Use Of Unsupervised Clustering Research Articles

Is Unsupervised Clustering Somehow Truer?

Scientists increasingly approach the world through machine learning techniques, but philosophers of science often question their epistemic status. Some philosophers have argued that the use of unsupervised clustering algorithms is more justified than the use of supervised classification, because supervised classification is more biased, and because (parametric) simplicity plays a different and more interesting role in unsupervised clustering. I call these arguments the No-Bias Argument and the Simplicity-Truth Argument. I show how both arguments are fallacious and how, on the contrary, the use of supervised classification is at least as justified as the use of unsupervised clustering.

Unsupervised machine learning for project stakeholder classification: Benefits and limitations

The literature has shown that an accurate classification of project stakeholders allows for more comprehensive planning of their management strategies. The most used classification methods have limitations stemming from using a small number of stakeholder attributes thus returning high-level and imprecise classification results. This work investigates the potential benefits and limitations of adopting unsupervised machine learning clustering as an alternative method to automatically recognize stakeholder groups. The paper demonstrates the application of a PAM algorithm for project stakeholder classification, employing qualitative and quantitative data collected from a real project in an IT Italian company. The results show that the use of unsupervised clustering leads to a more granular and detailed stakeholder grouping that enables the design of better refined and customized stakeholder management strategies. Furthermore, the results of the paper demonstrate that the use of this methodology, when data is taken from a structured dataset, reduces the degree of subjectivity in classification, promoting a data-driven approach to project stakeholder management.

Optimization of deep learning models for the prediction of gene mutations using unsupervised clustering.

Deep learning models are increasingly being used to interpret whole-slide images (WSIs) in digital pathology and to predict genetic mutations. Currently, it is commonly assumed that tumor regions have most of the predictive power. However, it is reasonable to assume that other tissues from the tumor microenvironment may also provide important predictive information. In this paper, we propose an unsupervised clustering-based multiple-instance deep learning model for the prediction of genetic mutations using WSIs of three cancer types obtained from The Cancer Genome Atlas. Our proposed model facilitates the identification of spatial regions related to specific gene mutations and exclusion of patches that lack predictive information through the use of unsupervised clustering. This results in a more accurate prediction of gene mutations when compared with models using all image patches on WSIs and two recently published algorithms for all three different cancer types evaluated in this study. In addition, our study validates the hypothesis that the prediction of gene mutations solely based on tumor regions on WSI slides may not always provide the best performance. Other tissue types in the tumor microenvironment could provide a better prediction ability than tumor tissues alone. These results highlight the heterogeneity in the tumor microenvironment and the importance of identification of predictive image patches in digital pathology prediction tasks.

Big data approaches to develop a comprehensive and accurate tool aimed at improving autism spectrum disorder diagnosis and subtype stratification

PurposeAutism spectrum disorder (ASD) is a complex neurodevelopmental disorder that is difficult to diagnose accurately due to its heterogeneous clinical manifestations. Comprehensive models combining different big data approaches (e.g. neuroimaging, genetics, eye tracking, etc.) may offer the opportunity to characterize ASD from multiple distinct perspectives. This paper aims to provide an overview of a novel diagnostic approach for ASD classification and stratification based on these big data approaches.Design/methodology/approachMultiple types of data were collected and recorded for three consecutive years, including clinical assessment, neuroimaging, gene mutation and expression and response signal data. The authors propose to establish a classification model for predicting ASD clinical diagnostic status by integrating the various data types. Furthermore, the authors suggest a data-driven approach to stratify ASD into subtypes based on genetic and genomic data.FindingsBy utilizing complementary information from different types of ASD patient data, the proposed integration model has the potential to achieve better prediction performance than models focusing on only one data type. The use of unsupervised clustering for the gene-based data-driven stratification will enable identification of more homogeneous subtypes. The authors anticipate that such stratification will facilitate a more consistent and personalized ASD diagnostic tool.Originality/valueThis study aims to utilize a more comprehensive investigation of ASD-related data types than prior investigations, including proposing longitudinal data collection and a storage scheme covering diverse populations. Furthermore, this study offers two novel diagnostic models that focus on case-control status prediction and ASD subtype stratification, which have been under-explored in the prior literature.

Exploring use of unsupervised clustering to associate signaling profiles of GPCR ligands to clinical response

Signaling diversity of G protein-coupled (GPCR) ligands provides novel opportunities to develop more effective, better-tolerated therapeutics. Taking advantage of these opportunities requires identifying which effectors should be specifically activated or avoided so as to promote desired clinical responses and avoid side effects. However, identifying signaling profiles that support desired clinical outcomes remains challenging. This study describes signaling diversity of mu opioid receptor (MOR) ligands in terms of logistic and operational parameters for ten different in vitro readouts. It then uses unsupervised clustering of curve parameters to: classify MOR ligands according to similarities in type and magnitude of response, associate resulting ligand categories with frequency of undesired events reported to the pharmacovigilance program of the Food and Drug Administration and associate signals to side effects. The ability of the classification method to associate specific in vitro signaling profiles to clinically relevant responses was corroborated using β2-adrenergic receptor ligands.

ACTINN: automated identification of cell types in single cell RNA sequencing.

Cell type identification is one of the major goals in single cell RNA sequencing (scRNA-seq). Current methods for assigning cell types typically involve the use of unsupervised clustering, the identification of signature genes in each cluster, followed by a manual lookup of these genes in the literature and databases to assign cell types. However, there are several limitations associated with these approaches, such as unwanted sources of variation that influence clustering and a lack of canonical markers for certain cell types. Here, we present ACTINN (Automated Cell Type Identification using Neural Networks), which employs a neural network with three hidden layers, trains on datasets with predefined cell types and predicts cell types for other datasets based on the trained parameters. We trained the neural network on a mouse cell type atlas (Tabula Muris Atlas) and a human immune cell dataset, and used it to predict cell types for mouse leukocytes, human PBMCs and human T cell sub types. The results showed that our neural network is fast and accurate, and should therefore be a useful tool to complement existing scRNA-seq pipelines. The codes and datasets are available at https://figshare.com/articles/ACTINN/8967116. Tutorial is available at https://github.com/mafeiyang/ACTINN. All codes are implemented in python. Supplementary data are available at Bioinformatics online.

Computed Tomography–Based Score Indicative of Lung Cancer Aggression (SILA) Predicts the Degree of Histologic Tissue Invasion and Patient Survival in Lung Adenocarcinoma Spectrum

ObjectiveMost computed tomography (CT)-detected lung cancers are adenocarcinomas (ACs), representing lesions with variable tissue invasion, aggressiveness, and clinical outcome. Visual radiologic characterization of AC pulmonary nodules is both inconsistent and inadequate to confidently predict histopathologic classification or prognosis. Comprehensive pathologic interpretation requires full nodule resection. We have described a computerized scoring system for AC detected on CT scans that can noninvasively estimate the degree of histologic invasion and simultaneously predict patient survival. MethodsThe Computer-Aided Nodule Assessment and Risk Yield has been validated to characterize CT-detected nodules across the spectrum of AC. With the use of unsupervised clustering, nine natural exemplars were identified as basic radiographic features of AC nodules. We now introduce the Score Indicative of Lung Cancer Aggression (SILA), which is a cumulative aggregate of normalized distributions of ordered Computer-Aided Nodule Assessment and Risk Yield exemplars. The SILA values for each of 237 unique nodules in AC were compared with the histopathologically defined maximum linear extent of tumor invasion. With use of the SILA, Kaplan-Meier survival and Cox proportionality analysis were performed on patients with stage I AC, who comprised a subset of our cohort. ResultsThe SILA discriminated between indolent and invasive AC (p < 0.0001). In addition, prediction of linear extent of histopathologic tumor invasion was possible. In stage I AC, three separate SILA prognosis groups were identified: indolent, intermediate, and poor, with 5-year survival rates of 100%, 79%, 58%, respectively. Cox proportionality hazard modeling predicted a 50% increase in mortality, for a 0.1 unit increase in the SILA over a median follow-up time of 3.6 years (p < 0.0002). ConclusionsThe SILA is a computer-based analytic measure allowing noninvasive approximation of histologic invasion and prediction of patient survival in CT-detected AC nodules.

Placental transcriptional and histologic subtypes of normotensive fetal growth restriction are comparable to preeclampsia

Infants born small for gestational age because of pathologic placenta-mediated fetal growth restriction can be difficult to distinguish from those who are constitutionally small. Additionally, even among fetal growth-restricted pregnancies with evident placental disease, considerable heterogeneity in clinical outcomes and long-term consequences has been observed. Gene expression studies of fetal growth-restricted placentas also have limited consistency in their findings, which is likely due to the presence of different molecular subtypes of disease. In our previous study on preeclampsia, another heterogeneous placenta-centric disorder of pregnancy, we found that, by clustering placentas based only on their gene expression profiles, multiple subtypes of preeclampsia, including several with co-occurring suspected fetal growth restriction, could be identified. The purpose of this study was to discover placental subtypes of normotensive small-for-gestational-age pregnancies with suspected fetal growth restriction through the use of unsupervised clustering of placental gene expression data and to investigate their relationships with hypertensive suspected fetal growth-restricted placental subtypes. A new dataset of 20 placentas from normotensive small-for-gestational-age pregnancies (birthweight <10th percentile for gestational age and sex) with suspected fetal growth restriction (ultrasound features of placental insufficiency) underwent genome-wide messenger RNA expression assessment and blinded detailed histopathologic evaluation. These samples were then combined with a subset of samples from our previously published preeclampsia cohort (n=77) to form an aggregate fetal growth-focused cohort (n=97) of placentas from normotensive small-for-gestational-age, hypertensive (preeclampsia and chronic hypertensive) small-for-gestational-age, and normotensive average-for-gestational-age pregnancies. Gene expression data were subjected to unsupervised clustering, and clinical and histopathologic features were correlated to the identified sample clusters. Clustering of the aggregate dataset revealed 3 transcriptional subtypes of placentas from normotensive small-for-gestational-age/suspected fetal growth-restricted pregnancies, with differential enrichment of clinical and histopathologic findings. The first subtype exhibited either no placental disease or mild maternal vascular malperfusion lesions, and, co-clustered with the healthy average-for-gestational-age control subjects; the second subtype showed more severe evidence of hypoxic damage and lesions of maternal vascular malperfusion, and the third subtype demonstrated an immune/inflammatory response and histologic features of a maternal-fetal interface disturbance. Furthermore, all 3 of these normotensive small-for-gestational-age subtypes co-clustered witha group of placentas from hypertensive small-for-gestational-age pregnancies with more severe clinical outcomes, but very comparable transcriptional and histologic placental profiles. Overall, this study provides evidence for at least 2 pathologic placental causes of normotensive small-for-gestational-age, likely representing true fetal growth restriction. These subtypes also show considerable similarity in gene expression and histopathology to our previously identified "canonical" and "immunologic" preeclampsia placental subtypes. Furthermore, we discovered a subtype of normotensive small-for-gestational-age (with suspected fetal growth restriction) with minimal placental disease that may represent both constitutionally small infants and mild fetal growth restriction, although these cannotbedistinguished with the currently available data. Future work that focuses on the identification of etiology-driven biomarkers and therapeutic interventions for each subtype of fetal growth restriction is warranted.

Pan-cancer analysis reveals presence of pronounced DNA methylation drift in CpG island methylator phenotype clusters.

To provide characteristics of major genomic correlates in CpG island methylator phenotype-high (CIMP-H) subgroups in relation to corresponding non-CIMP-H subgroups by use of phenotypic, DNA methylation and RNAseq data. Twenty-three datasets generated by The Cancer Genome Atlas project encompassing over 7200 unique samples were analyzed. We identified 23 CIMP-H clusters by use of unsupervised clustering. More than 90% of CIMP-H clusters were significantly associated with accelerated epigenetic mitotic clock, demethylation of enhancer sites, backbone and repetitive sequences. Pronounced epigenetic drift observed in majority of CIMP-H subgroups may be related to increased cell division rate, which leads to expansion of DNA methylation errors. This may explain pan-cancer mechanism of establishing CIMP-H in majority of tissue types.