Is hate speech detection the solution the world wants?

The machine learning (ML) research community has landed on automated hate speech detection as the vital tool in the mitigation of bad behavior online. However, it is not clear that this is a widely supported view outside of the ML world. Such a disconnect can have implications for whether automated detection tools are accepted or adopted. Here we lend insight into how other key stakeholders understand the challenge of addressing hate speech and the role automated detection plays in solving it. To do so, we develop and apply a structured approach to dissecting the discourses used by online platform companies, governments, and not-for-profit organizations when discussing hate speech. We find that, where hate speech mitigation is concerned, there is a profound disconnect between the computer science research community and other stakeholder groups-which puts progress on this important problem at serious risk. We identify urgent steps that need to be taken to incorporate computational researchers into a single, coherent, multistakeholder community that is working towards civil discourse online.

Identification of a novel endoplasmic reticulum stress response element regulated by XBP1.

Understanding the regulatory mechanisms mediating PRNP gene expression is highly relevant to elucidating normal cellular prion protein (PrP) function(s) and the transmissibility of prion protein neurodegenerative diseases. Here, luciferase reporter assays showed that an endoplasmic reticulum stress element (ERSE)-like element, CCAAT-N26-CCACG in the human PRNP promoter, is regulated by ER stress and X-box-binding protein 1 (XBP1) but not by activating transcription factor 6 α (ATF6α). Bioinformatics identified the ERSE-26 motif in 37 other human genes in the absence of canonical ERSE sites except for three genes. Several of these genes are associated with a synaptic function or are involved in oxidative stress. Brefeldin A, tunicamycin, and thapsigargin ER stressors induced gene expression of PRNP and four randomly chosen ERSE-26-containing genes, ERLEC1, GADD45B, SESN2, and SLC38A5, in primary human neuron cultures or in the breast carcinoma MCF-7 cell line, although the level of the response depends on the gene analyzed, the genetic background of the cells, the cell type, and the ER stressor. Overexpression of XBP1 increased, whereas siRNA knockdown of XBP1 considerably reduced, PRNP and ERLEC1 mRNA levels in MCF-7 cells. Taken together, these results identify a novel ER stress regulator, which implicates the ER stress response in previously unrecognized cellular functions.

Understanding social networks requires more than two dimensions.

The proposed framework is insufficient to categorize and understand current evidence on decision making. There are some ambiguities in the questions asked that require additional distinctions between correctness and accuracy, decision making and learning, accuracy and confidence, and social influence and empowerment. Social learning techniques are not all the same: Behavior copying is quite different from theory passing. Sigmoidal acquisition curves are not unique to social learning and are often mistaken for other accelerating curves.

On the contributions of topological features to transcriptional regulatory network robustness.

BACKGROUND: Because biological networks exhibit a high-degree of robustness, a systemic understanding of their architecture and function requires an appraisal of the network design principles that confer robustness. In this project, we conduct a computational study of the contribution of three degree-based topological properties (transcription factor-target ratio, degree distribution, cross-talk suppression) and their combinations on the robustness of transcriptional regulatory networks. We seek to quantify the relative degree of robustness conferred by each property (and combination) and also to determine the extent to which these properties alone can explain the robustness observed in transcriptional networks. RESULTS: To study individual properties and their combinations, we generated synthetic, random networks that retained one or more of the three properties with values derived from either the yeast or E. coli gene regulatory networks. Robustness of these networks were estimated through simulation. Our results indicate that the combination of the three properties we considered explains the majority of the structural robustness observed in the real transcriptional networks. Surprisingly, scale-free degree distribution is, overall, a minor contributor to robustness. Instead, most robustness is gained through topological features that limit the complexity of the overall network and increase the transcription factor subnetwork sparsity. CONCLUSIONS: Our work demonstrates that (i) different types of robustness are implemented by different topological aspects of the network and (ii) size and sparsity of the transcription factor subnetwork play an important role for robustness induction. Our results are conserved across yeast and E Coli, which suggests that the design principles examined are present within an array of living systems.

The epidemiological impact of the Canadian COVID Alert app.

OBJECTIVES: We analyzed the effectiveness of the Canadian COVID Alert app on reducing COVID-19 infections and deaths due to the COVID-19 virus. METHODS: Two separate but complementary approaches were taken. First, we undertook a comparative study to assess how the adoption and usage of the COVID Alert app compared to those of similar apps deployed in other regions. Next, we used data from the COVID Alert server and a range of plausible parameter values to estimate the numbers of infections and deaths averted in Canada using a model that combines information on number of notifications, secondary attack rate, expected fraction of transmissions that could be prevented, quarantine effectiveness, and expected size of the full transmission chain in the absence of exposure notification. RESULTS: The comparative analysis revealed that the COVID Alert app had among the lowest adoption levels among apps that reported usage. Our model indicates that use of the COVID Alert app averted between 6284 and 10,894 infections across the six Canadian provinces where app usage was highest during the March-July 2021 period. This range is equivalent to 1.6-2.9% of the total recorded infections across Canada in that time. Using province-specific case fatality rates, 57-101 deaths were averted during the same period. The number of cases and deaths averted was greatest in Ontario, whereas the proportion of cases and deaths averted was greatest in Newfoundland and Labrador. App impact measures were reported so rarely and so inconsistently by other regions that the relative assessment of impact is inconclusive. CONCLUSION: While the nationwide rates are low, provinces with widespread adoption of the app showed high ratios of averted cases and deaths (upper bound was greater than 60% of averted cases). This finding suggests that the COVID Alert app, when adopted at sufficient levels, can be an effective public health tool for combatting a pandemic such as COVID-19.

RéSUMé: OBJECTIFS: Nous avons analysé l'efficacité de l'application canadienne Alerte COVID sur la réduction des infections à la COVID-19 et des décès dus au virus COVID-19. MéTHODES: Deux approches distinctes mais complémentaires ont été adoptées. D'abord, nous avons entrepris une étude comparative pour évaluer comment l'adoption et l'utilisation de l'application Alerte COVID se comparent à celles d'applications similaires déployées dans d'autres régions. Ensuite, nous avons utilisé les données du serveur Alerte COVID et plusieurs valeurs de paramètres plausibles pour estimer le nombre d'infections et de décès évités au Canada à l'aide d'un modèle combinant des informations sur le nombre de notifications, le taux d'attaque secondaire, la fraction attendue des transmissions pouvant être prévenues, l'efficacité de la mise en quarantaine et la taille attendue de la chaîne de transmission complète en l'absence de notification d'exposition. RéSULTATS: L'analyse comparative a révélé que l'application Alerte COVID avait l'un des niveaux d'adoption les plus bas parmi les applications qui ont rapporté une utilisation. Notre modèle indique que l'utilisation de l'application Alerte COVID a évité entre 6 284 et 10 894 infections dans les six provinces canadiennes où l'utilisation de l'application était la plus élevée au cours de la période de mars à juillet 2021. Ces nombres correspondent à 1,6 % à 2,9 % du nombre total d'infections enregistrées au Canada pendant cette période. En utilisant les taux de létalité propres à chaque province, 57 à 101 décès ont été évités au cours de la même période. Le nombre de cas et de décès évités était le plus élevé en Ontario, tandis que la proportion de cas et de décès évités était la plus élevée à Terre-Neuve-et-Labrador. Les mesures d'impact des applications dans d'autres régions ont été rapportées si rarement et de manière si incohérente que l'évaluation relative de l'impact n'est pas concluante. CONCLUSION: Bien que les taux à l'échelle nationale soient faibles, les provinces où l'adoption de l'application était généralisée ont affiché des ratios élevés de cas et de décès évités (la limite supérieure était supérieure à 60 % des cas évités). Ces résultats suggèrent que l'application Alerte COVID, lorsqu'elle est adoptée à des niveaux suffisants, peut être un outil de santé publique efficace pour lutter contre une pandémie telle que la COVID-19.

Dilations and degeneracy in network controllability.

Network controllability asserts a perspective that the structure-the location of edges that connect nodes-of the network contains important information about fundamental characteristics of our ability to change the behavior that evolves on these networks. It can be used, for example, to determine the parts of the system that when influenced by outside controlling signals, can ultimately steer the behavior of the entire network. One of the challenges in utilizing the ideas from network controllability on real systems is that there is typically more than one potential solution (often many) suggested by the topology of the graph that perform equally well. Picking a single candidate from this degenerate solution set over others should be properly motivated, however, to-date our understanding of how these different options are related has been limited. In this work, we operationalize the existing notion of a dilation into a framework that provides clarity on the source of this control degeneracy and further elucidates many of the existing results surrounding degeneracy in the literature.

GS2: an efficiently computable measure of GO-based similarity of gene sets.

MOTIVATION: The growing availability of genome-scale datasets has attracted increasing attention to the development of computational methods for automated inference of functional similarities among genes and their products. One class of such methods measures the functional similarity of genes based on their distance in the Gene Ontology (GO). To measure the functional relatedness of a gene set, these measures consider every pair of genes in the set, and the average of all pairwise distances is calculated. However, as more data becomes available and gene sets used for analysis become larger, such pair-based calculation becomes prohibitive. RESULTS: In this article, we propose GS(2) (GO-based similarity of gene sets), a novel GO-based measure of gene set similarity that is computable in linear time in the size of the gene set. The measure quantifies the similarity of the GO annotations among a set of genes by averaging the contribution of each gene's GO terms and their ancestor terms with respect to the GO vocabulary graph. To study the performance of our method, we compared our measure with an established pair-based measure when run on gene sets with varying degrees of functional similarities. In addition to a significant speed improvement, our method produced comparable similarity scores to the established method. Our method is available as a web-based tool and an open-source Python library. AVAILABILITY: The web-based tools and Python code are available at: http://bioserver.cs.rice.edu/gs2.

International Research Symposium on Ankyloblepharon-Ectodermal Defects-Cleft Lip/Palate (AEC) syndrome.

Ankyloblepharon-ectodermal defects-cleft lip/palate (AEC) syndrome (Hay-Wells syndrome, MIM #106220) is a rare autosomal dominant ectodermal dysplasia syndrome. It is due to mutations in the TP63 gene, known to be a regulatory gene with many downstream gene targets. TP63 is important in the differentiation and proliferation of the epidermis, as well as many other processes including limb and facial development. It is also known that mutations in TP63 lead to skin erosions. These erosions, especially on the scalp, are defining features of AEC syndrome and cause significant morbidity and mortality in these patients. It was this fact that led to the 2003 AEC Skin Erosion Workshop. That conference laid the groundwork for the International Research Symposium for AEC Syndrome held at Texas Children's Hospital in 2006. The conference brought together the largest cohort of individuals with AEC syndrome, along with a multitude of physicians and scientists. The overarching goals were to define the clinical and pathologic findings for improved diagnostic criteria, to obtain tissue samples for further study and to define future research directions. The symposium was successful in accomplishing these aims as detailed in this conference report. Following our report, we also present 11 manuscripts within this special section that outline the collective clinical, pathologic, and mutational data from 18 individuals enrolled in the concurrent Baylor College of Medicine IRB-approved protocol: Characterization of AEC syndrome. These collaborative findings will hopefully provide a stepping-stone to future translational projects of TP63 and TP63-related syndromes.

The signaling petri net-based simulator: a non-parametric strategy for characterizing the dynamics of cell-specific signaling networks.

Reconstructing cellular signaling networks and understanding how they work are major endeavors in cell biology. The scale and complexity of these networks, however, render their analysis using experimental biology approaches alone very challenging. As a result, computational methods have been developed and combined with experimental biology approaches, producing powerful tools for the analysis of these networks. These computational methods mostly fall on either end of a spectrum of model parameterization. On one end is a class of structural network analysis methods; these typically use the network connectivity alone to generate hypotheses about global properties. On the other end is a class of dynamic network analysis methods; these use, in addition to the connectivity, kinetic parameters of the biochemical reactions to predict the network's dynamic behavior. These predictions provide detailed insights into the properties that determine aspects of the network's structure and behavior. However, the difficulty of obtaining numerical values of kinetic parameters is widely recognized to limit the applicability of this latter class of methods. Several researchers have observed that the connectivity of a network alone can provide significant insights into its dynamics. Motivated by this fundamental observation, we present the signaling Petri net, a non-parametric model of cellular signaling networks, and the signaling Petri net-based simulator, a Petri net execution strategy for characterizing the dynamics of signal flow through a signaling network using token distribution and sampling. The result is a very fast method, which can analyze large-scale networks, and provide insights into the trends of molecules' activity-levels in response to an external stimulus, based solely on the network's connectivity. We have implemented the signaling Petri net-based simulator in the PathwayOracle toolkit, which is publicly available at http://bioinfo.cs.rice.edu/pathwayoracle. Using this method, we studied a MAPK1,2 and AKT signaling network downstream from EGFR in two breast tumor cell lines. We analyzed, both experimentally and computationally, the activity level of several molecules in response to a targeted manipulation of TSC2 and mTOR-Raptor. The results from our method agreed with experimental results in greater than 90% of the cases considered, and in those where they did not agree, our approach provided valuable insights into discrepancies between known network connectivities and experimental observations.

PhyloNet: a software package for analyzing and reconstructing reticulate evolutionary relationships.

BACKGROUND: Phylogenies, i.e., the evolutionary histories of groups of taxa, play a major role in representing the interrelationships among biological entities. Many software tools for reconstructing and evaluating such phylogenies have been proposed, almost all of which assume the underlying evolutionary history to be a tree. While trees give a satisfactory first-order approximation for many families of organisms, other families exhibit evolutionary mechanisms that cannot be represented by trees. Processes such as horizontal gene transfer (HGT), hybrid speciation, and interspecific recombination, collectively referred to as reticulate evolutionary events, result in networks, rather than trees, of relationships. Various software tools have been recently developed to analyze reticulate evolutionary relationships, which include SplitsTree4, LatTrans, EEEP, HorizStory, and T-REX. RESULTS: In this paper, we report on the PhyloNet software package, which is a suite of tools for analyzing reticulate evolutionary relationships, or evolutionary networks, which are rooted, directed, acyclic graphs, leaf-labeled by a set of taxa. These tools can be classified into four categories: (1) evolutionary network representation: reading/writing evolutionary networks in a newly devised compact form; (2) evolutionary network characterization: analyzing evolutionary networks in terms of three basic building blocks - trees, clusters, and tripartitions; (3) evolutionary network comparison: comparing two evolutionary networks in terms of topological dissimilarities, as well as fitness to sequence evolution under a maximum parsimony criterion; and (4) evolutionary network reconstruction: reconstructing an evolutionary network from a species tree and a set of gene trees. CONCLUSION: The software package, PhyloNet, offers an array of utilities to allow for efficient and accurate analysis of evolutionary networks. The software package will help significantly in analyzing large data sets, as well as in studying the performance of evolutionary network reconstruction methods. Further, the software package supports the proposed eNewick format for compact representation of evolutionary networks, a feature that allows for efficient interoperability of evolutionary network software tools. Currently, all utilities in PhyloNet are invoked on the command line.

Confounding factors in HGT detection: statistical error, coalescent effects, and multiple solutions.

Prokaryotic organisms share genetic material across species boundaries by means of a process known as horizontal gene transfer (HGT). This process has great significance for understanding prokaryotic genome diversification and unraveling their complexities. Phylogeny-based detection of HGT is one of the most commonly used methods for this task, and is based on the fundamental fact that HGT may cause gene trees to disagree with one another, as well as with the species phylogeny. Using these methods, we can compare gene and species trees, and infer a set of HGT events to reconcile the differences among these trees. In this paper, we address three factors that confound the detection of the true HGT events, including the donors and recipients of horizontally transferred genes. First, we study experimentally the effects of error in the estimated gene trees (statistical error) on the accuracy of inferred HGT events. Our results indicate that statistical error leads to overestimation of the number of HGT events, and that HGT detection methods should be designed with unresolved gene trees in mind. Second, we demonstrate, both theoretically and empirically, that based on topological comparison alone, the number of HGT scenarios that reconcile a pair of species/gene trees may be exponential. This number may be reduced when branch lengths in both trees are estimated correctly. This set of results implies that in the absence of additional biological information, and/or a biological model of how HGT occurs, multiple HGT scenarios must be sought, and efficient strategies for how to enumerate such solutions must be developed. Third, we address the issue of lineage sorting, how it confounds HGT detection, and how to incorporate it with HGT into a single stochastic framework that distinguishes between the two events by extending population genetics theories. This result is very important, particularly when analyzing closely related organisms, where coalescent effects may not be ignored when reconciling gene trees. In addition to these three confounding factors, we consider the problem of enumerating all valid coalescent scenarios that constitute plausible species/gene tree reconciliations, and develop a polynomial-time dynamic programming algorithm for solving it. This result bears great significance on reducing the search space for heuristics that seek reconciliation scenarios. Finally, we show, empirically, that the locality of incongruence between a pair of trees has an impact on the numbers of HGT and coalescent reconciliation scenarios.

Correlations in the degeneracy of structurally controllable topologies for networks.

Many dynamic systems display complex emergent phenomena. By directly controlling a subset of system components (nodes) via external intervention it is possible to indirectly control every other component in the system. When the system is linear or can be approximated sufficiently well by a linear model, methods exist to identify the number and connectivity of a minimum set of external inputs (constituting a so-called minimal control topology, or MCT). In general, many MCTs exist for a given network; here we characterize a broad ensemble of empirical networks in terms of the fraction of nodes and edges that are always, sometimes, or never a part of an MCT. We study the relationships between the measures, and apply the methodology to the T-LGL leukemia signaling network as a case study. We show that the properties introduced in this report can be used to predict key components of biological networks, with potentially broad applications to network medicine.

Hypothesis generation in signaling networks.

Biological signaling networks comprise the chemical processes by which cells detect and respond to changes in their environment. Such networks have been implicated in the regulation of important cellular activities, including cellular reproduction, mobility, and death. Though technological and scientific advances have facilitated the rapid accumulation of information about signaling networks, utilizing these massive information resources has become infeasible except through computational methods and computer-based tools. To date, visualization and simulation tools have received significant emphasis. In this paper, we present a graph-theoretic formalization of biological signaling network models that are in wide but informal use, and formulate two problems on the graph: the Constrained Downstream and Minimum Knockout Problems. Solutions to these problems yield qualitative tools for generating hypotheses about the networks, which can then be experimentally tested in a laboratory setting. Using established graph algorithms, we provide a solution to the Constrained Downstream Problem. We also show that the Minimum Knockout Problem is NP-Hard, propose a heuristic, and assess its performance. In tests on the Epidermal Growth Factor Receptor (EGFR) network, we find that our heuristic reports the correct solution to the problem in seconds. Source code for the implementations of both solutions is available from the authors upon request.

Estimating the minimum control count of random network models.

The study of controllability of complex networks has introduced the minimum number of controls required for full controllability as a new network measure of interest. This network measure, like many others, is non-trivial to compute. As a result, establishing the significance of minimum control counts (MCCs) in real networks using random network null models is expensive. Here we derive analytic estimates for the expected MCCs of networks drawn from three commonly-used random network models. Our estimates show good agreement with exact control counts. Furthermore, the analytic expressions we derive offer insights into the structures within each random network model that induce the need for controls.

Topological constraints on network control profiles.

Network models are designed to capture properties of empirical networks and thereby provide insight into the processes that underlie the formation of complex systems. As new information concerning network structure becomes available, it becomes possible to design models that more fully capture the properties of empirical networks. A recent advance in our understanding of network structure is the control profile, which summarizes the structural controllability of a network in terms of source nodes, external dilations, and internal dilations. Here, we consider the topological properties-and their formation mechanisms-that constrain the control profile. We consider five representative empirical categories of internal-dilation dominated networks, and show that the number of source and sink nodes, the form of the in- and out-degree distributions, and local complexity (e.g., cycles) shape the control profile. We evaluate network models that are sufficient to produce realistic control profiles, and conclude that holistic network models should similarly consider these properties.

Network models. Response to Comment on "Control profiles of complex networks".

Campbell, Shea, and Albert propose an adaptation of the Barabási-Albert model of network formation that permits a level of tuning of the control profiles of these networks. We point out some limitations and generalizations of this method as well as highlight opportunities for future work to refine formation mechanisms to provide control profile tuning in synthetic networks.

Control profiles of complex networks.

Studying the control properties of complex networks provides insight into how designers and engineers can influence these systems to achieve a desired behavior. Topology of a network has been shown to strongly correlate with certain control properties; here we uncover the fundamental structures that explain the basis of this correlation. We develop the control profile, a statistic that quantifies the different proportions of control-inducing structures present in a network. We find that standard random network models do not reproduce the kinds of control profiles that are observed in real-world networks. The profiles of real networks form three well-defined clusters that provide insight into the high-level organization and function of complex systems.

A method for the automated, reliable retrieval of publication-citation records.

BACKGROUND: Publication records and citation indices often are used to evaluate academic performance. For this reason, obtaining or computing them accurately is important. This can be difficult, largely due to a lack of complete knowledge of an individual's publication list and/or lack of time available to manually obtain or construct the publication-citation record. While online publication search engines have somewhat addressed these problems, using raw search results can yield inaccurate estimates of publication-citation records and citation indices. METHODOLOGY: In this paper, we present a new, automated method that produces estimates of an individual's publication-citation record from an individual's name and a set of domain-specific vocabulary that may occur in the individual's publication titles. Because this vocabulary can be harvested directly from a research web page or online (partial) publication list, our method delivers an easy way to obtain estimates of a publication-citation record and the relevant citation indices. Our method works by applying a series of stringent name and content filters to the raw publication search results returned by an online publication search engine. In this paper, our method is run using Google Scholar, but the underlying filters can be easily applied to any existing publication search engine. When compared against a manually constructed data set of individuals and their publication-citation records, our method provides significant improvements over raw search results. The estimated publication-citation records returned by our method have an average sensitivity of 98% and specificity of 72% (in contrast to raw search result specificity of less than 10%). When citation indices are computed using these records, the estimated indices are within of the true value 10%, compared to raw search results which have overestimates of, on average, 75%. CONCLUSIONS: These results confirm that our method provides significantly improved estimates over raw search results, and these can either be used directly for large-scale (departmental or university) analysis or further refined manually to quickly give accurate publication-citation records.