Positions In Feature Space Research Articles

Abstract 139: cfDNA methylation profiling distinguishes lineage-specific hematologic malignancies

Abstract Introduction: Hematologic (heme) malignancies and their precursor conditions are highly prevalent. They are also diverse in biology, clinical presentation, and outcomes, underlining the importance of differentiating them. Previously, we demonstrated that a blood-based targeted methylation assay detected multiple cancer types across stages. Here, we examined test performance on various heme cancers, identifying specific methylation signatures. Methods: From the second substudy (training set) of the Circulating Cell-free Genome Atlas (CCGA) study (NCT02889978), we evaluated 325 participants from 17 different heme disease subtypes and 3,211 non-cancer controls enrolled without a cancer diagnosis. A cross-validated mutual information-based algorithm was used to identify features that discriminated heme subtypes. The resulting feature distribution was visualized using uniform manifold approximation and projection (UMAP) dimensionality reduction on held-out data. In cross validation with feature selection, we then trained a multinomial classifier to distinguish from among the major heme cancers and non-cancer and correlated the model's class probabilities to positions in feature space. Results: Dimensionality reduction and visualization of input features demonstrated that heme malignancies separated into five major clusters reflecting developmental lineages and disease ontogeny: myeloid, circulating lymphomas, hodgkin lymphomas, non-hodgkin lymphomas, and plasma cell neoplasm. The position of samples within each heme cluster correlated with the cancer signal strength. At 99.4% specificity [95% CI: 99.1, 99.7], heme cancer detection was 74.5% [69.4, 79.1] overall, 67.7% [41.1, 87.8] for myeloid, 77.9% [66.3, 86.9] for circulating lymphomas, 90.7% [73.2, 98.4] for hodgkin lymphomas, 68.6% [60.4, 76.1] for other non-hodgkin lymphomas, and 78.8% [67.0, 87.9] for plasma cell neoplasms. Of 18 non-cancer participants who were classified as having heme cancers, 4 were predicted as myeloid, 6 as circulating lymphoid, and 8 as other non-hodgkin lymphoid neoplasms (<1% false positive rate). Conclusion: Methylation features of cfDNA in patients with heme malignancies delineated five major clusters that reflected disease ontogeny and heme lineage. Lineage-specific signals followed a gradient suggestive of variation in disease-related methylation or tumor DNA shedding. These findings contribute to the understanding of biological signals that arise from various heme conditions. Since in general, most cfDNA arises from blood lineages, this knowledge will guide further efforts towards removing interfering biological signals from cfDNA-based cancer detection assays and achieving even more sensitive detection of multiple cancer types. Citation Format: Qinwen Liu, Rita Shaknovich, Xiaoji Chen, Zhao Dong, M. C. Maher, Samuel Gross, Alexander P. Fields, Jan Schellenberger, Kathryn N. Kurtzman, Eric T. Fung, Anne-Renee Hartman, Earl Hubbell, Arash Jamshidi, Alexander M. Aravanis, Oliver Venn. cfDNA methylation profiling distinguishes lineage-specific hematologic malignancies [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 139.

What Makes Popular Culture Popular? Product Features and Optimal Differentiation in Music

In this article, we propose a new explanation for why certain cultural products outperform their peers to achieve widespread success. We argue that products’ position in feature space significantly predicts their popular success. Using tools from computer science, we construct a novel dataset allowing us to examine whether the musical features of nearly 27,000 songs from Billboard’s Hot 100 charts predict their levels of success in this cultural market. We find that, in addition to artist familiarity, genre affiliation, and institutional support, a song’s perceived proximity to its peers influences its position on the charts. Contrary to the claim that all popular music sounds the same, we find that songs sounding too much like previous and contemporaneous productions—those that are highly typical—are less likely to succeed. Songs exhibiting some degree of optimal differentiation are more likely to rise to the top of the charts. These findings offer a new perspective on success in cultural markets by specifying how content organizes product competition and audience consumption behavior.

An Ecology of Market Categories

This paper proposes that social categorization is driven by an ecological dynamic that operates in two planes: feature space and label space. It develops a theoretical model that links positions in the two spaces. The first part of the theory predicts that an actor’s proximity in feature space to a labeled cluster increases its propensity to adopt the label. The second part predicts that the structure of label space affects this relationship: featurespace positions are more weakly related to labeling for lenient labels. An empirical analysis of software producers, based on their positions in a technical feature space (derived from portfolios of patents) and a space of market labels, supports these predictions. The results imply that one producer’s changed labeling can change the sets of feature values associated with a label, which then affects other producers in the domain. This coupled ecological dynamic is found even in a loosely governed system of self-categorization. Findings suggest that constraining categories become more constraining, while lenient categories become more lenient. When people judge whether an olive is a fruit or whether Pluto is a planet, their assessments do not depend on what the fruits and celestial bodies have to say about the matter. The situation differs sharply for categorization of humans and corporate actors. These kinds of agents can, and often do, dispute categorical assignments. Re-classification of one object can affect not only how it gets perceived but also the meaning of the category. We are struck by the possibility that this latter effect produces an ecological dynamic for social categorization. Classification of one entity, say a cultural product, affects the classification of others, potentially generating cascades of changes. This ecological dynamic operates in two planes: one involves feature values and the other involves profiles of label assignments. The two planes are connected through categorization when agents label objects. Sometimes labels are linked to feature values that are common to a set of objects. This is the first step to the label becoming infused with social meaning.1 The association between labels and feature values is straightforward in a static world. But social actors, unlike olives and other natural kinds, can change their characteristics. Moves in feature space by members of a category can make atypical Date: July 24, 2012; revised December 7, 2012. We appreciate the extremely helpful suggestions of Greta Hsu, Balazs Kovacs, Gael Le Mens, and Laszlo Polos. This study was supported by the Chicago-Booth Graduate School of Business, the Stanford Graduate School of Business, and the John Osterweis and Barbara Ravizzi Faculty Fellowship at the Stanford GSB. 1In our construction a feature is a dimension such as the form of authority in an organization. Positions in feature space are constructed from values of features, e.g., tradition, rational-legal, or charismatic forms of authority.

Toward an Ecology of Market Categories

This paper proposes that social categorization is driven by an ecological dynamic that operates in two planes: feature space and category space. It develops a theoretical model that links positions in feature space to label assignments in category space. The first part of the model predicts that movements in feature space affect label assignments in category space, with proximity to labeled clusters affecting label adoption. The second part predicts that the structure of category space affects this relationship. For lenient labels, positions in feature space are more weakly related to label adoption in category space. An empirical analysis of software producers, based on their positions in a feature space of patents, and a category space of market-label affiliations, supports these predictions. The results imply that social classification is characterized by this coupled ecological dynamic. Further, our findings have implications for the evolution of categories, suggesting that lenient categories will become more lenient, while constraining categories will become more constraining.

Self-Organizing Map with Generating and Moving Neurons in Visible Space

Kohonen’s Self-Organizing feature Map (SOM) is used to obtain topology-preserving mapping from high-dimensional feature space to visible space of two or fewer dimensions. The SOM algorithm uses a fixed structure of neurons in visible space and learns a dataset by updating reference points in feature space. The mapping result depends on mapping parameters fixed, which are the number and visible positions of neurons, and parameters of learning, which are the learning rate, total iteration, and the setting of neighboring radii. To obtain a satisfactory result, the user usually must try many combinations of parameters. It is wasteful, however, to set up every possible combination of parameters and to repeatedly run the algorithm from the beginning because the computation cost for learning is large, especially for a large-scale dataset. These problems arise due to the fixing of two types of mapping parameters, i.e., the number and visible positions of neurons. The high computation cost is mainly in the calculation of distances from each sample to all reference points. At the beginning of learning, reference points should be adjusted globally to preserve the topology well because they are initially set far from optimal positions in feature space, e.g. randomly. Such many reference points subdivides feature space into unnecessarily fine Voronoi regions. To avoid this computational waste, it is natural to start learning with a small number of neurons and increase the number of neurons during learning. We propose a new SOM method that varies the number and visible positions of neurons, and thus is applicable also to visible torus and sphere spaces. We apply our proposal to spherical visible space. We use central Voronoi tessellation to move visible positions for two reasons: to tessellate visible space evenly for easy visualization and to level the number of neighboring neurons and better preserve topology. We demonstrate the effect of generating neurons to reduce computation cost and of moving visible positions in visualization and topology preservation.

Analysis of pattern recognition by man using detection experiments.

This paper addresses the problem of analyzing biological pattern recognition systems. As no complete analysis is possible due to limited observability, the theoretical part of the paper examines some principles of construction for recognition systems. The relations between measurable and characteristic variables of these systems are described. The results of the study are: 1. Human recognition systems can always be described by a model consisting of an analyzer (FA) and a linear classifier. 2. The linearity of the classifier places no limits on the universal validity of the model. The principle of organization of such a system may be put into effect in many different ways. 3. The analyzer function FA determines the transformation of external patterns into their internal representations. For the experiments described in this paper, FA can be approximated by a filtering operation and a transformation of features (contour line filter). 4. Narrow band filtering (comb filter) in the space frequency domain is inadequate for pattern recognition because noise of different bandwidths and mean frequencies affects sinusoidal gratings differently. This excludes the use of a Fourier analyzer. 5. The relations between the measurable variables, which are the probabilities of detection (PD curves), and the characteristic variables of the recognition system are established analytically. 6. The probability of detection not only depends on signal energy but also on signal structure. This would not be the case in a simple matched filter system. 7. The differing probabilities of error in multiple detection experiments show that the interference is pattern specific and the bandwidth (steepness of the PD curves) is different for the different sets of patterns. 8. The distance between the reference vectors in feature space can be determined from the internal representation of the patterns defined by the model. Through multiple detection experiments it is possible to determine not only the relative distances between the patterns but also their absolute position in feature space.