Search

Nervous systems extract and process information from the environment to alter animal behavior and physiology. Despite progress in understanding how different stimuli are represented by changes in neuronal activity, less is known about how they affect broader neural network properties. We developed a framework for using graph-theoretic features of neural network activity to predict ecologically relevant stimulus properties, in particular stimulus identity. We used the transparent nematode, Caenorhabditis elegans, with its small nervous system to define neural network features associated with various chemosensory stimuli. We first immobilized animals using a microfluidic device and exposed their noses to chemical stimuli while monitoring changes in neural activity of more than 50 neurons in the head region. We found that graph-theoretic features, which capture patterns of interactions between neurons, are modulated by stimulus identity. Further, we show that a simple machine learning classifier trained using graph-theoretic features alone, or in combination with neural activity features, can accurately predict salt stimulus. Moreover, by focusing on putative causal interactions between neurons, the graph-theoretic features were almost twice as predictive as the neural activity features. These results reveal that stimulus identity modulates the broad, network-level organization of the nervous system, and that graph theory can be used to characterize these changes., Author summary Animals use their nervous systems to detect and respond to changes in their external environment. A central challenge in computational neuroscience is to determine how specific properties of these stimuli affect interactions between neurons. While most studies have focused on the neurons in the sensory periphery, recent advances allow us to probe how the rest of the nervous system responds to sensory stimulation. We recorded activity of neurons within the C. elegans head region while the animal was exposed to various chemosensory stimuli. We then used computational methods to identify various stimuli by analyzing neural activity. Specifically, we used a combination of population-level activity statistics (e.g., average, standard deviation, frequency-based measures) and graph-theoretic features of functional network structure (e.g., transitivity, which is the existence of strongly connected triplets of neurons) to accurately predict salt stimulus. Our method is general and can be used across species, particularly in instances where the identities of individual neurons are unknown. These results also suggest that neural activity downstream of the sensory periphery contains a signature of changes in the environment.​

A critical facet of adjusting one’s behaviour after succeeding or failing at a task is assigning responsibility for the ultimate outcome. Humans have trait- and state-like tendencies to implicate aspects of their own behaviour (called ‘internal’ ascriptions) or facets of the particular task or Lady Luck (’chance’). However, how these tendencies interact with actual performance is unclear. We designed a novel task in which subjects had to learn the likelihood of achieving their goals, and the extent to which this depended on their efforts. High internality (Levenson I-score) was associated with decision making patterns that are less vulnerable to failure. Our computational analyses suggested that this depended heavily on the adjustment in the perceived achievability of riskier goals following failure. We found beliefs about chance not to be explanatory of choice behaviour in our task. Beliefs about powerful others were strong predictors of behaviour, but only when subjects lacked substantial influence over the outcome. Our results provide an evidentiary basis for heuristics and learning differences that underlie the formation and maintenance of control expectations by the self., Author summary How success in tasks depends on our efforts is largely a feature of what is known as the controllability of the situation or environment. This quantity should determine the way we approach, adapt to, and perform tasks. In novel settings, it can only be our expectations about controllability that exert an effect, for instance determining the balance of our focus between features of achievability and potential reward; or affecting prior notions about what exactly is achievable or what we can or cannot do. All of these might be different between aversive and appetitive domains. To study these issues, we designed a novel task in which subjects have to learn about a new environment, and analyzed their behavior using a rich computational model. We found that expectations about controllability played a particularly important role in influencing learning, but in a way that differed between positive and negative outcomes. In particular, high expectations about controllability were tied to higher learning rates of unachievability given negative outcomes, guarding subjects from further loss, and preserving their subjective optimistic expectations about control. Our findings can be interpreted within a theoretical framework which implicates control expectations in individual learning differences, and fit well within modern theories of learning in aversive contexts and serotonergic function.

Biomarkers predict World Trade Center-Lung Injury (WTC-LI); however, there remains unaddressed multicollinearity in our serum cytokines, chemokines, and high-throughput platform datasets used to phenotype WTC-disease. To address this concern, we used automated, machine-learning, high-dimensional data pruning, and validated identified biomarkers. The parent cohort consisted of male, never-smoking firefighters with WTC-LI (FEV1, %Pred< lower limit of normal (LLN); n = 100) and controls (n = 127) and had their biomarkers assessed. Cases and controls (n = 15/group) underwent untargeted metabolomics, then feature selection performed on metabolites, cytokines, chemokines, and clinical data. Cytokines, chemokines, and clinical biomarkers were validated in the non-overlapping parent-cohort via binary logistic regression with 5-fold cross validation. Random forests of metabolites (n = 580), clinical biomarkers (n = 5), and previously assayed cytokines, chemokines (n = 106) identified that the top 5% of biomarkers important to class separation included pigment epithelium-derived factor (PEDF), macrophage derived chemokine (MDC), systolic blood pressure, macrophage inflammatory protein-4 (MIP-4), growth-regulated oncogene protein (GRO), monocyte chemoattractant protein-1 (MCP-1), apolipoprotein-AII (Apo-AII), cell membrane metabolites (sphingolipids, phospholipids), and branched-chain amino acids. Validated models via confounder-adjusted (age on 9/11, BMI, exposure, and pre-9/11 FEV1, %Pred) binary logistic regression had AUCROC [0.90(0.84–0.96)]. Decreased PEDF and MIP-4, and increased Apo-AII were associated with increased odds of WTC-LI. Increased GRO, MCP-1, and simultaneously decreased MDC were associated with decreased odds of WTC-LI. In conclusion, automated data pruning identified novel WTC-LI biomarkers; performance was validated in an independent cohort. One biomarker—PEDF, an antiangiogenic agent—is a novel, predictive biomarker of particulate-matter-related lung disease. Other biomarkers—GRO, MCP-1, MDC, MIP-4—reveal immune cell involvement in WTC-LI pathogenesis. Findings of our automated biomarker identification warrant further investigation into these potential pharmacotherapy targets., Author summary Disease related to air pollution causes millions of deaths annually. Large swathes of the general population, as well as certain occupations such as 1st responders and military personnel, are exposed to particulate matter (PM)—a major component of air pollution. Our longitudinal cohort of FDNY firefighters exposed to the World Trade Center dust cloud on 9/11 is a unique research opportunity to characterize the impact of a single, intense PM exposure by looking at pre- and post-exposure phenotype; however, PM-related lung disease and PM’s systemic effects are complex and call for a systems biological approach coupled with novel computational modelling techniques to fully understand pathogenesis. In the present study, we integrate clinical and environmental biomarkers with the serum metabolome, cytokines, and chemokines to develop a model for early disease detection and identification of potential signaling cascades of PM-related chronic lung disease.

The quantitative modeling of semantic representations in the brain plays a key role in understanding the neural basis of semantic processing. Previous studies have demonstrated that word vectors, which were originally developed for use in the field of natural language processing, provide a powerful tool for such quantitative modeling. However, whether semantic representations in the brain revealed by the word vector-based models actually capture our perception of semantic information remains unclear, as there has been no study explicitly examining the behavioral correlates of the modeled brain semantic representations. To address this issue, we compared the semantic structure of nouns and adjectives in the brain estimated from word vector-based brain models with that evaluated from human behavior. The brain models were constructed using voxelwise modeling to predict the functional magnetic resonance imaging (fMRI) response to natural movies from semantic contents in each movie scene through a word vector space. The semantic dissimilarity of brain word representations was then evaluated using the brain models. Meanwhile, data on human behavior reflecting the perception of semantic dissimilarity between words were collected in psychological experiments. We found a significant correlation between brain model- and behavior-derived semantic dissimilarities of words. This finding suggests that semantic representations in the brain modeled via word vectors appropriately capture our perception of word meanings., Author summary Word vectors, which have been originally developed in the field of engineering (natural language processing), have been extensively leveraged in neuroscience studies to model semantic representations in the human brain. These studies have attempted to model brain semantic representations by associating them with the meanings of thousands of words via a word vector space. However, there has been no study explicitly examining whether the brain semantic representations modeled by word vectors actually capture our perception of semantic information. To address this issue, we compared the semantic representational structure of words in the brain estimated from word vector-based brain models with that evaluated from behavioral data in psychological experiments. The results revealed a significant correlation between these model- and behavior-derived semantic representational structures of words. This indicates that the brain semantic representations modeled using word vectors actually reflect the human perception of word meanings. Our findings contribute to the establishment of word vector-based brain modeling as a useful tool in studying human semantic processing.

Analyses of gene set differential coexpression may shed light on molecular mechanisms underlying phenotypes and diseases. However, differential coexpression analyses of conceptually similar individual studies are often inconsistent and underpowered to provide definitive results. Researchers can greatly benefit from an open-source application facilitating the aggregation of evidence of differential coexpression across studies and the estimation of more robust common effects. We developed Meta Gene Set Coexpression Analysis (MetaGSCA), an analytical tool to systematically assess differential coexpression of an a priori defined gene set by aggregating evidence across studies to provide a definitive result. In the kernel, a nonparametric approach that accounts for the gene-gene correlation structure is used to test whether the gene set is differentially coexpressed between two comparative conditions, from which a permutation test p-statistic is computed for each individual study. A meta-analysis is then performed to combine individual study results with one of two options: a random-intercept logistic regression model or the inverse variance method. We demonstrated MetaGSCA in case studies investigating two human diseases and identified pathways highly relevant to each disease across studies. We further applied MetaGSCA in a pan-cancer analysis with hundreds of major cellular pathways in 11 cancer types. The results indicated that a majority of the pathways identified were dysregulated in the pan-cancer scenario, many of which have been previously reported in the cancer literature. Our analysis with randomly generated gene sets showed excellent specificity, indicating that the significant pathways/gene sets identified by MetaGSCA are unlikely false positives. MetaGSCA is a user-friendly tool implemented in both forms of a Web-based application and an R package “MetaGSCA”. It enables comprehensive meta-analyses of gene set differential coexpression data, with an optional module of post hoc pathway crosstalk network analysis to identify and visualize pathways having similar coexpression profiles., Author summary Analyses of gene set differential coexpression often shed light on molecular mechanisms underlying phenotypes and diseases. However, results from conceptually similar individual studies are often inconsistent and underpowered to reach definitive conclusions. We provide an open-source application facilitating the aggregation of evidence of differential coexpression across studies and the estimation of more robust common effects, with an optional module of post hoc pathway crosstalk network analysis to identify and visualize pathways having similar coexpression profiles. We established the usefulness of MetaGSCA via case studies of chronic kidney disease and non-small cell lung cancer, and applied it to a pan-cancer analysis of 11 cancer types. We further demonstrated the tool with 100 randomly generated gene sets and showed excellent specificity, indicating low false positive rates.

Transfer entropy (TE) is a widely used measure of directed information flows in a number of domains including neuroscience. Many real-world time series for which we are interested in information flows come in the form of (near) instantaneous events occurring over time. Examples include the spiking of biological neurons, trades on stock markets and posts to social media, amongst myriad other systems involving events in continuous time throughout the natural and social sciences. However, there exist severe limitations to the current approach to TE estimation on such event-based data via discretising the time series into time bins: it is not consistent, has high bias, converges slowly and cannot simultaneously capture relationships that occur with very fine time precision as well as those that occur over long time intervals. Building on recent work which derived a theoretical framework for TE in continuous time, we present an estimation framework for TE on event-based data and develop a k-nearest-neighbours estimator within this framework. This estimator is provably consistent, has favourable bias properties and converges orders of magnitude more quickly than the current state-of-the-art in discrete-time estimation on synthetic examples. We demonstrate failures of the traditionally-used source-time-shift method for null surrogate generation. In order to overcome these failures, we develop a local permutation scheme for generating surrogate time series conforming to the appropriate null hypothesis in order to test for the statistical significance of the TE and, as such, test for the conditional independence between the history of one point process and the updates of another. Our approach is shown to be capable of correctly rejecting or accepting the null hypothesis of conditional independence even in the presence of strong pairwise time-directed correlations. This capacity to accurately test for conditional independence is further demonstrated on models of a spiking neural circuit inspired by the pyloric circuit of the crustacean stomatogastric ganglion, succeeding where previous related estimators have failed., Author summary Transfer Entropy (TE) is an information-theoretic measure commonly used in neuroscience to measure the directed statistical dependence between a source and a target time series, possibly also conditioned on other processes. Along with measuring information flows, it is used for the inference of directed functional and effective networks from time series data. The currently-used technique for estimating TE on neural spike trains first time-discretises the data and then applies a straightforward plug-in information-theoretic estimation procedure. This approach has numerous drawbacks: it has high bias, cannot capture relationships occurring on both fine and large timescales simultaneously, converges very slowly as more data is obtained, and indeed does not even converge to the correct value for any practical non-vanishing discretisation scale. We present a new estimator for TE which operates in continuous time and demonstrate, via application to synthetic examples, that it addresses these problems and can reliably differentiate statistically significant flows from (conditionally) independent spike trains. Further, we also apply it to more biologically-realistic spike trains obtained from a biophysical model inspired by the pyloric circuit of the crustacean stomatogastric ganglion; our correct inference of directed conditional dependence and independence between neurons here provides an important validation for our approach where similar methods have previously failed.

Transcriptional bursts render substantial biological noise in cellular transcriptomes. Here, we investigated the theoretical extent of allelic expression resulting from transcriptional bursting and how it compared to the amount biallelic, monoallelic and allele-biased expression observed in single-cell RNA-sequencing (scRNA-seq) data. We found that transcriptional bursting can explain the allelic expression patterns observed in single cells, including the frequent observations of autosomal monoallelic gene expression. Importantly, we identified that the burst frequency largely determined the fraction of cells with monoallelic expression, whereas the burst size had little effect on monoallelic observations. The high consistency between the bursting model predictions and scRNA-seq observations made it possible to assess the heterogeneity of a group of cells as their deviation in allelic observations from the expected. Finally, both burst frequency and size contributed to allelic imbalance observations and reinforced that studies of allelic imbalance can be confounded from the inherent noise in transcriptional bursting. Altogether, we demonstrate that allele-level transcriptional bursting renders widespread, although predictable, amounts of monoallelic and biallelic expression in single cells and cell populations., Author summary Genes are transcribed into RNA and further translated into proteins. The maternal and paternal copy of each gene are typically transcribed independently, and transcription itself occur in discrete stochastic bursts (transcriptional bursts). Pioneering single-cell analysis of RNA across cells revealed abundant fluctuations in the amounts of maternal and paternal RNA in cells, with frequent observations of RNA from only the maternal or paternal gene copy (monoallelic expression). In this study, we investigated to which extent the observed monoallelic expression across single cells can be explained by transcriptional bursting. We demonstrate that the process of transcriptional bursting is sufficient to explain the amount of monoallelic expression, and we further demonstrate that the frequency of bursts mainly determines the frequency of monoallelic observations. Furthermore, we show that transcriptional bursts may lead to false positive observations of monoallelic expression across cell populations. Therefore, stochastic transcription renders large fluctuations in allelic origin of RNA in cells over time, including frequent monoallelic observations when profiling single cells.

The field of transcriptomics uses and measures mRNA as a proxy of gene expression. There are currently two major platforms in use for quantifying mRNA, microarray and RNA-Seq. Many comparative studies have shown that their results are not always consistent. In this study we aim to find a robust method to increase comparability of both platforms enabling data analysis of merged data from both platforms. We transformed high dimensional transcriptomics data from two different platforms into a lower dimensional, and biologically relevant dataset by calculating enrichment scores based on gene set collections for all samples. We compared the similarity between data from both platforms based on the raw data and on the enrichment scores. We show that the performed data transforms the data in a biologically relevant way and filters out noise which leads to increased platform concordance. We validate the procedure using predictive models built with microarray based enrichment scores to predict subtypes of breast cancer using enrichment scores based on sequenced data. Although microarray and RNA-Seq expression levels might appear different, transforming them into biologically relevant gene set enrichment scores significantly increases their correlation, which is a step forward in data integration of the two platforms. The gene set collections were shown to contain biologically relevant gene sets. More in-depth investigation on the effect of the composition, size, and number of gene sets that are used for the transformation is suggested for future research., Author summary The field of transcriptomics uses and measures mRNA as a proxy of gene expression. There are currently two major platforms in use for quantifying mRNA, microarray and RNA-Seq. Many comparative studies have shown that their results are not always consistent. In this study we aim to find a robust method to increase comparability of both platforms enabling data analysis of merged data from both platforms. We transformed the high dimensional transcriptomics data from the two different platforms into lower dimensional, and biologically relevant gene set scores. These gene sets were defined a-priori as specific combination of genes (e.g. up-regulated in a certain pathway). We observed that although microarray and RNA-Seq expression levels might appear different, using these gene sets to transform the data significantly increases their correlation. This is a step forward in data integration of the two platforms. More in-depth investigation on the effect of the composition, size, and number of gene sets that are used for the transformation is suggested for future research.

Learning transfer (i.e. accelerated learning over a series of structurally related learning tasks) differentiates species and age-groups, but the evolutionary and developmental implications of such differences are unclear. To this end, the relational schema induction paradigm employing tasks that share algebraic (group-like) structures was introduced to contrast stimulus-independent (relational) versus stimulus-dependent (associative) learning processes. However, a theory explaining this kind of relational learning transfer has not been forthcoming beyond a general appeal to some form of structure-mapping, as typically assumed in models of analogy. In this paper, we provide a theory of relational schema induction as a “reconstruction” process: the algebraic structure underlying transfer is reconstructed by comparing stimulus relations, learned within each task, for structural consistency across tasks—formally, the theory derives from a category theory version of Tannakian reconstruction. The theory also applies to non-human studies of relational concepts, thereby placing human and non-human transfer on common ground for sharper comparison and contrast. As the theory and paradigm do not depend on linguistic ability, we also have a way for pinpointing where aspects of human learning diverge from other species without begging the question of language., Author summary Learning transfer—an improvement in the rate of learning over a series of learning tasks—differs between species and age-groups, but the implications of such differences are unclear. Relational schema induction is a form of learning transfer that highlights relational aspects of learning. However, a theory explaining this kind of induction and transfer has not been forthcoming beyond a general appeal to some form of analogical mapping, as typically assumed in models of analogy. I present a theory of relational schema induction as a “reconstruction” process: the common structure affording transfer is reconstructed by comparing stimulus relations learned within each task for structural consistency across tasks. The theory also applies to non-human studies of relational concepts, thereby placing human and non-human studies on common ground for a clearer comparison and contrast.

Information transfer, measured by transfer entropy, is a key component of distributed computation. It is therefore important to understand the pattern of information transfer in order to unravel the distributed computational algorithms of a system. Since in many natural systems distributed computation is thought to rely on rhythmic processes a frequency resolved measure of information transfer is highly desirable. Here, we present a novel algorithm, and its efficient implementation, to identify separately frequencies sending and receiving information in a network. Our approach relies on the invertible maximum overlap discrete wavelet transform (MODWT) for the creation of surrogate data in the computation of transfer entropy and entirely avoids filtering of the original signals. The approach thereby avoids well-known problems due to phase shifts or the ineffectiveness of filtering in the information theoretic setting. We also show that measuring frequency-resolved information transfer is a partial information decomposition problem that cannot be fully resolved to date and discuss the implications of this issue. Last, we evaluate the performance of our algorithm on simulated data and apply it to human magnetoencephalography (MEG) recordings and to local field potential recordings in the ferret. In human MEG we demonstrate top-down information flow in temporal cortex from very high frequencies (above 100Hz) to both similarly high frequencies and to frequencies around 20Hz, i.e. a complex spectral configuration of cortical information transmission that has not been described before. In the ferret we show that the prefrontal cortex sends information at low frequencies (4-8 Hz) to early visual cortex (V1), while V1 receives the information at high frequencies (> 125 Hz)., Author summary Systems in nature that perform computations typically consist of a large number of relatively simple but interacting parts. In human brains, for example, billions of neurons work together to enable our cognitive abilities. This well-orchestrated teamwork requires information to be exchanged very frequently. In many cases this exchange happens rhythmically and, therefore, it seems beneficial for our understanding of physical systems if we could link the information exchange to specific rhythms. We here present a method to determine which rhythms send, and which rhythms receive information. Since many rhythms can interact at both sender and receiver side, we show that the above problem is tightly linked to partial information decomposition—an intriguing problem from information theory only solved recently, and only partly. We applied our novel method to information transfer in the human inferior temporal cortex, a brain region relevant for object perception, and unexpectedly found information transfer originating at very high frequencies at 100Hz and then forking to be received at both similarly high but also much lower frequencies around 20Hz. These results overturn the current standard assumption that low frequencies send information to high frequencies.

We consider the following general family of algorithmic problems that arises in transcriptomics, metabolomics and other fields: given a weighted graph G and a subset of its nodes S, find subsets of S that show significant connectedness within G. A specific solution to this problem may be defined by devising a scoring function, the Maximum Clique problem being a classic example, where S includes all nodes in G and where the score is defined by the size of the largest subset of S fully connected within G. Major practical obstacles for the plethora of algorithms addressing this type of problem include computational efficiency and, particularly for more complex scores which take edge weights into account, the computational cost of permutation testing, a statistical procedure required to obtain a bound on the p-value for a connectedness score. To address these problems, we developed CTD, “Connect the Dots”, a fast algorithm based on data compression that detects highly connected subsets within S. CTD provides information-theoretic upper bounds on p-values when S contains a small fraction of nodes in G without requiring computationally costly permutation testing. We apply the CTD algorithm to interpret multi-metabolite perturbations due to inborn errors of metabolism and multi-transcript perturbations associated with breast cancer in the context of disease-specific Gaussian Markov Random Field networks learned directly from respective molecular profiling data., Author summary A frequently encountered “omic” analysis problem is to identify a subset of nodes within a weighted graph G that is both highly connected in G and belongs to S, a subset of nodes in G. For example, G may represent a biological pathway, kinetic network model, biological interaction network, or a network learned directly from data, where edges represent co-variation relationships between abundances of molecular variables. S may be the set of molecular variables that are perturbed in an individual case or in a set of disease cases relative to controls. In this work, we develop a novel information-theoretic formulation of this problem and a local search algorithm that obviate the need for computationally costly permutation testing, a statistical procedure which is typically required to establish rigorous p-value bounds for other scoring-based methods.

Existing studies have demonstrated that dysregulation of microRNAs (miRNAs or miRs) is involved in the initiation and progression of cancer. Many efforts have been devoted to identify microRNAs as potential biomarkers for cancer diagnosis, prognosis and therapeutic targets. With the rapid development of miRNA sequencing technology, a vast amount of miRNA expression data for multiple cancers has been collected. These invaluable data repositories provide new paradigms to explore the relationship between miRNAs and cancer. Thus, there is an urgent need to explore the complex cancer-related miRNA-gene patterns by integrating multi-omics data in a pan-cancer paradigm. In this study, we present a tensor sparse canonical correlation analysis (TSCCA) method for identifying cancer-related miRNA-gene modules across multiple cancers. TSCCA is able to overcome the drawbacks of existing solutions and capture both the cancer-shared and specific miRNA-gene co-expressed modules with better biological interpretations. We comprehensively evaluate the performance of TSCCA using a set of simulated data and matched miRNA/gene expression data across 33 cancer types from the TCGA database. We uncover several dysfunctional miRNA-gene modules with important biological functions and statistical significance. These modules can advance our understanding of miRNA regulatory mechanisms of cancer and provide insights into miRNA-based treatments for cancer., Author summary MicroRNAs (miRNAs) are a class of small non-coding RNAs. Previous studies have revealed that miRNA-gene regulatory modules play key roles in the occurrence and development of cancer. However, little has been done to discover miRNA-gene regulatory modules from a pan-cancer view. Thus, it is urgently needed to develop new methods to explore the complex cancer-related miRNA-gene patterns by integrating multi-omics data of multi-cancers. To build the connections between miRNA-gene regulatory modules across different cancer types, we propose a tensor sparse canonical correlation analysis (TSCCA) method. Our specific contributions are two-fold: (1) We propose a sparse statistical learning model TSCCA and an efficient block-coordinate descent algorithm to solve it. (2) We apply TSCCA to a multi-omics data set of 33 cancer types from TCGA and identify some cancer-related miRNA-gene modules with important biological functions and statistical significance.

The prevailing paradigm for the analysis of biological data involves comparing groups of replicates from different conditions (e.g. control and treatment) to statistically infer features that discriminate them (e.g. differentially expressed genes). However, many situations in modern genomics such as single-cell omics experiments do not fit well into this paradigm because they lack true replicates. In such instances, spectral techniques could be used to rank features according to their degree of consistency with an underlying metric structure without the need to cluster samples. Here, we extend spectral methods for feature selection to abstract simplicial complexes and present a general framework for clustering-independent analysis. Combinatorial Laplacian scores take into account the topology spanned by the data and reduce to the ordinary Laplacian score when restricted to graphs. We demonstrate the utility of this framework with several applications to the analysis of gene expression and multi-modal genomic data. Specifically, we perform differential expression analysis in situations where samples cannot be grouped into distinct classes, and we disaggregate differentially expressed genes according to the topology of the expression space (e.g. alternative paths of differentiation). We also apply this formalism to identify genes with spatial patterns of expression using fluorescence in-situ hybridization data and to establish associations between genetic alterations and global expression patterns in large cross-sectional studies. Our results provide a unifying perspective on topological data analysis and manifold learning approaches to the analysis of large-scale biological datasets., Author summary Manifold learning methods have emerged as a way of analyzing the large high-dimensional data sets that are currently generated in many areas of science. They assume the data has been sampled from an unknown manifold which is approximated with a graph and utilize spectral graph techniques to perform unsupervised feature selection and dimensionality reduction. However, graphs provide only partial approximations to manifolds, precluding the application to features with a complex combinatorial structure. Relatedly, these methods cannot take into account the topology of the manifold. In this work, we extend spectral methods for feature selection to topological spaces built from data and present a general framework for feature selection. We present specific applications of this framework to clustering-independent analysis of gene expression and multi-modal genomic data. In particular, using these methods, we perform differential expression analysis in situations where samples cannot be grouped into distinct classes, and we disaggregate the results according to topological features of the expression space. In addition, we identify genes with spatial patterns of expression using spatially-resolved transcriptomic data and establish associations between genetic alterations and global expression patterns in large cross-sectional cancer studies.

T cells use their T-cell receptors (TCRs) to scan other cells for antigenic peptides presented by MHC molecules (pMHC). If a TCR encounters a pMHC, it can trigger a signalling pathway that could lead to the activation of the T cell and the initiation of an immune response. It is currently not clear how the binding of pMHC to the TCR initiates signalling within the T cell. One hypothesis is that conformational changes in the TCR lead to further downstream signalling. Here we investigate four different TCRs in their free state as well as in their pMHC bound state using large scale molecular simulations totalling 26 000 ns. We find that the dynamical features within TCRs differ significantly between unbound TCR and TCR/pMHC simulations. However, apart from expected results such as reduced solvent accessibility and flexibility of the interface residues, these features are not conserved among different TCR types. The presence of a pMHC alone is not sufficient to cause cross-TCR-conserved dynamical features within a TCR. Our results argue against models of TCR triggering involving conserved allosteric conformational changes., Author summary The interaction between T-cells and other cells is one of the most important interactions in the human immune system. If T-cells are not triggered major parts of the immune system cannot be activated or are not working effectively. Despite many years of research the exact mechanism of how a T-cell is initially triggered is not clear. One hypothesis is that conformational changes within the T-cell receptor (TCR) can cause further downstream signalling within the T-cell. In this study we computationally investigate the dynamics of four different TCRs in their free and bound configuration. Our large scale simulations show that all four TCRs react to binding in different ways. In some TCRs mainly the areas close to the binding region are affected while in other TCRs areas further apart from the binding region are also affected. Our results argue against a conserved structural activation mechanism across different types of TCRs.

Identification of differentially expressed genes (DEGs) is well recognized to be variable across independent replications of genome-wide transcriptional studies. These are often employed to characterize disease state early in the process of discovery and prioritize novel targets aimed at addressing unmet medical need. Increasing reproducibility of biological findings from these studies could potentially positively impact the success rate of new clinical interventions. This work demonstrates that statistically sound combination of gene expression data with prior knowledge about biology in the form of large protein interaction networks can yield quantitatively more reproducible observations from studies characterizing human disease. The novel concept of Well-Associated Proteins (WAPs) introduced herein—gene products significantly associated on protein interaction networks with the differences in transcript levels between control and disease—does not require choosing a differential expression threshold and can be computed efficiently enough to enable false discovery rate estimation via permutation. Reproducibility of WAPs is shown to be on average superior to that of DEGs under easily-quantifiable conditions suggesting that they can yield a significantly more robust description of disease. Enhanced reproducibility of WAPs versus DEGs is first demonstrated with four independent data sets focused on systemic sclerosis. This finding is then validated over thousands of pairs of data sets obtained by random partitions of large studies in several other diseases. Conditions that individual data sets must satisfy to yield robust WAP scores are examined. Reproducible identification of WAPs can potentially benefit drug target selection and precision medicine studies., Author summary Gene expression studies are commonly used to characterize biological systems. Genes identified in such experiments as expressed at different levels between conditions (e.g. healthy vs. disease) can indicate biological functions that are important in this context. However, it is well-recognized that such findings can vary substantially across independent investigations. We quantified reproducibility here under a conservative control scenario that partitions a given data set in two, independently of experimental conditions, for multiple data sets characterizing several diseases in humans. Furthermore, we have shown that it is possible to obtain more reproducible findings than DEGs, which we term Well-Associated Proteins, characterizing differences in gene expression between healthy and disease states. This was accomplished by combining gene expression and prior knowledge of functional relationships between gene products accumulated over many studies and publications. Resulting Well-Associated Proteins can be computed efficiently enough to enable permutation controls and demonstrate on average higher reproducibility than differentially expressed genes, both within and across data sets. This suggests that Well-Associated Proteins may better reflect differences in biology when comparing disease and healthy states than DEGs, thus representing an important step towards identification of key disease drivers.

To develop a complete description of sensory encoding, it is necessary to account for trial-to-trial variability in cortical neurons. Using a linear model with terms corresponding to the visual stimulus, mouse running speed, and experimentally measured neuronal correlations, we modeled short term dynamics of L2/3 murine visual cortical neurons to evaluate the relative importance of each factor to neuronal variability within single trials. We find single trial predictions improve most when conditioning on the experimentally measured local correlations in comparison to predictions based on the stimulus or running speed. Specifically, accurate predictions are driven by positively co-varying and synchronously active functional groups of neurons. Including functional groups in the model enhances decoding accuracy of sensory information compared to a model that assumes neuronal independence. Functional groups, in encoding and decoding frameworks, provide an operational definition of Hebbian assemblies in which local correlations largely explain neuronal responses on individual trials., Author summary V1 neurons exhibit substantial response variability while being massively interconnected, complicating the development of a comprehensive theory of visual encoding. Some work has pointed to a global description, in which stimuli and spontaneous behaviors are the main drivers of neuronal responses. However, other work concerning the retina and elsewhere suggest that local correlations between neighboring neurons shape a target neuron’s single trial responses to a much greater extent than global variables. In this work, we compare and evaluate local and global descriptions of variable single trial responses in mouse V1. We compute pairwise correlations of neuronal activity after accounting for global variables, and define the “functional group” of a neuron to be those neurons exhibiting correlated activity over short time scales. We find that the knowledge of the activity of a target neuron’s functional group enables much more accurate predictions of single trial responses as compared to knowledge of the stimulus condition or mouse locomotion. Moreover, we show that the informative correlations are strong, positive, and are exhibited on short time scales. Finally, we find that local correlations enable accurate decoding of the visual stimulus. These results point to a local description of the visual system over short time scales.

Individual instances of cancer are primarily a result of a combination of a small number of genetic mutations (hits). Knowing the number of such mutations is a prerequisite for identifying specific combinations of carcinogenic mutations and understanding the etiology of cancer. We present a mathematical model for estimating the number of hits based on the distribution of somatic mutations. The model is fundamentally different from previous approaches, which are based on cancer incidence by age. Our somatic mutation based model is likely to be more robust than age-based models since it does not require knowing or accounting for the highly variable mutation rate, which can vary by over three orders of magnitude. In fact, we find that the number of somatic mutations at diagnosis is weakly correlated with age at cancer diagnosis, most likely due to the extreme variability in mutation rates between individuals. Comparing the distribution of somatic mutations predicted by our model to the actual distribution from 6904 tumor samples we estimate the number of hits required for carcinogenesis for 17 cancer types. We find that different cancer types exhibit distinct somatic mutational profiles corresponding to different numbers of hits. Why might different cancer types require different numbers of hits for carcinogenesis? The answer may provide insight into the unique etiology of different cancer types., Author summary Cancer is primarily a result of genetic mutations. Each individual instance of cancer is initiated by a specific combination of a small number of mutations (hits). In trying to identify these combinations of mutations, it is important to know how many hits to look for. However, there are conflicting estimates for the number of hits. We present a fundamentally different model for estimating the number of hits. We found that the number hits ranges from two-eight depending on cancer type. These findings may provide insight into the unique characteristics of different cancer types.

Chemical graph generators are software packages to generate computer representations ofchemical structuresadhering to certainboundary conditions. Their development is a research topic ofcheminformatics. Chemical graph generators are used in areas such as virtuallibrarygeneration indrug design, inmolecular designwith specified properties, called inverseQSAR/QSPR, as well as inorganic synthesis design,retrosynthesisor in systems forcomputer-assisted structure elucidation(CASE). CASE systems again have regained interest for the structure elucidation of unknowns in computationalmetabolomics, a current area ofcomputational biology.

The limited capacity of recent memory inevitably leads to partial memory of past stimuli. There is also evidence that behavioral and neural responses to novel or rare stimuli are dependent on one’s memory of past stimuli. Thus, these responses may serve as a probe of different individuals’ remembering and forgetting characteristics. Here, we utilize two lossy compression models of stimulus sequences that inherently involve forgetting, which in addition to being a necessity under many conditions, also has theoretical and behavioral advantages. One model is based on a simple stimulus counter and the other on the Information Bottleneck (IB) framework which suggests a more general, theoretically justifiable principle for biological and cognitive phenomena. These models are applied to analyze a novelty-detection event-related potential commonly known as the P300. The trial-by-trial variations of the P300 response, recorded in an auditory oddball paradigm, were subjected to each model to extract two stimulus-compression parameters for each subject: memory length and representation accuracy. These parameters were then utilized to estimate the subjects’ recent memory capacity limit under the task conditions. The results, along with recently published findings on single neurons and the IB model, underscore how a lossy compression framework can be utilized to account for trial-by-trial variability of neural responses at different spatial scales and in different individuals, while at the same time providing estimates of individual memory characteristics at different levels of representation using a theoretically-based parsimonious model., Author summary Surprise responses reflect expectations based on preceding stimuli representations, and hence can be used to infer the characteristics of memory utilized for a task. We suggest a quantitative method for extracting an individual estimate of effective memory capacity dedicated for a task based on the correspondence between a theoretical surprise model and electrophysiological single-trial surprise responses. We demonstrate this method on EEG responses recorded while participants were performing a simple auditory task; we show the correspondence between the theoretical and physiological surprise, and calculate an estimate of the utilized memory. The generality of this framework allows it to be applied to different EEG features that reflect different modes and levels of the processing hierarchy, as well as other physiological measures of surprise responses. Future studies may use this framework to construct a handy diagnostic tool for a quantitative, individualized characterization of memory-related disorders.

Machine learning algorithms are becoming increasingly popular for decoding psychological constructs based on neural data. However, as a step towards bridging the gap between theory-driven cognitive neuroscience and data-driven decoding approaches, there is a need for methods that allow to interpret trained decoding models. The present study demonstrates grouped model reliance as a model-agnostic permutation-based approach to this problem. Grouped model reliance indicates the extent to which a trained model relies on conceptually related groups of variables, such as frequency bands or regions of interest in electroencephalographic (EEG) data. As a case study to demonstrate the method, random forest and support vector machine models were trained on within-participant single-trial EEG data from a Sternberg working memory task. Participants were asked to memorize a sequence of digits (0–9), varying randomly in length between one, four and seven digits, where EEG recordings for working memory load estimation were taken from a 3-second retention interval. The present results confirm previous findings insofar as both random forest and support vector machine models relied on alpha-band activity in most subjects. However, as revealed by further analyses, patterns in frequency and topography varied considerably between individuals, pointing to more pronounced inter-individual differences than previously reported., Author summary Modern machine learning algorithms currently receive considerable attention for their predictive power in neural decoding applications. However, there is a need for methods that make such predictive models interpretable. In the present work, we address the problem of assessing which aspects of the input data a trained model relies upon to make predictions. We demonstrate the use of grouped model reliance as a generally applicable method for interpreting neural decoding models. Illustrating the method on a case study, we employed an experimental design in which a comparably small number of participants (10) completed a large number of trials (972) over three electroencephalography (EEG) recording sessions from a Sternberg working memory task. Trained decoding models consistently relied on predictor variables from the alpha frequency band, which is in line with existing research on the relationship between neural oscillations and working memory. However, our analyses also indicate large inter-individual variability with respect to the relation between activity patterns and working memory load in frequency and topography. We argue that grouped model reliance provides a useful tool to better understand the workings of (sometimes otherwise black box) decoding models.