from genetic information using the UKB GWAS data. This is a perfect example of the problems in the high-dimensional regime: approximately 8.4 million dimensions and 40,000 samples, where the high noise, low sample count, and large size make it hard for ML methods to work. To get the optimal prediction accuracy we consider a two-step inference. We first consider a pre-processed version of the data, of about 880,000 dimensions, on which we run the gVAMP algorithm; this allows us to find predictive regions, taking into account joint effects between different regions, and for most proteins we end up selecting only about 0.1% of the dimensions. We then consider the analogue of this small subset, on which we can now run a standard tree algorithm (XGB) that no longer suffers from the aforementioned problems. This pipeline yields better prediction accuracy than recent nonlinear deep learning approaches applied to this data, as we are able to handle the pre-processing in a principled Bayesian way.}, author = {Kögler, Kevin}, issn = {2663-337X}, pages = {222}, publisher = {Institute of Science and Technology Austria}, title = {{Compression and inference in high dimensions: An approximate message passing perspective}}, doi = {10.15479/AT-ISTA-23072}, year = {2026}, } @phdthesis{22857, abstract = {Artificial intelligence and machine learning have undergone an unprecedented evolution in the past decade, motivating a research effort toward a theory able to capture the qualitative behavior of large-scale neural systems. A central puzzle has been the clear benefit of scaling architecture size and overfitting the training set in supervised learning tasks. This evidence, in apparent contradiction with classical statistical learning theory, pushed researchers to develop a new theory capturing the interplay between the algorithmic and architectural bias of training and the specific target function, differently from previous methods rooted in uniform stability. This approach has enabled a grounded understanding of novel learning regimes, typically through formal limits where the number of training samples $n$, data dimensions $d$, and model parameters $p$ grow to infinity at different rates. \\ In this thesis, we follow this approach, focusing on the trustworthiness of high-dimensional models: properties that are difficult to control during training or deployment and often emerge under unpredictable or adversarial conditions. In such settings, it is crucial to formally ensure a priori the reliability of machine learning systems. First, we study data memorization, both as label fitting and as the storage of private information about training samples in trained parameters. We prove that $p = \Omega(n)$ parameters are sufficient for a deep neural network to memorize a generic set of labels, and for a model to memorize spurious features across training data. We then give evidence that $p = \Omega(dn)$ parameters are instead necessary for an adversary to reconstruct the full training set from the trained parameters. Second, we study robustness, both to adversarial perturbations and to distribution shift. We first prove that $p = \Omega(dn)$ parameters can be sufficient for a class of neural networks to overfit the training data while guaranteeing robustness to adversarial perturbations. Then, we focus on spurious correlations learning in high-dimensional regression, studying the effect of the ridge regularization parameter in the proportional regime $n = \Theta(d)$, and connecting it via an equivalence argument to the role of over-parameterization $p = \Omega(n)$ in neural networks. We also investigate the architectural bias of attention-based networks, showing that they are sensitive to the replacement of individual words in an embedded sentence, allowing them to generalize on sentences where the contextual meaning depends on one or few words. Finally, we study differentially private optimization in high-dimensional regimes. We prove that standard private gradient methods do not suffer in the over-parameterized regime $p = \Omega(n)$, challenging the current wisdom based on stability-derived generalization bounds. We then consider linear regression in the proportional regime $n = \Theta(d)$, showing that standard private gradient descent can achieve optimal rates under appropriate hyper-parameter scaling, such as sufficiently small gradient clipping constants, whose role is still debated in practice.}, author = {Bombari, Simone}, isbn = {978-3-99078-091-6}, issn = {2663-337X}, keywords = {machine learning, high-dimensional statistics, deep learning theory, privacy, memorization, robustness}, pages = {446}, publisher = {Institute of Science and Technology Austria}, title = {{Trustworthy machine learning in high dimensions}}, doi = {10.15479/AT-ISTA-22857}, year = {2026}, } @article{23043, abstract = {Mendelian randomization, the standard instrument variable causal-inference approach for genomics, can be severely biased by inadequate selection of appropriate instrument variables and horizontal pleiotropy. We introduce Causal Inference GWAS, a graphical approach that selects genetic instruments from summary statistics controlling for linkage and pleiotropy, accommodates rare variants and binary outcomes, distinguishes direct from indirect risk factors in high dimensions, and flags latent confounding. Applied to 9 risk factors, 4 metabolic disease outcomes, and 8.4M variants across 458,747 UK Biobank individuals, our method runs in 20 minutes, identifying only 696 genome-wide valid instruments. We replicate nearly all previously reported risk factor-outcome paths, but find that most are indistinguishable from unmeasured confounding. This suggests robust biobank causal inference will require longitudinal and family data to resolve temporal precedence from reverse causation. Our approach offers a principled first step toward screening genuine causal signal from phenotypic correlation in biobank data.}, author = {Machnik, Nick N and Mahmoudi, Seyed Mahdi and Borczyk, Malgorzata and Krätschmer, Ilse and Bauer, Markus J. and Robinson, Matthew Richard}, issn = {2041-1723}, journal = {Nature Communications}, publisher = {Springer Nature}, title = {{Causal inference for multiple risk factors and diseases from genomics data}}, doi = {10.1038/s41467-026-76877-7}, volume = {17}, year = {2026}, } @phdthesis{21423, abstract = {Biological systems, such as cytoskeletal filaments and multicellular collectives, exhibit rich self-organization driven by their intrinsic activity and physical constraints. Across scales, geometry plays a central role: the shape of active units and the structure of their environment strongly influence how order emerges in these far-from-equilibrium systems. In this thesis, we combine minimal active-matter models with bottom-up in vitro experiments to identify the principles by which geometry and activity shape collective behavior in biological matter. At the subcellular scale, we investigate how the physical properties of chiral FtsZ filaments govern their large-scale organization. Using a mesoscopic model validated by reconstituted filament assays, we show that filament shape, intrinsic curvature, and flexibility, together with activity-driven straightening, determine whether filaments form chiral rings, polar bands, or nematic-like states. Our model shows that small changes in filament mechanics produce large shifts in collective structures, revealing how the geometry of individual filaments amplifies through activity to control patterns relevant for bacterial cytokinesis. At the cellular scale, we examine how geometry of the external environment influences collective cancer cell migration. Experiments in controlled microfabricated obstacle environments demonstrate that geometric disorder alone can induce a transition from collective to single-cell invasion. A minimal model of active particles in disordered geometries reproduces these effects, demonstrating that cell dissemination results both from local geometrical constraints and a global integration of spatial disorder over time. Surprisingly, this transition is accompanied by a change in the universality class of interface dynamics, revealing that cell-environment interactions can shift the collective behavior of an invasive front purely through mechanics. In the last system, we study how cell-intrinsic geometry and chirality affect collective motion in confined and unconfined settings. Using a minimal model inspired by in vitro experiments, we analyze motile cells that exhibit persistent rotation due to intrinsic left-right asymmetry. In circular confinement, this simplified chiral model reproduces several experimentally observed features within specific parameter regimes. At high densities in unconfined environments, however, intrinsic chirality becomes masked even before the onset of the jamming transition. These results show that single-cell geometric asymmetry does not necessarily translate into large-scale chiral flows, offering new insight into how forces and symmetries shape collective motion in dense cellular systems. Overall, this work uncovers how geometric constraints – embedded in the intrinsic architecture of an active filament or encoded in the complexity of a microenvironment couple to active forces to generate different classes of complex behaviors. }, author = {Dunajova, Zuzana}, isbn = {978-3-99078-076-3}, issn = {2663-337X}, pages = {110}, publisher = {Institute of Science and Technology Austria}, title = {{Geometry-driven self-organization of migrating cells and chiral filaments}}, doi = {10.15479/AT-ISTA-21423}, year = {2026}, } @misc{21439, abstract = {These files contain supplementary movies accompanying the PhD thesis “Geometry-driven self-organization of migrating cells and chiral filaments” by Zuzana Dunajova (2026). The videos provide additional visual material supporting the experiments and results described in the thesis.}, author = {Dunajova, Zuzana}, publisher = {Institute of Science and Technology Austria}, title = {{Supplementary movies to PhD thesis “Geometry-driven self-organization of migrating cells and chiral filaments”}}, doi = {10.15479/AT-ISTA-21439}, year = {2026}, } @article{20847, abstract = {We report on an experimental active matter system with motion restricted to four cardinal directions. Our particles are magnetite-doped colloidal spheres driven by the Quincke electrorotational instability. The absence of a magnetic field (|𝑩|=0) leads to circular trajectories interspersed with short spontaneous runs. Intermediate fields (|𝑩|≲20mT) linearize the motion along the axis perpendicular to 𝑩. At high magnetic fields, we observe the surprising emergence of a second, distinct linearization along the axis parallel to 𝑩. With numerical simulations, we show that this behavior can be explained by anisotropic magnetic susceptibility.}, author = {Fitzgerald, Eavan and Clavaud, Cécile and Das, Debasish and Lenton, Isaac C and Waitukaitis, Scott R}, issn = {2470-0053}, journal = {Physical Review E}, number = {6}, publisher = {American Physical Society}, title = {{Rolling at right angles: Magnetic anisotropy enables dual-anisotropic active matter}}, doi = {10.1103/1ss8-31rb}, volume = {112}, year = {2025}, } @inproceedings{21327, abstract = {Proteins exist as a dynamic ensemble of multiple conformations, and these motions are often crucial for their functions. However, current structure prediction methods predominantly yield a single conformation, overlooking the conformational heterogeneity revealed by diverse experimental modalities. Here, we present a framework for building experiment-grounded protein structure generative models that infer conformational ensembles consistent with measured experimental data. The key idea is to treat stateof-the-art protein structure predictors (e.g., AlphaFold3) as sequence-conditioned structural priors, and cast ensemble modeling as posterior inference of protein structures given experimental measurements. Through extensive real-data experiments, we demonstrate the generality of our method to incorporate a variety of experimental measurements. In particular, our framework uncovers previously unmodeled conformational heterogeneity from crystallographic densities, and generates high-accuracy NMR ensembles orders of magnitude faster than the status quo. Notably, we demonstrate that our ensembles outperform AlphaFold3 (Abramson et al., 2024) and sometimes better fit experimental data than publicly deposited structures to the Protein Data Bank (PDB, Burley et al. (2017)). We believe that this approach will unlock building predictive models that fully embrace experimentally observed conformational diversity.}, author = {Maddipatla, Sai A and Sellam, Nadav E and Bojan, Meital I and Vedula, Sanketh and Schanda, Paul and Marx, Ailie and Bronstein, Alexander}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, issn = {2640-3498}, location = {Vancouver, Canada}, pages = {42366 -- 42393}, publisher = {ML Research Press}, title = {{Inverse problems with experiment-guided AlphaFold}}, volume = {267}, year = {2025}, } @misc{18697, abstract = {The information-processing capability of the brain’s cellular network depends on the physical wiring pattern between neurons and their molecular and functional characteristics. Mapping neurons and resolving their individual synaptic connections can be achieved by volumetric imaging at nanoscale resolution with dense cellular labelling. Light microscopy is uniquely positioned to visualize specific molecules but dense, synapse-level circuit reconstruction by light microscopy has been out of reach due to limitations in resolution, contrast, and volumetric imaging capability. Here we developed light-microscopy based connectomics (LICONN). We integrated specifically engineered hydrogel embedding and expansion with comprehensive deep-learning based segmentation and analysis of connectivity, thus directly incorporating molecular information in synapse-level brain tissue reconstructions. LICONN will allow synapse-level brain tissue phenotyping in biological experiments in a readily adoptable manner.}, author = {Danzl, Johann G and Lyudchik, Julia and Kreuzinger, Caroline}, publisher = {Institute of Science and Technology Austria}, title = {{Light-microscopy based connectomic reconstruction of mammalian brain tissue}}, doi = {10.15479/AT:ISTA:18697}, year = {2025}, } @misc{18712, abstract = {This file contains the code associated with the manuscript 'Effect of assortative mating and sexual selection on polygenic barriers to gene flow'. }, author = {Surendranadh, Parvathy and Sachdeva, Himani}, publisher = {Institute of Science and Technology Austria}, title = {{Mathematica notebook and Fortran code for 'Effect of assortative mating and sexual selection on polygenic barriers to gene flow'}}, doi = {10.15479/AT:ISTA:17344}, year = {2025}, } @article{18778, abstract = {Transcription by RNA polymerase II (Pol II) can be repressed by noncoding RNA, including the human RNA Alu. However, the mechanism by which endogenous RNAs repress transcription remains unclear. Here we present cryogenic-electron microscopy structures of Pol II bound to Alu RNA, which reveal that Alu RNA mimics how DNA and RNA bind to Pol II during transcription elongation. Further, we show how distinct domains of the general transcription factor TFIIF control repressive activity. Together, we reveal how a noncoding RNA can regulate mammalian gene expression.}, author = {Tluckova, Katarina and Kaczmarek, Beata M and Testa Salmazo, Anita P and Bernecky, Carrie A}, issn = {1545-9985}, journal = {Nature Structural & Molecular Biology}, pages = {607--612}, publisher = {Springer Nature}, title = {{Mechanism of mammalian transcriptional repression by noncoding RNA}}, doi = {10.1038/s41594-024-01448-7}, volume = {32}, year = {2025}, } @article{18879, abstract = {Our brain has remarkable computational power, generating sophisticated behaviors, storing memories over an individual’s lifetime, and producing higher cognitive functions. However, little of our neuroscience knowledge covers the human brain. Is this organ truly unique, or is it a scaled version of the extensively studied rodent brain? Combining multicellular patch-clamp recording with expansion-based superresolution microscopy and full-scale modeling, we determined the cellular and microcircuit properties of the human hippocampal CA3 region, a fundamental circuit for memory storage. In contrast to neocortical networks, human hippocampal CA3 displayed sparse connectivity, providing a circuit architecture that maximizes associational power. Human synapses showed unique reliability, high precision, and long integration times, exhibiting both species- and circuit-specific properties. Together with expanded neuronal numbers, these circuit characteristics greatly enhanced the memory storage capacity of CA3. Our results reveal distinct microcircuit properties of the human hippocampus and begin to unravel the inner workings of our most complex organ. }, author = {Watson, Jake and Vargas Barroso, Victor M and Morse, Rebecca and Navas Olivé, Andrea C and Tavakoli, Mojtaba and Danzl, Johann G and Tomschik, Matthias and Rössler, Karl and Jonas, Peter M}, issn = {1097-4172}, journal = {Cell}, number = {2}, pages = {501--514.e18}, publisher = {Elsevier}, title = {{Human hippocampal CA3 uses specific functional connectivity rules for efficient associative memory}}, doi = {10.1016/j.cell.2024.11.022}, volume = {188}, year = {2025}, }