High-throughput metabolomics data provide a detailed molecular window into biological processes. We consider the problem of assessing how the association of metabolite levels with individual (sample) characteristics such as sex or treatment may depend on metabolite characteristics such as pathway. Typically this is one in a two-step process: In the first step we assess the association of each metabolite with individual characteristics. In the second step an enrichment analysis is performed by metabolite characteristics among significant associations. We combine the two steps using a bilinear model based on the matrix linear model (MLM) framework we have previously developed for high-throughput genetic screens. Our framework can estimate relationships in metabolites sharing known characteristics, whether categorical (such as type of lipid or pathway) or numerical (such as number of double bonds in triglycerides). We demonstrate how MLM offers flexibility and interpretability by applying our method to three metabolomic studies. We show that our approach can separate the contribution of the overlapping triglycerides characteristics, such as the number of double bonds and the number of carbon atoms. The proposed method have been implemented in the open-source Julia package, MatrixLM. Data analysis scripts with example data analyses are also available.

Although we must prioritize the processing of task-relevant information to navigate life, our ability to do so fluctuates across time. Previous work has identified fMRI functional connectivity (FC) networks that predict an individual's ability to sustain attention and vary with attentional state from one minute to the next. However, traditional dynamic FC approaches typically lack the temporal precision to capture moment-by-moment network fluctuations. Recently, researchers have 'unfurled' traditional FC matrices in 'edge cofluctuation time series' which measure time point-by-time point cofluctuations between regions. Here we apply event-based and parametric fMRI analyses to edge time series to capture high-frequency fluctuations in networks related to attention. In two independent fMRI datasets in which participants performed a sustained attention task, we identified a reliable set of edges that rapidly deflects in response to rare task events. Another set of edges varies with continuous fluctuations in attention and overlaps with a previously defined set of edges associated with individual differences in sustained attention. Demonstrating that edge-based analyses are not simply redundant with traditional regions-of-interest based approaches, up to one-third of reliably deflected edges were not predicted from univariate activity patterns alone. These results reveal the large potential in combining traditional fMRI analyses with edge time series to identify rapid reconfigurations in networks across the brain.

The rapid proliferation of trajectory inference methods for single-cell RNA-seq data has allowed researchers to investigate complex biological processes by examining underlying gene expression dynamics. After estimating a latent cell ordering, statistical models are used to determine which genes exhibit changes in expression that are significantly associated with progression through the biological trajectory. While a few techniques for performing trajectory differential expression exist, most rely on the flexibility of generalized additive models in order to account for the inherent nonlinearity of changes in gene expression. As such, the results can be difficult to interpret, and biological conclusions often rest on subjective visual inspections of the most dynamic genes. To address this challenge, we propose scLANE testing, which is built around an interpretable generalized linear model and handles nonlinearity with basis splines chosen empirically for each gene. In addition, extensions to estimating equations and mixed models allow for reliable trajectory testing under complex experimental designs. After validating the accuracy of scLANE under several different simulation scenarios, we apply it to a set of diverse biological datasets and display its ability to provide novel biological information when used downstream of both pseudotime and RNA velocity estimation methods.

Emergence of SARS-CoV-2 variants diminishes the efficacy of vaccines and antiviral monoclonal antibodies. Continued development of immunotherapies and vaccine immunogens resilient to viral evolution is therefore necessary. Using coldspot-guided antibody discovery, a screening approach that focuses on portions of the virus spike that are both functionally relevant and averse to change, we identified human neutralizing antibodies to highly conserved viral epitopes. Antibody fp.006 binds the fusion peptide and cross-reacts against coronaviruses of the four genera , including the nine human coronaviruses, through recognition of a conserved motif that includes the S2' site of proteolytic cleavage. Antibody hr2.016 targets the stem helix and neutralizes SARS-CoV-2 variants. Antibody sd1.040 binds to subdomain 1, synergizes with antibody rbd.042 for neutralization and, like fp.006 and hr2.016, protects mice when present as bispecific antibody. Thus, coldspot-guided antibody discovery reveals donor-derived neutralizing antibodies that are cross-reactive with Orthocoronavirinae , including SARS-CoV-2 variants.

Identifying cellular identities (both novel and well-studied) is one of the key use cases in single-cell transcriptomics. While supervised machine learning has been leveraged to automate cell annotation predictions for some time, there has been relatively little progress both in scaling neural networks to large data sets and in constructing models that generalize well across diverse tissues and biological contexts up to whole organisms. Here, we propose scTab, an automated, feature-attention-based cell type prediction model specific to tabular data, and train it using a novel data augmentation scheme across a large corpus of single-cell RNA-seq observations (22.2 million human cells in total). In addition, scTab leverages deep ensembles for uncertainty quantification. Moreover, we account for ontological relationships between labels in the model evaluation to accommodate for differences in annotation granularity across datasets. On this large-scale corpus, we show that cross-tissue annotation requires nonlinear models and that the performance of scTab scales in terms of training dataset size as well as model size - demonstrating the advantage of scTab over current state-of-the-art linear models in this context. Additionally, we show that the proposed data augmentation schema improves model generalization. In summary, we introduce a de novo cell type prediction model for single-cell RNA-seq data that can be trained across a large-scale collection of curated datasets from a diverse selection of human tissues and demonstrate the benefits of using deep learning methods in this paradigm. Our codebase, training data, and model checkpoints are publicly available at https://github.com/theislab/scTab to further enable rigorous benchmarks of foundation models for single-cell RNA-seq data.

Mutations in the phosphatidylinositol glycan biosynthesis class A (PIGA) gene cause a rare, X-linked recessive congenital disorder of glycosylation (CDG). PIGA-CDG is characterized by seizures, intellectual and developmental delay, and congenital malformations. The PIGA gene encodes an enzyme involved in the first step of GPI anchor biosynthesis. There are over 100 GPI anchored proteins that attach to the cell surface and are involved in cell signaling, immunity, and adhesion. Little is known about the pathophysiology of PIGA-CDG. Here we describe the first Drosophila model of PIGA-CDG and demonstrate that loss of PIG-A function in Drosophila accurately models the human disease. As expected, complete loss of PIG-A function is larval lethal. Heterozygous null animals appear healthy, but when challenged, have a seizure phenotype similar to what is observed in patients. To identify the cell-type specific contributions to disease, we generated neuron- and glia-specific knockdown of PIG-A. Neuron-specific knockdown resulted in reduced lifespan and a number of neurological phenotypes, but no seizure phenotype. Glia-knockdown also reduced lifespan and, notably, resulted in a very strong seizure phenotype. RNAseq analyses demonstrated that there are fundamentally different molecular processes that are disrupted when PIG-A function is eliminated in different cell types. In particular, loss of PIG-A in neurons resulted in upregulation of glycolysis, but loss of PIG-A in glia resulted in upregulation of protein translation machinery. Here we demonstrate that Drosophila is a good model of PIGA-CDG and provide new data resources for future study of PIGA-CDG and other GPI anchor disorders.

Chromosomal instability (CIN) is the persistent reshuffling of cancer karyotypes via chromosome mis-segregation during cell division. In cancer, CIN exists at varying levels that have differential effects on tumor progression. However, mis-segregation rates remain challenging to assess in human cancer despite an array of available measures. We evaluated measures of CIN by comparing quantitative methods using specific, inducible phenotypic CIN models of chromosome bridges, pseudobipolar spindles, multipolar spindles, and polar chromosomes. For each, we measured CIN fixed and timelapse fluorescence microscopy, chromosome spreads, 6-centromere FISH, bulk transcriptomics, and single cell DNA sequencing (scDNAseq). As expected, microscopy of tumor cells in live and fixed samples correlated well (R=0.77; p<0.01) and sensitively detect CIN. Cytogenetics approaches include chromosome spreads and 6-centromere FISH, which also correlate well (R=0.77; p<0.01) but had limited sensitivity for lower rates of CIN. Bulk genomic DNA signatures and bulk transcriptomic scores, CIN70 and HET70, did not detect CIN. By contrast, single-cell DNA sequencing (scDNAseq) detects CIN with high sensitivity, and correlates very well with imaging methods (R=0.83; p<0.01). In summary, single-cell methods such as imaging, cytogenetics, and scDNAseq can measure CIN, with the latter being the most comprehensive method accessible to clinical samples. To facilitate comparison of CIN rates between phenotypes and methods, we propose a standardized unit of CIN: Mis-segregations per Diploid Division (MDD). This systematic analysis of common CIN measures highlights the superiority of single-cell methods and provides guidance for measuring CIN in the clinical setting.

The use of deep learning algorithms to extract meaningful diagnostic features from biomedical images holds the promise to improve patient care given the expansion of digital pathology. Among these deep learning models, Vision Transformer (ViT) models have been demonstrated to capture long-range spatial relationships with more robust prediction power for image classification tasks than regular convolutional neural network (CNN) models, and also better model interpretability. Model interpretation is important for understanding and elucidating how a deep learning model makes predictions, especially for developing transparent models for digital pathology. However, like other deep learning algorithms, with limited annotated biomedical imaging datasets, ViT models are prone to poor performance due to overfitting, which can lead to false predictions due to random noise. Overfitting affects model interpretation when predictions are made out of random noise. To address this issue, we introduce a novel metric - Training Attention and Validation Attention Consistency (TAVAC) - for evaluating ViT model degree of overfitting on imaging datasets and quantifying the reproducibility of interpretation. Specifically, the model interpretation is performed by comparing the high-attention regions in the image between training and testing. We test the method on four publicly available image classification datasets and two independent breast cancer histological image datasets. All overfitted models exhibited significantly lower TAVAC scores than the good-fit models. The TAVAC score quantitatively measures the level of generalization of model interpretation on a fine-grained level for small groups of cells in each H&E image, which cannot be provided by traditional performance evaluation metrics like prediction accuracy. Furthermore, the application of TAVAC extends beyond medical diagnostic AI models; it enhances the monitoring of model interpretative reproducibility at pixel-resolution in basic research, to reveal critical spatial patterns and cellular structures essential to understanding biological processes and disease mechanisms. TAVAC sets a new standard for evaluating the performance of deep learning model interpretation and provides a method for determining the significance of high-attention regions detected from the attention map of the biomedical images.

Computational models that predict an individual's response to a vaccine offer the potential for mechanistic insights and personalized vaccination strategies. These models are increasingly derived from systems vaccinology studies that generate immune profiles from human cohorts pre- and post-vaccination. Most of these studies involve relatively small cohorts and profile the response to a single vaccine. The ability to assess the performance of the resulting models would be improved by comparing their performance on independent datasets, as has been done with great success in other areas of biology such as protein structure predictions. To transfer this approach to system vaccinology studies, we established a prototype platform that focuses on the evaluation of Computational Models of Immunity to Pertussis Booster vaccinations (CMI-PB). A community resource, CMI-PB generates experimental data for the explicit purpose of model evaluation, which is performed through a series of annual data releases and associated contests. We here report on our experience with the first such 'dry run' for a contest where the goal was to predict individual immune responses based on pre-vaccination multi-omic profiles. Over 30 models adopted from the literature were tested, but only one was predictive, and was based on age alone. The performance of new models built using CMI-PB training data was much better, but varied significantly based on the choice of pre-vaccination features used and the model building strategy. This suggests that previously published models developed for other vaccines do not generalize well to Pertussis Booster vaccination. Overall, these results reinforced the need for comparative analysis across models and datasets that CMI-PB aims to achieve. We are seeking wider community engagement for our first public prediction contest, which will open in early 2024.

Diabetic neuropathy (DN) is a debilitating disorder characterized by mechanical allodynia and sensory loss. It has traditionally been considered a small-fiber neuropathy, defined by the loss of free nerve endings in the epidermis. Free nerve endings, however, are nociceptors which may not be the only sensor for mechanical pain. To investigate the role of mechanoreceptors, specifically Meissner corpuscles, in the development of diabetic mechanical allodynia, our study focused on the keratinocyte-secreted brain-derived neurotrophic factor (BDNF) and its transcriptional regulator sirtuin 1 (SIRT1). Wild-type DN mice demonstrated decreased SIRT1 deacetylase activity, leading to a decrease in BDNF expression and Meissner corpuscle densities in foot skin. Epidermal SIRT1 knockout (KO) mice developed exacerbated DN phenotypes including severe mechanical allodynia, markedly reduced Meissner corpuscles, and subcutaneous Aß axon degeneration. Among the major skin-derived neurotrophic factors, only BDNF was down-regulated in epidermal SIRT1 KO mice. With similar KO phenotypes, epidermal BDNF appeared to belong to the same pathway as SIRT1 in modulating diabetic mechanical allodynia. Furthermore, mice overexpressing epidermal SIRT1 showed BDNF up-regulation and improved DN phenotypes, supporting an important role of epidermal SIRT1 and BDNF in skin sensory apparatus regeneration and functional recovery in the setting of diabetes.

Metabolic pathways are a human-defined grouping of life sustaining biochemical reactions, metabolites being both the reactants and products of these reactions. But many public datasets include identified metabolites whose pathway involvement is unknown, hindering metabolic interpretation. To address these shortcomings, various machine learning models, including those trained on data from the Kyoto Encyclopedia of Genes and Genomes (KEGG), have been developed to predict the pathway involvement of metabolites based on their chemical descriptions; however, these prior models are based on old metabolite KEGG-based datasets, including one benchmark dataset that is invalid due to the presence of over 1500 duplicate entries. Therefore, we have developed a new benchmark dataset derived from the KEGG following optimal standards of scientific computational reproducibility and including all source code needed to update the benchmark dataset as KEGG changes. We have used this new benchmark dataset with our atom coloring methodology to develop and compare the performance of Random Forest, XGBoost, and multilayer perceptron with autoencoder models generated from our new benchmark dataset. Best overall weighted average performance across 1000 unique folds was an F1-score of 0.8180 and Matthews correlation coefficient of 0.7933, which was provided by XGBoost binary classification models for 11 KEGG-defined pathway categories.

Birth-death models play a key role in phylodynamic analysis for their interpretation in terms of key epidemiological parameters. In particular, models with piecewise-constant rates varying at different epochs in time, to which we refer as episodic birth-death-sampling (EBDS) models, are valuable for their reflection of changing transmission dynamics over time. A challenge, however, that persists with current time-varying model inference procedures is their lack of computational efficiency. This limitation hinders the full utilization of these models in large-scale phylodynamic analyses, especially when dealing with high-dimensional parameter vectors that exhibit strong correlations. We present here a linear-time algorithm to compute the gradient of the birth-death model sampling density with respect to all time-varying parameters, and we implement this algorithm within a gradient-based Hamiltonian Monte Carlo (HMC) sampler to alleviate the computational burden of conducting inference under a wide variety of structures of, as well as priors for, EBDS processes. We assess this approach using three different real world data examples, including the HIV epidemic in Odesa, Ukraine, seasonal influenza A/H3N2 virus dynamics in New York state, America, and Ebola outbreak in West Africa. HMC sampling exhibits a substantial efficiency boost, delivering a 10- to 200-fold increase in minimum effective sample size per unit-time, in comparison to a Metropolis-Hastings-based approach. Additionally, we show the robustness of our implementation in both allowing for flexible prior choices and in modeling the transmission dynamics of various pathogens by accurately capturing the changing trend of viral effective reproductive number.

Due to the importance of 4R tau in the pathogenicity of primary tauopathies, it has been challenging to model these diseases in iPSC-derived neurons, which express very low levels of 4R tau. To address this problem we have developed a panel of isogenic iPSC lines carrying the MAPT splice-site mutations S305S, S305I or S305N, derived from four different donors. All three mutations significantly increased the proportion of 4R tau expression in iPSC-neurons and astrocytes, with up to 80% 4R transcripts in S305N neurons from as early as 4 weeks of differentiation. Transcriptomic and functional analyses of S305 mutant neurons revealed shared disruption in glutamate signaling and synaptic maturity, but divergent effects on mitochondrial bioenergetics. In iPSC-astrocytes, S305 mutations induced lysosomal disruption and inflammation and exacerbated internalization of exogenous tau that may be a precursor to the glial pathologies observed in many tauopathies. In conclusion, we present a novel panel of human iPSC lines that express unprecedented levels of 4R tau in neurons and astrocytes. These lines recapitulate previously characterized tauopathy-relevant phenotypes, but also highlight functional differences between the wild type 4R and mutant 4R proteins. We also highlight the functional importance of MAPT expression in astrocytes. These lines will be highly beneficial to tauopathy researchers enabling a more complete understanding of the pathogenic mechanisms underlying 4R tauopathies across different cell types.

The "innate-like" T cell compartment, known as Tinn, represents a diverse group of T cells that straddle the boundary between innate and adaptive immunity, having the ability to mount rapid responses following activation. In mice, this ability is acquired during thymic development. We explored the transcriptional landscape of Tinn compared to conventional T cells (Tconv) in the human thymus and blood using single cell RNA sequencing and flow cytometry. We reveal that in human blood, the majority of Tinn cells, including iNKT, MAIT, and Vδ2+Vγ9+ T cells, share an effector program characterized by the expression of unique chemokine and cytokine receptors, and cytotoxic molecules. This program is driven by specific transcription factors, distinct from those governing Tconv cells. Conversely, only a fraction of thymic Tinn cells displays an effector phenotype, while others share transcriptional features with developing Tconv cells, indicating potential divergent developmental pathways. Unlike the mouse, human Tinn cells do not differentiate into multiple effector subsets but develop a mixed type I/type III effector potential. To conduct a comprehensive cross-species analysis, we constructed a murine Tinn developmental atlas and uncovered additional species-specific distinctions, including the absence of type II Tinn cells in humans, which implies distinct immune regulatory mechanisms across species. The study provides insights into the development and functionality of Tinn cells, emphasizing their role in immune responses and their potential as targets for therapeutic interventions.

Non-alcoholic fatty liver disease (NAFLD) is a liver manifestation of metabolic syndrome, and is estimated to affect one billion individuals worldwide. An increased intake of a high-fat diet (HFD) and sugar-sweetened beverages are risk-factors for NAFLD development, but how their combined intake promotes progression to a more severe form of liver injury is unknown. Here we show that fructose metabolism via ketohexokinase (KHK) C isoform increases endoplasmic reticulum (ER) stress in a dose dependent fashion, so when fructose is coupled with a HFD intake it leads to unresolved ER stress. Conversely, a liver-specific knockdown of KHK in C57BL/6J male mice consuming fructose on a HFD is adequate to improve the NAFLD activity score and exert a profound effect on the hepatic transcriptome. Overexpression of KHK-C in cultured hepatocytes is sufficient to induce ER stress in fructose free media. Upregulation of KHK-C is also observed in genetically obesity ob/ob, db/db and lipodystrophic FIRKO male mice, whereas KHK knockdown in these mice improves metabolic function. Additionally, in over 100 inbred strains of male or female mice hepatic KHK expression correlates positively with adiposity, insulin resistance, and liver triglycerides. Similarly, in 241 human subjects and their controls, hepatic Khk expression is upregulated in early, but not late stages of NAFLD. In summary, we describe a novel role of KHK-C in triggering ER stress, which offers a mechanistic understanding of how the combined intake of fructose and a HFD propagates the development of metabolic complications.

Recent deep learning techniques hold promise to enable IMU-driven kinetic assessment; however, they require large extents of ground reaction force (GRF) data to serve as labels for supervised model training. We thus propose using existing self-supervised learning (SSL) techniques to leverage large IMU datasets to pre-train deep learning models, which can improve the accuracy and data efficiency of IMU-based GRF estimation.

Newly formed memories are not passively stored for future retrieval; rather, they are reactivated offline and thereby strengthened and transformed. However, reactivation is not a uniform process: it occurs throughout different states of consciousness, including conscious rehearsal during wakefulness and unconscious processing during both wakefulness and sleep. In this study, we explore the consequences of reactivation during conscious and unconscious awake states. Forty-one participants learned associations consisting of adjective-object-position triads. Objects were clustered into distinct semantic groups (e.g., multiple fruits, vehicles, musical instruments) which allowed us to examine the consequences of reactivation on semantically-related memories. After an extensive learning phase, some triads were reactivated consciously, through cued retrieval, or unconsciously, through subliminal priming. In both conditions, the adjective was used as the cue. Reactivation impacted memory for the most distal association (i.e., the spatial position of associated objects) in a consciousness-dependent and memory-strength-dependent manner. First, conscious reactivation of a triad resulted in a weakening of other semantically related memories, but only those that were initially more accurate (i.e., memories with lower pre-reactivation spatial errors). This is similar to what has been previously demonstrated in studies employing retrieval-induced forgetting designs. Unconscious reactivation, on the other hand, benefited memory selectively for weak cued items. Semantically linked associations were not impaired, but rather integrated with the reactivated memory. Taken together, our results demonstrate that conscious and unconscious reactivation of memories during wakefulness have qualitatively different consequences on memory for distal associations. Effects are memory-strength-dependent, as has been shown for reactivation during sleep. Results support a consciousness-dependent inhibition account, according to which unconscious reactivation involves less inhibitory dynamics than conscious reactivation, thus allowing more liberal spread of activation. Our findings set the stage for additional exploration into the role of consciousness in memory structuring.

Monoclonal antibodies (mAbs) are the basis of treatments and diagnostics for pathogens and other biological phenomena. We conducted a structural characterization of mAbs against the N-terminal domain of nucleocapsid protein (NP NTD ) from SARS-CoV-2 using small angle X-ray scattering (SAXS). Our solution-based results distinguished the mAbs' flexibility and how this flexibility impacts the assembly of multiple mAbs on an antigen. By pairing two mAbs that bind different epitopes on the NP NTD , we show that flexible mAbs form a closed sandwich-like complex. With rigid mAbs, a juxtaposition of the Fabs is prevented, enforcing a linear arrangement of the mAb pair, which facilitates further mAb polymerization. In a modified sandwich ELISA, we show the rigid mAb-pairings with linear polymerization led to increased NP NTD detection sensitivity. These enhancements can expedite the development of more sensitive and selective antigen-detecting point-of-care lateral flow devices (LFA), key for early diagnosis and epidemiological studies of SARS-CoV-2 and other pathogens.

DNA methylation data offers valuable insights into various aspects of mammalian biology. The recent introduction and large-scale application of the mammalian methylation array has significantly expanded the availability of such data across conserved sites in many mammalian species. In our study, we consider 13,245 samples profiled on this array encompassing 348 species and 59 tissues from 746 species-tissue combinations. While having some coverage of many different species and tissue types, this data captures only 3.6% of potential species-tissue combinations. To address this gap, we developed CMImpute (Cross-species Methylation Imputation), a method based on a Conditional Variational Autoencoder, to impute DNA methylation for non-profiled species-tissue combinations. In cross-validation, we demonstrate that CMImpute achieves a strong correlation with actual observed values, surpassing several baseline methods. Using CMImpute we imputed methylation data for 19,786 new species-tissue combinations. We believe that both CMImpute and our imputed data resource will be useful for DNA methylation analyses across a wide range of mammalian species.

Time series analysis is critical for understanding brain signals and their relationship to behavior and cognition. Cluster-based permutation tests (CBPT) are commonly used to analyze a variety of electrophysiological signals including EEG, MEG, ECoG, and sEEG data without a priori assumptions about specific temporal effects. However, two major limitations of CBPT include the inability to directly analyze experiments with multiple fixed effects and the inability to account for random effects (e.g. variability across subjects). Here, we propose a flexible multi-step hypothesis testing strategy using CBPT with Linear Mixed Effects Models (LMEs) and Generalized Linear Mixed Effects Models (GLMEs) that can be applied to a wide range of experimental designs and data types.