Homogeneous Clusters Research Articles

Decentralized and robust privacy-preserving model using blockchain-enabled Federated Deep Learning in intelligent enterprises

In Federated Deep Learning (FDL), multiple local enterprises are allowed to train a model jointly. Then, they submit their local updates to the central server, and the server aggregates the updates to create a global model. However, trained models usually perform worse than centralized models, especially when the training data distribution is non-independent and identically distributed (non-IID). Because non-IID data harms the accuracy and performance of the model. Second, due to the centrality of federated learning (FL) and the untrustworthiness of enterprises, traditional FL solutions are vulnerable to security and privacy attacks. Therefore, to tackle this issue, we propose FEDANIL, a secure blockchain-enabled Federated Deep LeArning Model that improves enterprise models’ decentralized, performance, and tamper-proof properties, including two main phases. The first phase is proposed to address the non-IID challenge (label and feature distribution skew). In this phase, local models with similar data distributions are grouped into homogeneous clusters using the cosine similarity (CS) and affinity propagation (AP) techniques. Then, for each homogeneous cluster, Wasserstein Generative Adversarial Networks (WGAN) are used to deal with label and feature distribution skew. The second phase was adopted to address security and privacy concerns against poisoning and inference attacks via three steps. In the first step, data poisoning attacks are prevented by using CS. Then, in the second step, collude attacks were prevented by randomly selecting enterprises in the consortium blockchain. Finally, in the third step, model poisoning, membership inference, and reconstruction attacks have been prevented using the CKKS Fully Homomorphic Encryption (CKKS-FHE) technique and consortium blockchain. Extensive experiments were conducted using the Sent140, Fashion-MNIST, FEMNIST, and CIFAR-10 new real-world datasets to evaluate FedAnil’s robustness and performance. The simulation results demonstrate that FEDANIL satisfies FDL privacy-preserving requirements. In terms of convergence analysis, the model parameter obtained with the FEDANIL converges to the optimum of the model parameter. In addition, it performs better in terms of accuracy (more than 11, 15, and 24%) and computation overhead (less than 8, 10, and 15%) compared with baseline approaches, namely SHIELDFL, RVPFL, and RFA, respectively. The FEDANIL source code can be found on GitHub.11Code available on GitHub Repository: https://github.com/rezafotohi/FedAnil. For any questions about the code, please contact Fotohi.reza@gmail.com.

Investigating the contributing factors of crashes on interstate bridges in Louisiana using latent class clustering and association rule mining

Drivers on long bridges on interstates often face unique challenges, such as restricted lane width, inadequate shoulders, and a lack of clear zones for safe recoveries. Studies on understanding the factors that contribute to crash severity on such high-risk sections of interstates are limited. This research study applies latent class clustering (LCC) to detect homogeneous clusters while accounting for unobserved heterogeneity in a dataset of 10,036 crashes that occurred over a 6-year period (2015–2020) on eight selected bridges. Utilizing the LCC method, the research identifies four optimal clusters in bridge crashes, characterized by attributes such as ‘4-lane,’ ‘6-lane,’ ‘single-vehicle crashes,’ and ‘unknown driver.’ The Association Rule Mining (ARM) approach is used to identify the important collective factors to visible injury (KAB – fatal, severe and moderate) and property damage only (PDO or no injury). In Cluster 1 (4-lane), KAB and PDO crashes differ in collision type and visibility conditions, with rear-end crashes linked to KAB and sideswipe crashes to PDO. Cluster 2 (6-lane) shows similar distinctions but lacks specific lighting associations for PDO. In Cluster 3 (single-vehicle crashes), KAB involves moderate traffic and low visibility, while PDO has lower speed limits and non-dry surfaces. Cluster 4 (unknown driver), despite overrepresenting hit-and-run cases, underscores challenges in injury crash data collection in high-volume mobility scenarios. The discussions of the findings on the severity factors in this study are expected to help traffic safety engineers, policymakers, and planners to identify effective safety countermeasures on major elevated sections.

CROWD VIDEO SEQUENCES PROCESSING METHODS FOR DETERMINING THE CROWD MOTION PATTERNS

Nowadays, homogeneous objects clusters motion is one of the most important and rapidly developing computer vision and machine learning application. In this paper, we consider the crowd motion patterns determination by using motion maps that we calculate with FlowNet, a neural network examining motion of objects in a video sequence. This approach allows us to get information on the crowd direction and velocity with relation to other objects of scene, which plays the key role in behavior analysis and security establishment. Besides, we consider methods for preliminary video sequence processing, including frame combination, to estimate motion maps more precisely and improve the effectiveness of the dynamic scenes analysis.

Randomized feature selection based semi-supervised latent Dirichlet allocation for microbiome analysis

Health and disease are fundamentally influenced by microbial communities and their genes (the microbiome). An in-depth analysis of microbiome structure that enables the classification of individuals based on their health can be crucial in enhancing diagnostics and treatment strategies to improve the overall well-being of an individual. In this paper, we present a novel semi-supervised methodology known as Randomized Feature Selection based Latent Dirichlet Allocation (RFSLDA) to study the impact of the gut microbiome on a subject’s health status. Since the data in our study consists of fuzzy health labels, which are self-reported, traditional supervised learning approaches may not be suitable. As a first step, based on the similarity between documents in text analysis and gut-microbiome data, we employ Latent Dirichlet Allocation (LDA), a topic modeling approach which uses microbiome counts as features to group subjects into relatively homogeneous clusters, without invoking any knowledge of observed health status (labels) of subjects. We then leverage information from the observed health status of subjects to associate these clusters with the most similar health status making it a semi-supervised approach. Finally, a feature selection technique is incorporated into the model to improve the overall classification performance. The proposed method provides a semi-supervised topic modelling approach that can help handle the high dimensionality of the microbiome data in association studies. Our experiments reveal that our semi-supervised classification algorithm is effective and efficient in terms of high classification accuracy compared to popular supervised learning approaches like SVM and multinomial logistic model. The RFSLDA framework is attractive because it (i) enhances clustering accuracy by identifying key bacteria types as indicators of health status, (ii) identifies key bacteria types within each group based on estimates of the proportion of bacteria types within the groups, and (iii) computes a measure of within-group similarity to identify highly similar subjects in terms of their health status.

Public opinion about solar radiation management: A cross-cultural study in 20 countries around the world

Some argue that complementing climate change mitigation measures with solar radiation management (SRM) might prove a last resort to limit global warming to 1.5 °C. To make a socially responsible decision on whether to use SRM, it is important to consider also public opinion, across the globe and particularly in the Global South, which would face the greatest risks from both global warming and SRM. However, most research on public opinion about SRM stems from the Global North. We report findings from the first large-scale, cross-cultural study on the public opinion about SRM among the general public (N = 2,248) and students (N = 4,583) in 20 countries covering all inhabited continents, including five countries from the Global South and five ‘non-WEIRD’ (i.e. not Western, Educated, Industrialised, Rich, and Democratic) countries from the Global North. As public awareness of SRM is usually low, we provided participants with information on SRM, including key arguments in favour of and against SRM that appear in the scientific debate. On average, acceptability of SRM was significantly higher in the Global South than in the ‘non-WEIRD’ Global North, while acceptability in the ‘WEIRD’ Global North was in between. However, we found substantial variation within these clusters, especially in the ‘non-WEIRD’ Global North, suggesting that countries do not form homogenous clusters and should thus be considered individually. Moreover, the average participants’ views, while generally neither strong nor polarised, differed from some expert views in important ways, including that participants perceived SRM as only slightly effective in limiting global warming. Still, our data suggests overall a conditional, reluctant acceptance. That is, while on average, people think SRM would have mostly negative consequences, they may still be willing to tolerate it as a potential last resort to fight global warming, particularly if they think SRM has only minor negative (or even positive) impacts on humans and nature.

Abstract 900: Development and validation of an artificial-intelligence-based pathomics biomarker to predict resistance to first-line treatment in metastatic colorectal cancer

Abstract Background: The backbone chemotherapy of first-line standard of care (SOC) for microsatellite stable (MSS) metastatic colorectal cancer (mCRC) combines 5-Fluorouracil to oxaliplatin and/or irinotecan. There are no biomarkers to predict response, which is complete or long-lasting (CR/LLR) in 20-25% of patients, while 10-15% are primary refractory. The aim of this work is to develop a predictive biomarker, based on digital pathology images, that can help stratify patients according to their risk of resistance. Methods: We trained a supervised bag-of-words artificial intelligence (AI)-based model on a cohort of response outliers mCRC patients classified as “really sensitive” (RS) if they achieved CR or LLR >10 months to any SOC, or “really resistant” (RR) if progression occurred at first disease reassessment. Whole-slide Imaging (WSI) of the resected primary tumors were tiled into patches of 224 × 224 pixel (0.5μm/pixel). First-order and texture features were subsequently extracted from all tumoral tiles and grouped into homogenous clusters through a k-means algorithm (k=6). For each patient, the percentage of tiles belonging to each tiles’ cluster was computed to represent new features (called bag of words) with whom different machine learning classifiers were trained. Main clinicopathological features were matched to treatment response by Fisher’s exact test. Results: To date, we analyzed 82 response outlier patients, of whom 35 were classified as RS and 46 RR. Of them, patients identified at Italian centers were used as construction cohort (N=47; 27 RR and 20 RS) and those identified at Spanish centers were used a validation cohort (N=35; 19 RR and 16 RS). The best result was obtained using a stepwise logistic regression, reaching a negative predictive value (NPV) of 90% (18/20; 95% CI=70-97%) and 71% (10/14; 95% CI=49-87%), in the construction and validation sets. No standard clinicopathological features (including stage, RAS/BRAF status, histology and sidedness) was associated with the chance of being RR. Conclusions: We demonstrated that a pathomics signature has the potential to predict resistance to SOC in MSS mCRC. Further validation of these preliminary findings on a larger cohort of response outlier patients is ongoing. Citation Format: Luca Lazzari, Gianluca Mauri, Valentina Giannini, Debora Cafaro, Giulia Nicoletti, Caterina Marchiò, Andrea Sartore-Bianchi, Federica Marmorino, Maria Nieva Munoz, Nadia Saoudi Gonzalez, Alberto Puccini, Chiara Cremolini, Clara Montagut, Elena Elez, Stefania Sciallero, Enrico Berrino, Martina Carullo, Pietro Paolo Vitiello, Maria Costanza Aquilano, Martina Di Como, Emanuela Bonoldi, Salvatore Siena, Alberto Bardelli, Daniele Regge, Silvia Marsoni. Development and validation of an artificial-intelligence-based pathomics biomarker to predict resistance to first-line treatment in metastatic colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 900.

EU Countries Tax Evasion Clustering

The problem of tax evasion is becoming more and more prominent, and it is causing large losses in individual state budgets. However, European policies have taken significant steps to eliminate them over the last two decades. National policies also contribute to reducing the estimated extent of tax evasion, as measured by Schneider's MIMIC model. However, these tendencies are not the same in individual countries. Therefore, in our paper, we try to find pattern in tax evasion time trajectories of European Union countries and thus to create homogenous clusters that include countries with similar tax evasion situation. To meet our goal, we use panel data clustering method on our data with information about tax evasion of EU countries from 2000 to 2019. As a next step, we compare created cluster from the perspective of quality indicators of public institutions.

Overlapping coefficient in network-based semi-supervised clustering

AbstractNetwork-based Semi-Supervised Clustering (NeSSC) is a semi-supervised approach for clustering in the presence of an outcome variable. It uses a classification or regression model on resampled versions of the original data to produce a proximity matrix that indicates the magnitude of the similarity between pairs of observations measured with respect to the outcome. This matrix is transformed into a complex network on which a community detection algorithm is applied to search for underlying community structures which is a partition of the instances into highly homogeneous clusters to be evaluated in terms of the outcome. In this paper, we focus on the case the outcome variable to be used in NeSSC is numeric and propose an alternative selection criterion of the optimal partition based on a measure of overlapping between density curves as well as a penalization criterion which takes accounts for the number of clusters in a candidate partition. Next, we consider the performance of the proposed method for some artificial datasets and for 20 different real datasets and compare NeSSC with the other three popular methods of semi-supervised clustering with a numeric outcome. Results show that NeSSC with the overlapping criterion works particularly well when a reduced number of clusters are scattered localized.

Identification of hydrologically homogenous watersheds and climate-vegetation dynamics in the Blue Nile Basin of Ethiopia

Abstract Identification of hydrologically homogenous watersheds in the Upper Blue Nile Basin of Ethiopia is challenging due to the large number of watersheds and the lack of consistent and reliable data. Traditional methods, such as expert-based classification, are time-consuming, subjective, and often not reproducible. Therefore, this study aims to identify homogenous gauged watersheds using hydrometeorological and remote sensing data. In this study 76 watersheds were delineated from a 30-m digital elevation model (SRTM-DEM). Twelve watershed characteristics were selected to aid the classification process. Three homogenous climate regions were identified using rainfall data from 42 stations, and for each homogeneous climate region, gauged watersheds were identified. Principal component analysis (PCA) and K-means clustering were used for classification. The PCA reduced 12 watershed characteristics into three principal components using a threshold of 80% accounted variance and eigenvalues greater than one. K-means clustering classified the 76 watersheds into nine homogenous clusters. In the classified regions, vegetation dynamics within three decades have also been analyzed. This helped identify trends in vegetation cover and its spatial and temporal dynamics. The results of the investigation will potentially be used for runoff prediction of ungauged watersheds and for water resource management models in the future.

Integrating joint latent class mixed models and Bayesian network for uncovering clinical subgroups of COVID-19 patients

When modelling the dynamics of biomarkers in biomedical studies, it is essential to identify homogeneous clusters of patients and analyse them from a precision medicine perspective. This need has emerged as crucial and urgent during the COVID-19 pandemic: early understanding of symptoms and patient heterogeneity has significant implications for prevention, early diagnosis, effective management, and treatment. Additionally, biomarker progression may be associated with clinically relevant time-toevent data. Therefore, statistical models are necessary to gain insight into complex disease mechanisms by properly accounting for unobservable heterogeneity in patients while jointly modelling longitudinal and time-to-event data. In this study, we leverage the key features of Latent Class modelling and Bayesian Network approaches and propose a unified framework to (a) uncover homogeneous subgroups of patients concerning their longitudinal and survival data and (b) describe patient subgroups within a multivariate framework.

Anti-correlated feature selection prevents false discovery of subpopulations in scRNAseq

While sub-clustering cell-populations has become popular in single cell-omics, negative controls for this process are lacking. Popular feature-selection/clustering algorithms fail the null-dataset problem, allowing erroneous subdivisions of homogenous clusters until nearly each cell is called its own cluster. Using real and synthetic datasets, we find that anti-correlated gene selection reduces or eliminates erroneous subdivisions, increases marker-gene selection efficacy, and efficiently scales to millions of cells.

Clustering High-Dimensional Noisy Categorical Data

Clustering is a widely used unsupervised learning technique that groups data into homogeneous clusters. However, when dealing with real-world data that contain categorical values, existing algorithms can be computationally costly in high dimensions and can struggle with noisy data that has missing values. Furthermore, except for one algorithm, no others provide theoretical guarantees of clustering accuracy. In this article, we propose a general categorical data encoding method and a computationally efficient spectral-based algorithm to cluster high-dimensional noisy categorical data (nominal or ordinal). Under a statistical model for data on m attributes from n subjects in r clusters with missing probability ϵ, we show that our algorithm exactly recovers the true clusters with high probability when m n ( 1 − ϵ ) ≥ C M r 2 log 3 M , with M = max ( n , m ) and a fixed constant C. In addition, we show that m n ( 1 − ϵ ) 2 ≥ r δ / 2 with 0 < δ < 1 is necessary for any algorithm to succeed with probability at least ( 1 + δ ) / 2 . In cases where m = n and r are fixed, the sufficient condition matches with the necessary condition up to a polylog ( n ) factor. In numerical studies our algorithm outperforms several existing algorithms in both clustering accuracy and computational efficiency. Supplementary materials for this article are available online.

Regulation of multiple signaling pathways promotes the consistent expansion of human pancreatic progenitors in defined conditions

The unlimited expansion of human progenitor cells in vitro could unlock many prospects for regenerative medicine. However, it remains an important challenge as it requires the decoupling of the mechanisms supporting progenitor self-renewal and expansion from those mechanisms promoting their differentiation. This study focuses on the expansion of human pluripotent stem (hPS) cell-derived pancreatic progenitors (PP) to advance novel therapies for diabetes. We obtained mechanistic insights into PP expansion requirements and identified conditions for the robust and unlimited expansion of hPS cell-derived PP cells under GMP-compliant conditions through a hypothesis-driven iterative approach. We show that the combined stimulation of specific mitogenic pathways, suppression of retinoic acid signaling, and inhibition of selected branches of the TGFβ and Wnt signaling pathways are necessary for the effective decoupling of PP proliferation from differentiation. This enabled the reproducible, 2000-fold, over 10 passages and 40–45 d, expansion of PDX1+/SOX9+/NKX6-1+ PP cells. Transcriptome analyses confirmed the stabilization of PP identity and the effective suppression of differentiation. Using these conditions, PDX1+/SOX9+/NKX6-1+ PP cells, derived from different, both XY and XX, hPS cell lines, were enriched to nearly 90% homogeneity and expanded with very similar kinetics and efficiency. Furthermore, non-expanded and expanded PP cells, from different hPS cell lines, were differentiated in microwells into homogeneous islet-like clusters (SC-islets) with very similar efficiency. These clusters contained abundant β-cells of comparable functionality as assessed by glucose-stimulated insulin secretion assays. These findings established the signaling requirements to decouple PP proliferation from differentiation and allowed the consistent expansion of hPS cell-derived PP cells. They will enable the establishment of large banks of GMP-produced PP cells derived from diverse hPS cell lines. This approach will streamline SC-islet production for further development of the differentiation process, diabetes research, personalized medicine, and cell therapies.

Regulation of multiple signaling pathways promotes the consistent expansion of human pancreatic progenitors in defined conditions.

The unlimited expansion of human progenitor cells in vitro could unlock many prospects for regenerative medicine. However, it remains an important challenge as it requires the decoupling of the mechanisms supporting progenitor self-renewal and expansion from those mechanisms promoting their differentiation. This study focuses on the expansion of human pluripotent stem (hPS) cell-derived pancreatic progenitors (PP) to advance novel therapies for diabetes. We obtained mechanistic insights into PP expansion requirements and identified conditions for the robust and unlimited expansion of hPS cell-derived PP cells under GMP-compliant conditions through a hypothesis-driven iterative approach. We show that the combined stimulation of specific mitogenic pathways, suppression of retinoic acid signaling, and inhibition of selected branches of the TGFβ and Wnt signaling pathways are necessary for the effective decoupling of PP proliferation from differentiation. This enabled the reproducible, 2000-fold, over 10 passages and 40-45 d, expansion of PDX1+/SOX9+/NKX6-1+ PP cells. Transcriptome analyses confirmed the stabilization of PP identity and the effective suppression of differentiation. Using these conditions, PDX1+/SOX9+/NKX6-1+ PP cells, derived from different, both XY and XX, hPS cell lines, were enriched to nearly 90% homogeneity and expanded with very similar kinetics and efficiency. Furthermore, non-expanded and expanded PP cells, from different hPS cell lines, were differentiated in microwells into homogeneous islet-like clusters (SC-islets) with very similar efficiency. These clusters contained abundant β-cells of comparable functionality as assessed by glucose-stimulated insulin secretion assays. These findings established the signaling requirements to decouple PP proliferation from differentiation and allowed the consistent expansion of hPS cell-derived PP cells. They will enable the establishment of large banks of GMP-produced PP cells derived from diverse hPS cell lines. This approach will streamline SC-islet production for further development of the differentiation process, diabetes research, personalized medicine, and cell therapies.

Understanding key contributing factors on the severity of traffic violations by elderly drivers: a hybrid approach of latent class analysis and XGBoost based SHAP

Traffic violation is one of the leading causes of traffic crashes. In the context of global aging, it is important to study traffic violations by elderly drivers for improving traffic safety in preparation for a worldwide aging population. In this study, a hybrid approach of Latent Class Analysis (LCA) and XGBoost based SHAP is proposed to identify hidden clusters and to understand the key contributing factors on the severity of traffic violations by elderly drivers, based on the police-reported traffic violation dataset of Beijing (China). First, LCA is applied to segment the dataset into several latent homogeneous clusters, then XGBoost based SHAP is established on each cluster to identify feature contributions and the interaction effects of the key contributing factors on the severity of traffic violations by elderly drivers. Two comparison groups were set up to analyze factors, which are responsible for the different severities of traffic violations. The results show that elderly drivers can be classified into four groups by age, urban or not, license, and season; factors such as less annual number of traffic violations, national & provincial highway, night and winter are key contributing factors for higher severity of traffic violations, which are consistent with common cognition; key contributing factors for all clusters are similar but not identical, for example, more annual number of traffic violations contribute to more severe violation for all clusters except for Cluster 2; some factors which are not key contributing factors may affect the severity of traffic violations when they are combined with other factors, for example, the combination of lower annual number of traffic violations and county & township highway contributes to more severe violation for Cluster 1. These findings can help government to formulate targeted countermeasures to decrease the severity of traffic violations by specific elderly groups and improve road service for the driving population.

Resampling estimation of discrete choice models

In the context of discrete choice modeling, the extraction of potential behavioral insights from large datasets is often limited by the poor scalability of maximum likelihood estimation. This paper proposes a simple and fast dataset-reduction method that is specifically designed to preserve the richness of observations originally present in a dataset, while reducing the computational complexity of the estimation process. Our approach, called LSH-DR, leverages locality-sensitive hashing to create homogeneous clusters, from which representative observations are then sampled and weighted. We demonstrate the efficacy of our approach by applying it on a real-world mode choice dataset: the obtained results show that the samples generated by LSH-DR allow for substantial savings in estimation time while preserving estimation efficiency at little cost.

Regulating solute partitioning via submerged friction stirring technology for improving corrosion resistance in an Al[sbnd]Mg alloy

Recently, compared to friction stirring processing (FSP) conducted in air, the potential of submerged FSP (SFS) for enhancing mechanical properties by immersing the workpiece in water has garnered significant attention. However, this enhancement was achieved at the expense of the corrosion resistance of alloys. In this study, the innovative use of a 10 wt% CaCl2 solution as a medium (SFS(CaCl2)) was expected to improve the corrosion resistance. Electrochemical tests conducted in 3.5 wt% NaCl solution showed that SFS(CaCl2) treatment improved the corrosion resistance of alloy compared to FSP treatment, including its contribution to pitting resistance. By summarizing the characteristics of microstructure, the mechanism of excellent corrosion resistance for SFS(CaCl2) sample was explained in detail. Nanoscale microstructure played a decisive role. By atomic-resolution TEM, the SFS(CaCl2) samples uniquely formed homogeneous clusters (1– 2 nm). The high-density clusters reduced the size of the solute depletion zone, and homogenized the local strength of the microstructure.

Big data clustering through fusion of FCM, optimized encoder-decoder CNN, and BiLSTM

Clustering Big Data, as a fundamental component in the processing and analysis of massive datasets, holds crucial importance in addressing complex challenges inherent in handling extensive data sets. Falling within the realm of unsupervised learning methods, the primary objective of clustering is to efficiently organize substantial datasets into homogeneous clusters without relying on pre-existing labels. Our innovative approach seeks to optimize this process by synergistically combining three techniques: the fuzzy C-Means (FCM) methodology, the optimized encoder–decoder CNN model, and the bidirectional recurrent neural network (BiLSTM). This synergy represents a strategic convergence between supervised and unsupervised paradigms. The introduction of BiLSTM is of significant importance, leveraging its capability to sequentially process data from both sides using LSTM cells. This bidirectional approach enhances the understanding of data sequences, a crucial feature in the demanding context of Big Data clustering. Simultaneously, FCM benefits from substantial improvement through the introduction of a function that calculates the separation between the cluster center and the instance, thereby reinforcing the precision of clustering. To optimize performance and reduce computation time, our methodology advocates for the use of the Optimized Encoder–Decoder CNN model. This refined architecture promotes more efficient extraction of data features, thereby enhancing the intrinsic quality of clustering. The rigorous evaluation of our approach revolves around specific data sources, namely fashion MNIST. Performance criteria such as accuracy, adjusted rand index (ARI), and normalized mutual information (NMI) convincingly attest to the remarkable capability of our methodology. In comparative analyses, our approach significantly outperforms existing models, demonstrating its effectiveness and relevance in the complex domain of Big Data clustering.

Brain texture as a marker of transdiagnostic clinical profiles in patients with recent-onset psychosis and depression

The inter-relationships of voxels can be captured by the radiomics texture features across multiple spatial scales. Prediction models of brain texture changes captured by the contrast texture feature in recent-onset psychosis (ROP) and recent-onset depression (ROD) have recently been proposed, although the validation of these models transdiagnostically at the individual level and the investigation of the variability in clinical profiles are lacking. Established prevention and treatment approaches focus on specific diagnoses and do not address the heterogeneity and manifold potential outcomes of patients. Here we aimed to investigate the cross-sectional utility of brain texture changes for (1) identification of the psychopathological state (ROP and ROD) and (2) the association of individualized brain texture maps with clinical symptom severity and outcome profiles. We developed transdiagnostic models based on structural magnetic resonance imaging data for 116 patients with ROD, 122 patients with ROP and 197 healthy control participants from the PRONIA (Personalized pROgNostic tools for early psychosIs mAnagement) study by applying a set of tools and frameworks to explain the classification decisions of the deep-learning algorithm (named explainable artificial intelligence) and clustering analysis. We investigated the contrast texture feature as the key feature for the identification of a general psychopathological state. The discrimination power of the trained prediction model was >72% and was validated in a second independent age- and sex-matched sample of 137 ROP, 94 ROD and 159 healthy control participants. Clustering analysis was implemented to map the changes in texture brain produced from an explainable artificial intelligence algorithm, in a group fashion. The explained individualized brain contrast map grouped into eight homogeneous clusters. In the clinical group, we investigated the association between the explained brain contrast texture map and clinical symptom severity as well as outcome profiles. Different patterns in the explained brain contrast texture map showed unique associations of brain alterations with clinical symptom severity and clinical outcomes, that is, age, positive, negative and depressive symptoms, and functionality. In some clusters, the mean explained brain contrast texture map values and/or brain contrast texture voxels that contributed significantly to the classification decision predicted accurately the PANSS (positive and negative symptom scale) scores, functionality and change in functionality over time. In conclusion, we created homogeneous clusters which predict the clinical severity and outcome profile in ROP and ROD patients.

Heterogeneity in the projections and excitability of tyraminergic/octopaminergic neurons that innervate the Drosophila reproductive tract.

Aminergic nuclei in mammals are generally composed of relatively small numbers of cells with broad projection patterns. Despite the gross similarity of many individual neurons, recent transcriptomic, anatomic and behavioral studies suggest previously unsuspected diversity. Smaller clusters of aminergic neurons in the model organism Drosophila melanogaster provide an opportunity to explore the ramifications of neuronal diversity at the level of individual cells. A group of approximately 10 tyraminergic/octopaminergic neurons innervates the female reproductive tract in flies and has been proposed to regulate multiple activities required for fertility. The projection patterns of individual neurons within the cluster are not known and it remains unclear whether they are functionally heterogenous. Using a single cell labeling technique, we show that each region of the reproductive tract is innervated by a distinct subset of tyraminergic/octopaminergic cells. Optogenetic activation of one subset stimulates oviduct contractions, indicating that the cluster as a whole is not required for this activity, and underscoring the potential for functional diversity across individual cells. Using whole cell patch clamp, we show that two adjacent and morphologically similar cells are tonically inhibited, but each responds differently to injection of current or activation of the inhibitory GluCl receptor. GluCl appears to be expressed at relatively low levels in tyraminergic/octopaminergic neurons within the cluster, suggesting that it may regulate their excitability via indirect pathways. Together, our data indicate that specific tyraminergic/octopaminergic cells within a relatively homogenous cluster have heterogenous properties and provide a platform for further studies to determine the function of each cell.