Shared Nearest Neighbor Research Articles

Multiple Extended Target Tracking Algorithm Based on Spatio-Temporal Correlation

In the clutter environment, the measurement of a set of multiple extended targets, with an unknown number of targets, poses challenges in partitioning, and the computational cost is high. In particular, the multiple extended target tracking method, based on distance partition, has obvious potential estimation errors when the extended targets intersect. This paper proposes a partition algorithm, based on spatio-temporal correlation, which considers the correlation between adjacent moments of the extended target and uses this prior information to divide the measurement set into a survival target measurement set and a born target measurement set for the first time. Then, the survival target measurement set is clustered by the K-means++ algorithm, and the extended target tracking is transformed into point target tracking. The born target measurement undergoes preprocessing by the DBSCAN clustering algorithm, and then uses the directed graph with shared nearest neighbors (SNN) dividing the measurement set. The method proposed in this paper significantly reduces the number of partitions and the computational time. The effectiveness of the algorithm is demonstrated through experimental simulations.

Dynamical substructures of local metal-poor halo

ABSTRACT Based on 4098 very metal poor (VMP) stars with six-dimensional phase-space and chemical information from Gaia Data Release 3 (DR3) and Large sky Area Multi-Object fiber Spectroscopic Telescope DR9 as tracers, we apply an unsupervised machine learning algorithm, shared nearest neighbour, to identify stellar groups in the action–energy (J–E) space. We detect seven previously known mergers in local samples, including Helmi Stream, Gaia-Sausage/Enceladus (GSE), metal-weak thick disc (MWTD), Pontus, Wukong, Thamnos, and I’itoi + Sequoia + Arjuna. According to energy, we further divide GSE and Wukong into smaller parts to explore the orbital characteristics of individual fragments. Similarly, the division of Thamnos is based on action. It can be found that the apocentric distances of GSE parts of high and medium energy levels are located at $29.5\pm 3.6$ and $13.0\pm 2.7\ {\rm kpc}$, respectively, which suggests that GSE could account for breaks in the density profile of the Galactic halo at both ≈30 and 15–18 kpc. The VMP stars of MWTD move along prograde orbits with larger eccentricities than those of its more metal-rich stars, which indicates that the VMP part of MWTD may be formed by accreting with dwarf galaxies. Finally, we summarize all substructures discovered in our local VMP samples. Our results provide a reference for the formation and evolution of the inner halo of the Milky Way.

A Fault Detection and Isolation Method via Shared Nearest Neighbor for Circulating Fluidized Bed Boiler

Accurate and timely fault detection and isolation (FDI) improve the availability, safety, and reliability of target systems and enable cost-effective operations. In this study, a shared nearest neighbor (SNN)-based method is proposed to identify the fault variables of a circulating fluidized bed boiler. SNN is a derivative method of the k-nearest neighbor (kNN), which utilizes shared neighbor information. The distance information between these neighbors can be applied to FDI. In particular, the proposed method can effectively detect faults by weighing the distance values based on the number of neighbors they share, thereby readjusting the distance values based on the shared neighbors. Moreover, the data distribution is not constrained; therefore, it can be applied to various processes. Unlike principal component analysis and independent component analysis, which are widely used to identify fault variables, the main advantage of SNN is that it does not suffer from smearing effects, because it calculates the contributions from the original input space. The proposed method is applied to two case studies and to the failure case of a real circulating fluidized bed boiler to confirm its effectiveness. The results show that the proposed method can detect faults earlier (1 h 39 min 46 s) and identify fault variables more effectively than conventional methods.

Rapid trajectory clustering based on neighbor spatial analysis

• Position information and velocity direction are used for trajectory segmentation. • A new similarity matrix is construct for clustering. • The spatial R-tree index is used to improve the efficiency of neighborhood query. • The real Atlantic hurricane and elk datasets verify that the proposed algorithm. The existing trajectory clustering algorithms only use the position information in trajectory segmentation, which makes the selection of segment points unreliable. Meanwhile, the adopted distance metrics are originally designed to compare the whole trajectory, leads to the inaccuracy similarity of trajectory segments, and hence causes subsequent clustering risks. Moreover, the execution time of clustering significantly increases with the amount of data. To address these issues, the direction of velocity is introduced for improving the discrimination of segment points. Next, the shared nearest neighbor ( S N N ) similarity and Trajectory-Hausdorff distance are combined to construct the similarity matrix for overcoming the limitations of existing distance measures. Then, based on the R-tree index strategy, the neighbored trajectory segments are extracted and stored for fastening segment indexing. Finally, the Atlantic hurricane and elk datasets verify that the proposed algorithm can not only improve the clustering efficiency but also extract the trajectory model accurately.

Data-driven estimation of TBM performance in soft soils using density-based spatial clustering and random forest

This study proposed a hybrid approach that integrates supervised and unsupervised learning to estimate the tunnel boring machine (TBM) performance in soft soil under limited geological information. By combining the shared nearest neighbor (SNN) algorithm and the density-based spatial clustering of applications with the noise (DBSCAN) method, an unsupervised learning approach, SNN-DBSCAN method, is performed to extract useful knowledge from the TBM logged data. The supervised random forest (RF) method is further combined with the SNN-DBSCAN method to predict the key TBM performance indicator. A realistic mass rapid transit (MRT) project in Singapore is adopted to examine the efficiency of the proposed methodology. The results from this case study indicate that: (1) The proposed SNN-DBSCAN method is suitable to perform data mining tasks on TBM logged data as the clustering result has an average of 85.03% similarity with site observation; (2) The knowledge extracted from the proposed approach could assist on soil identification as well as operational parameters determination; (3) Compared to the conventional RF method, the proposed approach achieves a high prediction accuracy with the coefficient of determination (R2) increasing from 0.78 to 0.92.

Shared Nearest Neighbour in Text Mining for Classification Material in Online Learning Using Mobile Application

There are many resources for media learning in online learning that all of the teachers made many media which it made a problem if there have the same subject and material. This problem made online learning having a big database and many materials made useless because the material has the same purpose. The big problem in overload database is that online learning can't be accessed by everyone. This research to fix this problem developed an algorithm in Artificial Intelligence for the classification of material in online learning with the same subject and purpose so that teachers can use already media. This algorithm is text mining and Shared Nearest Neighbour (SSN) that is embedded in the mobile application to display the classification and the location of searching media in database online learning. The testing in this research applied in 142 media with 130 data training and 12 data testing is the result of testing is 94,7% of the accuracy of the algorithm and The average of validation is 73,33%.

Abstract P009: A shared nearest neighbors approach for integrated, multi-platform networks and its application to the exploration of multiomics data from early-stage non-small cell lung cancers

Abstract Background: The ImmunogenomiC prOfiling of Non-small cell lung cancer (NSCLC) Project (ICON) represents an ambitious undertaking to comprehensively characterize immuno-genomic diversity in NSCLC across diverse platforms. The depth and breadth of this cohort presented a unique opportunity to develop a specialized method for multi-platform data integration and exploration, which can be broadly applied to forthcoming large-scale patient profiling studies. Such a holistic approach can unlock insights for therapeutic targets, biomarkers, and treatment plans by providing a more complete view of phenomena driving disease pathogenesis and evolution. Purpose: We developed a novel shared nearest neighbors (SNN) approach to create an integrated network of ICON’s multi-platform data and identified collections of closely related measurements within the resulting network tied to noteworthy patient characteristics, including recurrence and oncogenotype. Methods: The ICON dataset is derived from tumor and normal lung tissue samples collected from 150 patients at time of resection as well as blood samples collected then and at intervals during the year following. Tissue samples underwent RNA-sequencing (RNA-seq), whole exome sequencing, T-cell receptor sequencing, multiplex immunofluorescence for immune cells, and reverse phase protein array profiling; flow cytometry for immune cells was performed on tissue and blood samples. From these data, the ICON data network was built using an integrative approach based on the SNN algorithm in which genes were linked on the basis of their shared top correlates in orthogonal datasets. Results: The ICON data network currently includes over 20,000 genes linked by over 500,000 connections derived from correlations between RNA-seq and orthogonal platforms. We captured established associations between cancer-related genes and examined these along with new ones in the network. To do so, we used the InfoMap algorithm to extract more interpretable sub-networks, termed modules, from the ICON data network. Single sample gene set enrichment scores for each module were used in multivariate analysis to highlight modules linked to clinical characteristics of interest. As an example, we found modules significantly tied to disease recurrence. The most notable of these was strongly associated with metabolic pathways, and other modules associated with platelets and ion channels were also identified. The metabolic pathway module is being explored as a prognostic biomarker, underscoring the opportunites enabled by mining the network. Conclusions: Through the framework developed, we identified modules in the ICON data network significantly associated with important patient characteristics like recurrence and oncogenotype. We are validating the gene sets identified as potential biomarkers and are developing an interactive application to facilitate further mining of the network. Taken together, our SNN network-building approach enables the integration and exploration of patient data from diverse platforms. Citation Format: Stephanie T. Schmidt, Neal Akhave, Alexandre Reuben, Tina Cascone, Jianhua Zhang, Jun Li, Junya Fujimoto, Lauren A. Byers, Beatriz Sanchez-Espiridion, Lixia Diao, Jing Wang, Lorenzo Federico, Marie-Andree Forget, Daniel J McGrail, Annikka Weissferdt, Shiaw-Yih Lin, Younghee Lee, Natalie Vokes, Carmen Behrens, Ignacio I. Wistuba, Andrew Futreal, Ara Vaporciyan, Boris Sepesi, John V. Heymach, Chantale Bernatchez, Cara Haymaker, Jianjun Zhang, Christopher A. Bristow, Timothy P. Heffernan, Marcelo V. Negrao, Don L. Gibbons. A shared nearest neighbors approach for integrated, multi-platform networks and its application to the exploration of multiomics data from early-stage non-small cell lung cancers [abstract]. In: Proceedings of the AACR-NCI-EORTC Virtual International Conference on Molecular Targets and Cancer Therapeutics; 2021 Oct 7-10. Philadelphia (PA): AACR; Mol Cancer Ther 2021;20(12 Suppl):Abstract nr P009.

An Adaptive Ensemble Data Driven Approach for Nonparametric Probabilistic Forecasting of Electricity Load

Probabilistic load forecasting that provides uncertainty information involved in load forecasting is crucial for various decision-making tasks in power systems. This paper proposes a novel adaptive ensemble data driven (AEDD) approach for nonparametric probabilistic forecasting of electricity load by mining the uncertainty distribution from the historical observations based on conditional historical dataset construction and adaptive weighted ensemble. The pertinent patterns similar to the forecasting condition are searched from the numerous historical observations. The similarity degree measurement method is established based on shared nearest neighbors. Moreover, the uncertainty degree of the predictive load is quantified via information entropy, and then the number of similar patterns is determined depending to its uncertainty degree. After obtaining the conditional historical dataset, an adaptive weighted ensemble method is proposed for estimating the uncertainty distribution more correctly, where the weight for each similar pattern is set based on its similarity degree with the predictive load. Comprehensive numerical studies based on realistic load data validate the superiority of the proposed AEDD method in both accuracy and computational efficiency.

Evidence for Radial Expansion at the Core of the Orion Complex with Gaia EDR3

Abstract We present a phase-space study of two stellar groups located at the core of the Orion Complex: Briceño-1 and Orion Belt Population-near (OBP-near). We identify the groups with the unsupervised clustering algorithm, Shared Nearest Neighbor (SNN), which previously identified 12 new stellar substructures in the Orion Complex. For each of the two groups, we derive the 3D space motions of individual stars using Gaia EDR3 proper motions supplemented by radial velocities from Gaia DR2, APOGEE-2, and GALAH DR3. We present evidence for radial expansion of the two groups from a common center. Unlike previous work, our study suggests that evidence of stellar group expansion is confined only to OBP-near and Briceño-1, whereas the rest of the groups in the complex show more complicated motions. Interestingly, the stars in the two groups lie at the center of a dust shell, as revealed via an extant 3D dust map. The exact mechanism that produces such coherent motions remains unclear, while the observed radial expansion and dust shell suggest that massive stellar feedback could have influenced the star formation history of these groups.

Peak Density Clustering Algorithm Combining Natural and Shared Nearest Neighbor

Peak Density Clustering Algorithm Combining Natural and Shared Nearest Neighbor

DTI-SNNFRA: Drug-target interaction prediction by shared nearest neighbors and fuzzy-rough approximation.

In-silico prediction of repurposable drugs is an effective drug discovery strategy that supplements de-nevo drug discovery from scratch. Reduced development time, less cost and absence of severe side effects are significant advantages of using drug repositioning. Most recent and most advanced artificial intelligence (AI) approaches have boosted drug repurposing in terms of throughput and accuracy enormously. However, with the growing number of drugs, targets and their massive interactions produce imbalanced data which may not be suitable as input to the classification model directly. Here, we have proposed DTI-SNNFRA, a framework for predicting drug-target interaction (DTI), based on shared nearest neighbour (SNN) and fuzzy-rough approximation (FRA). It uses sampling techniques to collectively reduce the vast search space covering the available drugs, targets and millions of interactions between them. DTI-SNNFRA operates in two stages: first, it uses SNN followed by a partitioning clustering for sampling the search space. Next, it computes the degree of fuzzy-rough approximations and proper degree threshold selection for the negative samples’ undersampling from all possible interaction pairs between drugs and targets obtained in the first stage. Finally, classification is performed using the positive and selected negative samples. We have evaluated the efficacy of DTI-SNNFRA using AUC (Area under ROC Curve), Geometric Mean, and F1 Score. The model performs exceptionally well with a high prediction score of 0.95 for ROC-AUC. The predicted drug-target interactions are validated through an existing drug-target database (Connectivity Map (Cmap)).

An Exhaustive Research on the Application of Intrusion Detection Technology in Computer Network Security in Sensor Networks

Intrusion detection is crucial in computer network security issues; therefore, this work is aimed at maximizing network security protection and its improvement by proposing various preventive techniques. Outlier detection and semisupervised clustering algorithms based on shared nearest neighbors are proposed in this work to address intrusion detection by converting it into a problem of mining outliers using the network behavior dataset. The algorithm uses shared nearest neighbors as similarity, judges whether it is an outlier according to the number of nearest neighbors of a data point, and performs semisupervised clustering on the dataset where outliers are deleted. In the process of semisupervised clustering, vast prior knowledge is added, and the dataset is clustered according to the principle of graph segmentation. The novelty of the proposed algorithm lies in outlier detection while effectively avoiding the dependence on parameters, thus eliminating the influence of outliers on clustering. This article uses real datasets: lypmphography and glass for simulation purposes. The simulation results show that the algorithm proposed in this paper can effectively detect outliers and has a good clustering effect. Furthermore, the experimentation reveals that the outlier detection‐based SCA‐SNN algorithm has the best practical effect on the dataset without outliers, clearly validating the clustering performance of the outlier detection‐based SCA‐SNN algorithm. Furthermore, compared to the other state‐of‐the‐art anomaly detection method, it was revealed that the anomaly detection technology based on outlier mining does not require a training process. Thus, they overcome the current anomaly detection problems caused due to incomplete normal patterns in training samples.

Fast Searching Density Peak Clustering Algorithm Based on Shared Nearest Neighbor and Adaptive Clustering Center

The clustering analysis algorithm is used to reveal the internal relationships among the data without prior knowledge and to further gather some data with common attributes into a group. In order to solve the problem that the existing algorithms always need prior knowledge, we proposed a fast searching density peak clustering algorithm based on the shared nearest neighbor and adaptive clustering center (DPC-SNNACC) algorithm. It can automatically ascertain the number of knee points in the decision graph according to the characteristics of different datasets, and further determine the number of clustering centers without human intervention. First, an improved calculation method of local density based on the symmetric distance matrix was proposed. Then, the position of knee point was obtained by calculating the change in the difference between decision values. Finally, the experimental and comparative evaluation of several datasets from diverse domains established the viability of the DPC-SNNACC algorithm.

Single-Cell Analysis of the Classical Hodgkin Lymphoma Immune Environment Reveals a Clonally-Expanded CD8+ T Cell Population with a Cytotoxic Phenotype

Introduction: Classical Hodgkin lymphoma (cHL) is characterized by a robust and complex immune cell infiltrate and the rare presence of malignant Hodgkin-Reed-Sternberg (HRS) cells. At the genetic level, HRS cells recurrently acquire alterations that lead to defective antigen presentation (β2 microglobulin mutations) and mediate T cell dysfunction (PD-L1 copy gains/amplifications) in order to subvert host immune surveillance. The clinical relevance of PD-L1 protein over-expression in cHL is clear, as response rates to PD-1 blockade therapy are extremely high among patients with relapsed/refractory (r/r) disease. Despite its remarkable efficacy, the cells that mediate response to anti-PD-1 therapy in cHL remain undefined. Recent analyses have highlighted a possible role for CD4+ T cells in mediating the clinical activity of anti-PD-1 therapy in cHL. CD4+ T cells significantly outnumber CD8+ T cells in cHL lesions and are more frequently juxtaposed to HRS cells in situ. Furthermore, HLA class II expression on HRS cells predicted higher complete response rates to PD-1 blockade therapy in r/r cHL patients. However, a candidate T cell population capable of specific reactivity to antigens expressed by HRS cells has yet to be identified. This information is critical as such T cells might be functionally reinvigorated to mediate HRS cell elimination following PD-1 blockade therapy. In order to address this key knowledge gap, we analyzed data at single cell (sc) resolution using paired RNA and T cell receptor (TCR) sequencing in 9 diagnostic cHL and 5 reactive lymph node (RLN) specimens. Methods: Sequencing was performed using the 10x Genomics Chromium Single Cell 5' Gene Expression and V(D)J workflows. B-cell depletion of each sample was achieved using CD19 microbeads and negative selection to enrich T cell populations. Reads were analyzed and aligned with CellRanger (v3.1.0) and Seurat (v3.2.0) was used to conduct clustering by a shared nearest neighbor (SNN) graph on scRNA data. TCR sequencing data was integrated using scRepertoire (v1.0.0). Results: A detailed map of the immune cell states in cHL was created using scRNA-seq (10X) data on 79,085 cells from 9 cHL (52,602 cells) and 5 RLN samples (26,484 cells) expressing a total of 21,421 genes (mean 5649 cells/sample; mean 2849 mRNA reads/cell). Dimensionality reduction and unsupervised graph-based clustering revealed 21 distinct cell type and activation state clusters, including T cells, NK cells, macrophages, and dendritic cells (Fig 1A-B). A cluster identifying HRS cells was not observed, consistent with a recently published report. Ten T cell clusters were identified (47,573 cells), including naive- and memory-like T cells, effector/cytotoxic CD8+ T cells, regulatory T cells, and T follicular helper cells. Unexpectedly, a putative exhausted T cell cluster was not clearly observed. The relative contributions of cHL and RLNs cases to these clusters are shown in Fig 1C. Paired TCR sequencing was available for 23,943 cells. Overall TCR diversity was lower among cHL samples compared to RLN specimens (Fig 1D). In cHL samples, modest clonal expansion within regulatory T cell and memory CD4+ T cell clusters was observed, but the most striking clonal expansion occurred among cells assigned to effector/cytotoxic CD8+ T cell clusters - a finding not observed in most RLN specimens (Fig 1E). Clonally-expanded effector/cytotoxic CD8+ T cells displayed high expression of granzymes (GZMA, GZMH, GZMK), cytokines (TNF, IFNG) and chemokines (CCL4/CCL5), and modest expression of exhaustion markers (PDCD1, ENTPD1, HAVCR2, ITGAE), contrasting with data from single-cell analyses of solid tumors. Clonal expansion of effector/cytotoxic CD8+ T cells was particularly robust in EBV-positive cHLs, likely due to recognition of viral-derived epitopes displayed on HRS cells (Fig 1F). Phenotypic and functional validation of key immune cell clusters in cHL specimens using spectral cytometry is underway and will be reported at the meeting. Conclusions: For the first time, our data have unveiled the nature of the T cell repertoire in cHL at single cell resolution. Our results reveal a recurrent pattern of clonal expansion within effector CD8+ cells, which may be the HRS antigen-specific T cells that mediate response to PD-1 blockade. This hypothesis requires confirmation through similar analyses of pre- and on-treatment biopsies of cHL patients receiving anti-PD-1 therapy. Disclosures Godfrey: Gilead: Research Funding; Merck: Research Funding; Verastem: Research Funding. Venkataraman:EUSA Pharma: Speakers Bureau. Smith:Janssen: Consultancy; BMS: Consultancy; TG Therapeutics: Consultancy, Research Funding; Genentech/Roche: Consultancy, Other: Support of parent study and funding of editorial support, Research Funding; Karyopharm: Consultancy, Research Funding; FortySeven: Research Funding; Pharmacyclics: Research Funding; Acerta: Research Funding; Celgene: Consultancy, Research Funding. Kline:Kite/Gilead: Speakers Bureau; Seattle Genetics: Membership on an entity's Board of Directors or advisory committees; Merck: Research Funding; Karyopharm: Membership on an entity's Board of Directors or advisory committees; Verastem: Membership on an entity's Board of Directors or advisory committees, Research Funding.

Discovery of new stellar groups in the Orion complex

We test the ability of two unsupervised machine learning algorithms,EnLinkand Shared Nearest Neighbor (SNN), to identify stellar groupings in the Orion star-forming complex as an application to the 5D astrometric data fromGaiaDR2. The algorithms represent two distinct approaches to limiting user bias when selecting parameter values and evaluating the relative weights among astrometric parameters.EnLinkadopts a locally adaptive distance metric and eliminates the need for parameter tuning through automation. The original SNN relies only on human input for parameter tuning so we modified SNN to run in two stages. We first ran the original SNN 7000 times, each with a randomly generated sample according to within-source co-variance matrices provided inGaiaDR2 and random parameter values within reasonable ranges. During the second stage, we modified SNN to identify the most repeating stellar groups from the 25 798 we obtained in the first stage. We recovered 22 spatially and kinematically coherent groups in the Orion complex, 12 of which were previously unknown. The groups show a wide distribution of distances extending as far as about 150 pc in front of the star-forming Orion molecular clouds, to about 50 pc beyond them, where we, unexpectedly, find several groups. Our results reveal the wealth of sub-structure in the OB association, within and beyond the classical Blaauw Orion OBI sub-groups. A full characterization of the new groups is essential as it offers the potential to unveil how star formation proceeds globally in large complexes such as Orion.

Clustering by Detecting Density Peaks and Assigning Points by Similarity-First Search Based on Weighted K-Nearest Neighbors Graph

This paper presents an improved clustering algorithm for categorizing data with arbitrary shapes. Most of the conventional clustering approaches work only with round-shaped clusters. This task can be accomplished by quickly searching and finding clustering methods for density peaks (DPC), but in some cases, it is limited by density peaks and allocation strategy. To overcome these limitations, two improvements are proposed in this paper. To describe the clustering center more comprehensively, the definitions of local density and relative distance are fused with multiple distances, including K-nearest neighbors (KNN) and shared-nearest neighbors (SNN). A similarity-first search algorithm is designed to search the most matching cluster centers for noncenter points in a weighted KNN graph. Extensive comparison with several existing DPC methods, e.g., traditional DPC algorithm, density-based spatial clustering of applications with noise (DBSCAN), affinity propagation (AP), FKNN-DPC, and K-means methods, has been carried out. Experiments based on synthetic data and real data show that the proposed clustering algorithm can outperform DPC, DBSCAN, AP, and K-means in terms of the clustering accuracy (ACC), the adjusted mutual information (AMI), and the adjusted Rand index (ARI).

Application of an Efficient Graph-Based Partitioning Algorithm for Extended Target Tracking Using GM-PHD Filter

One of the main tasks involving the extended target tracking is how to partition the measurement set accurately and efficiently. In this article, an efficient graph-based partitioning algorithm is introduced for extended target tracking. To reduce the computational load and the interference of clutter on the measurement set partition, a measurement set preprocessing method based on density-based clustering algorithm is presented. An intuitive directed $k$ -nearest neighbor ( $k$ NN) graph model based on graph theory is established to represent the relationship between different measurements in the measurement set that needs to be segmented. In the framework of directed $k$ NN graph, a novel similarity metric based on shared nearest neighbor (SNN) is used, and a pairwise similarity that integrates the number of elements in the set of SNN and the closeness of data points is constructed. The spectral clustering algorithm is used to process the multiway cut in the directed $k$ NN graph. The graph-based partitioning algorithm is applied to the extended target Gaussian mixture probability hypothesis density filter. Simulation results illustrate the advantages of our proposed graph-based partitioning algorithm in performance and computational efficiency.

Single-Cell Clustering Based on Shared Nearest Neighbor and Graph Partitioning.

Clustering of single-cell RNA sequencing (scRNA-seq) data enables discovering cell subtypes, which is helpful for understanding and analyzing the processes of diseases. Determining the weight of edges is an essential component in graph-based clustering methods. While several graph-based clustering algorithms for scRNA-seq data have been proposed, they are generally based on k-nearest neighbor (KNN) and shared nearest neighbor (SNN) without considering the structure information of graph. Here, to improve the clustering accuracy, we present a novel method for single-cell clustering, called structural shared nearest neighbor-Louvain (SSNN-Louvain), which integrates the structure information of graph and module detection. In SSNN-Louvain, based on the distance between a node and its shared nearest neighbors, the weight of edge is defined by introducing the ratio of the number of the shared nearest neighbors to that of nearest neighbors, thus integrating structure information of the graph. Then, a modified Louvain community detection algorithm is proposed and applied to identify modules in the graph. Essentially, each community represents a subtype of cells. It is worth mentioning that our proposed method integrates the advantages of both SNN graph and community detection without the need for tuning any additional parameter other than the number of neighbors. To test the performance of SSNN-Louvain, we compare it to five existing methods on 16 real datasets, including nonnegative matrix factorization, single-cell interpretation via multi-kernel learning, SNN-Cliq, Seurat and PhenoGraph. The experimental results show that our approach achieves the best average performance in these datasets.

High-Throughput BCR Sequencing and Single-Cell Transcriptomics Reveal Distinct Transcriptional Profiles Associated with Subclonal Evolution of Follicular Lymphoma

Objectives: Follicular lymphoma (FL) typically originates from premalignant mature B cells that carry the founder t(14;18) BCL2 translocation. Mutations in epigenetic modifiers and acquisition of N-glycosylation sites in CDR regions of the B-cell receptor (BCR) are recurrent secondary events in FL pathogenesis. Despite these oncogenic drivers, FL can remain indolent and clinically stable for years. The molecular events driving subclonal evolution into symptomatic progression and eventual transformation to aggressive lymphoma are insufficiently understood. FL cells are frozen in their B-cell development at the germinal center stage and undergo continuous somatic hypermutation mediated by expression of activation-induced deaminase (AID). We aim to identify crucial drivers of subclonal FL evolution by high-throughput mapping at single-cell resolution. Methods: Viable FL cells were isolated and cryopreserved from 23 histologically or immunocytologically confirmed FL samples from 13 patients with informed consent. Full-length VDJ/VJ transcripts were isolated by unbiased template-switching ARTISAN PCR and massive parallel NGS sequencing on the PacBio platform. The clonal primordial FL BCR (pBCR) was reconstructed from unmutated IGV/IGJ sequences with the CDR3 of the least mutated BCR. Since the IgTree program was unable to process the obtained numbers of BCR sequences, we developed the WILLOW algorithm for analysis of BCR evolution based on the principle of maximum parsimony and on distance from the pBCR. Intraclonal BCR variability was quantified by Shannon's diversity index. 5' single cell transcriptomics and VDJ/VJ sequencing was performed on 2 pools of highly purified FL cells from 5 lymph node biopsies on the 10x Genomics platform. Data were deconvoluted based on expressed variants by the Single Cell Sample Matcher (SCSM) algorithm. Clustering based on gene expression profiles was performed by shared nearest neighbour (SNN) modularity optimization within the R Seurat package. Genes whose expression differed significantly (adjusted p<0.05) between clusters were assigned to gene ontology terms. Results: ARTISAN PCR/PacBio NGS yielded a median of 743 full-length VDJ and VJ sequences (range 62-12782) per BCR chain with expected high intraclonal diversity (median 200 subclones, range 15-3301). WILLOW revealed dominant FL subclones with a subclonal hierarchy wherein multiple routes converged to offspring nodes with identical additional mutations rather than tree-like branching (Figure). In serial samples of 4 patients, lymph node biopsies had only marginally higher subclonal diversity than blood or bone marrow samples (p=0,055; Wilcoxon's matched-pairs signed rank test). Overall BCR mutational burden increased over time in sequential biopsies. Two cases of histological FL transformation were dominated by a single subclone (65% and 80% of all VDJ/VJ sequences, respectively) that was rare in the preceding FL BCR network (0.2% and 1.8%). Pooled transcriptomics data from 6050-6500 cells were assigned to individual samples by SCSM and revealed up to seven transcriptional clusters per FL. In 9 of 10 FL, genes assigned to immune function strongly contributed to separation into one or more clusters. Single cell VDJ/VJ sequencing yielded combined heavy and light chain BCR sequences for a median of 502 FL cells per biopsy (range 22 - 1919) that permitted mapping of subclonal evolution by WILLOW based on complete BCR information. Transcriptome clusters were not distributed evenly throughout the WILLOW FL BCR networks but rather statistically associated with distinct major FL subclones. Vice versa, major FL subclones within the same biopsy were distinguished by particular gene expression profiles. Conclusions: WILLOW facilitates mapping of subclonal FL evolution based on high-throughput BCR sequencing. FL evolution proceeds in networks rather than tree-like branching, whereby acquisition of certain combinations of several BCR mutations can occur in parallel in different trajectories. Transcriptomic profiling of single FL cells identifies distinct clusters within a single biopsy. Mapping of these clusters to the FL cell position in the subclonal FL evolutionary network identifies putative mechanisms that are associated with subclonal progression. These mechanisms involve physiological B-cell signalling pathways. Figure Disclosures No relevant conflicts of interest to declare.

Longitudinal Tracking of Acute Myeloid Leukemia Stem Cells in Xenografts and Patients By microRNA Reporters and Single Cell RNA Sequencing

Current understanding of acute myeloid leukemia (AML) assumes a developmental hierarchy, in which a minor fraction of primitive and quiescent leukemia stem cells (LSC) sustain clonal propagation of disease. These therapy-resistant LSC may be the basis of relapse, as supported by data correlating the presence of LSC gene expression signatures at diagnosis with poor prognosis (Ng et al, Nature 2016). Novel approaches tracing LSC fates at single cell level before, during and after chemotherapy (CTX) are needed to confirm their biological relevance and derive new, LSC-focused diagnostic and therapeutic strategies. We and others have previously linked key LSC properties, such as quiescence and therapy resistance, to complex transcriptional regulation orchestrated by miR-126 (Lechman et al, Cancer Cell 2016). Furthermore, we provided proof of concept that LSCs could be prospectively isolated as miR-126(high) cells exploiting a lentiviral reporter vector capturing miR-126 bioactivity in live cells with single cell resolution. To extend these studies, we transduced primary blasts from n=3 AML patients (pts) carrying NPM1 and FLT3-ITD mutations with the miR-126 reporter, followed by xenografting (PDX). Blasts showed intra-tumor heterogeneity (ITH) in terms of miR-126 activity, with a minor fraction identified as miR-126(high). Limiting dilution secondary transplantation of FACS sorted miR-126(high) and -(low) blasts proved strong enrichment of repopulating activity within the miR-126(high) compartment in all 3 pts. On the contrary, no LSC enrichment could be verified in the CD34+CD38- fraction in 1 patient, suggesting that high miR-126 activity represents a more robust LSC identifier than commonly used surface markers. Next, we investigated the impact of daunorubicin and cytarabine CTX on miR-126-reporter+ blasts (n=3 AML) in PDX. Surprisingly, overall miR-126 activity diminished in post CTX residual AML compared to controls, compatible with a loss of blast quiescence. CTX accentuated ITH by uncovering a subset of blasts with very high levels of miR-126, distinct from the bulk population, which may correspond to residual quiescent LSC (Fig A). We then performed bulk RNA sequencing on miR-126(high) and miR-126(low) subsets from CTX and control PDX. While miR-126(low) blasts from both groups expressed markers of myeloid differentiation, miR-126(high) blasts were enriched for published hematopoietic stem cell hallmark signatures. Integrating differentially expressed genes between miR-126(high) and-(low) subsets at steady state and post CTX, we extrapolated a novel 8 gene signature associated with miR-126(high) blasts. Of note, patients from the AML TCGA PanCancer Atlas Cohort (n=161) harboring overexpression in one or more of these 8 genes had significantly decreased overall survival (10 vs 19 months, Logrank test p-value = 0.018). To further test whether our 8-gene miR-126(high) signature reveals ITH in patients, we performed single cell RNA sequencing (scRNAseq) of AML patient BM aspirates at diagnosis (n= 6). Blasts were identified based on the detection of mutated NPM1 transcripts in single cells. In 5 out of 7 patients expression of the miR-126 signature mapped to specific clusters of blasts identified by unsupervised shared nearest neighbor algorithm, confirming that it identifies ITH in patient samples. Interestingly, we detected miR-126 signature(high) blasts in pts with poor prognosis (n=4) and not in those with favorable outcome (n=2) (Fig B). To investigate ITH across longitudinal samples, we next performed scRNAseq of residual AML from a representative patient assessed early after CTX. In line with our PDX CTX model displaying increased miR-126 ITH, blasts on day14 of induction CTX segregated into 2 different clusters: cluster 1 containing LSC-like cells with miR-126(high) signature and similar transcriptional profile to the diagnosis counterpart; cluster 2, instead, was composed of actively cycling blasts. Residual blasts at day30, in addition to uniformly expressing the miR-126(high) signature, differed from diagnosis and day14 blasts by displaying cell cycle quiescence and induction of oxidative phosphorylation genes (Fig C). In summary, we have set up and applied PDX LSC modeling to clinically relevant patient samples to address AML intra-tumor heterogeneity and to pinpoint novel relevant transcriptomic features of LSC at diagnosis and after chemotherapy. Figure Disclosures Gentner: Genenta Science: Consultancy, Equity Ownership, Research Funding.