Dataset Information

Element-centric clustering comparison unifies overlaps and hierarchy.

ABSTRACT: Clustering is one of the most universal approaches for understanding complex data. A pivotal aspect of clustering analysis is quantitatively comparing clusterings; clustering comparison is the basis for many tasks such as clustering evaluation, consensus clustering, and tracking the temporal evolution of clusters. In particular, the extrinsic evaluation of clustering methods requires comparing the uncovered clusterings to planted clusterings or known metadata. Yet, as we demonstrate, existing clustering comparison measures have critical biases which undermine their usefulness, and no measure accommodates both overlapping and hierarchical clusterings. Here we unify the comparison of disjoint, overlapping, and hierarchically structured clusterings by proposing a new element-centric framework: elements are compared based on the relationships induced by the cluster structure, as opposed to the traditional cluster-centric philosophy. We demonstrate that, in contrast to standard clustering similarity measures, our framework does not suffer from critical biases and naturally provides unique insights into how the clusterings differ. We illustrate the strengths of our framework by revealing new insights into the organization of clusters in two applications: the improved classification of schizophrenia based on the overlapping and hierarchical community structure of fMRI brain networks, and the disentanglement of various social homophily factors in Facebook social networks. The universality of clustering suggests far-reaching impact of our framework throughout all areas of science.

SUBMITTER: Gates AJ

PROVIDER: S-EPMC6561975 | biostudies-literature | 2019 Jun

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Element-centric clustering comparison unifies overlaps and hierarchy.

Gates Alexander J AJ Wood Ian B IB Hetrick William P WP Ahn Yong-Yeol YY

Scientific reports 20190612 1

Clustering is one of the most universal approaches for understanding complex data. A pivotal aspect of clustering analysis is quantitatively comparing clusterings; clustering comparison is the basis for many tasks such as clustering evaluation, consensus clustering, and tracking the temporal evolution of clusters. In particular, the extrinsic evaluation of clustering methods requires comparing the uncovered clusterings to planted clusterings or known metadata. Yet, as we demonstrate, existing cl ...[more]

PMID: 31189888

Similar Datasets

Project description:ImportanceThe current program-centric algorithm for the National Resident Matching Program (NRMP) primarily uses the program's ranking of students to determine a match. Concerns that the existing algorithm favors programs over students, recent findings that the program's ranking of applicants is not associated with resident performance, and disruptions of existing screening methods and metrics have prompted reevaluation of the current algorithm relative to a student-centric algorithm, in which student ranking of programs is primary and program ranking of students is secondary.ObjectiveTo compare program-centric and student-centric algorithms for the NRMP participants.Design, setting, and participantsThis cross-sectional study used randomized computer-generated data reflecting the NRMP match for 2018, 2019, and 2020, capturing more than 50 000 students and more than 4000 programs in 23 specialties, to compare the 2 algorithms.ExposuresThe same simulated students, programs, and rankings were exposed to the 2 algorithms, running 2300 simulations in the overall analysis and 1000 simulations in each of 23 specialties.Main outcomes and measuresThe percentage of students who did and did not match, the percentage of students who matched to their top-ranked and top-5-ranked programs, and the program's rank of the last student matched per position were examined.ResultsThe 2 algorithms were not different in percentage of students matched overall (eg, for 2020, program-centric: 59% [95% CI, 57%-61%]; student-centric: 58% [95% CI, 56%-60%]; P = .73). The student-centric algorithm, relative to the program-centric algorithm, matched a significantly higher percentage of students to their first-ranked program (eg, for 2020, 50% [95% CI, 48%-52%] vs 14% [95% CI, 13%-15%]; P < .001) and to their top-5-ranked programs (eg, for 2020, 60% [95% CI, 58%-62%] vs 46% [95% CI, 44%-48%]; P < .001). However, the last position was filled with students who had lower program rankings in the student-centric algorithm vs the program-centric algorithm (2 [95% CI, 1-2] vs 8 [95% CI, 6-10]; P < .001).Conclusions and relevanceIn this study, the 2 algorithms were not different in the percentage of students matched overall. However, the student-centric algorithm matched a significantly higher percentage of students to their preferred programs. The program-centric algorithm was associated with a lower program's last matched student rank. Further research is needed on the algorithms' associations with cost and time demands in the match, postmatch resident and program performance, and fit with a changing environment.

Project description:PurposeTo determine the effect of daily shifts based on rigid registration to intraprostatic markers on coverage of boost doses delivered to gross nodal disease for prostate cancer.Methods and materialsSeventy-five cone beam computed tomographies (CBCTs) from 15 patients treated with definitive radiation for clinically node-positive prostate cancer underwent fiducial-based and pelvic bony-based registration to the initial planning scans. Gross tumor volumes of nodal boost targets were contoured directly on each CBCT registration. The nodal displacement (3-dimensional translation from the node centroid on planning CT to node centroid on registered CBCT) and dose coverage (minimum dose [Dmin], mean dose [Dmean], dose delivered to 95% of the gross tumor volumes [D95]) were calculated for each registration on all nodal targets. All doses for each node were normalized to its intended prescription dose (dose covering 95% of a 3 mm planning target volume [PTV] expansion).ResultsForty-one gross nodal targets were analyzed. Most boosted nodes (80.5%, 33/41) were treated with conventional fractionation using volumetric-arc radiation therapy, and 19.5% (8/41) underwent stereotactic body radiation therapy (SBRT). Dmin, Dmean, and D95 were all significantly lower with fiducial-based registration compared with bony-based registration (P < .0001). Nodal displacement was significantly higher for fiducial-based registrations (P < .0001). The 3-dimensional translation between the fiducial-based and bony-based registrations (bony-to-fiducial vector) was the most significant predictor of nodal displacement (P < .0001). On fiducial-based registrations, a 3 to 5 mm gross nodal PTV margin is sufficient in most directions; however, superior and posterior margins of 8 to 9 mm are required as a result of asymmetrical prostatic motion.ConclusionsLarge and anisotropic PTV margins are likely needed to adequately dose gross nodal targets when patient setup is based on rigid registration to intraprostatic markers. Alternative approaches such as adaptive replanning may be required to overcome these limitations.

Dataset Information

Element-centric clustering comparison unifies overlaps and hierarchy.

Publications

Element-centric clustering comparison unifies overlaps and hierarchy.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets