Browse views: by Year, by Function, by GLF, by Subfunction, by Conference, by Journal

Benchmarking of Multivariate Similarity Measures for High Content Screening Fingerprints in Phenotypic Drug Discovery

Reisen, Felix and Zhang, Xian and Gabriel, Daniela and Selzer, Paul (2013) Benchmarking of Multivariate Similarity Measures for High Content Screening Fingerprints in Phenotypic Drug Discovery. Journal of Biomolecular Screening, 18 (10). pp. 1284-1297. ISSN 1087-05711552-454X

Abstract

With the evolving technology during the past decade High Content Screening (HCS) has become a powerful tool in drug discovery as it is amenable to high throughput measuring of cellular responses to chemical disturbance while providing a highly multiplexed and quantitative phenotypic readout. These image-based readouts such as cell size, shape, intensity and texture characterize the corresponding cell phenotype and are thus defined as HCS fingerprints. Systematic analysis of HCS fingerprints allows for objective computational comparisons of cellular responses that enable the detection of phenotypic outcomes of large scale small molecule screens and consequently compound hit candidates. Feature selection methods and similarity metrics, as the basis for phenotype identification and clustering, are critical for the quality of such computational analyses. Here, we present a systematic evaluation of more than 15 different similarity measures, such as Mahalanobis distance, distance correlation, maximum information coefficient, or cosine similarity, in combination with unsupervised linear or non-linear feature selection methods. We evaluate their potential to capture biologically relevant image features and their applicability in HCS and drug discovery with data from a high-throughput HCS campaign. The results of the experiments highlight the benefits and drawbacks of the different vector comparison methods in respect to the application under consideration. We show that non-linear correlation based similarity measures such as Kendall’s τ and Spearman’s ρ perform well in most of the tested scenarios and outperform other non-parameterized measures as the Euclidian or Manhattan distance. Our results also demonstrate that measures based on comparing the correlations of selected up- and down regulated variables can be high-performing as well. In this context, we present four novel modifications of the frequently used connectivity map similarity measure which surpass the original version in our experiments. This study provides a basis for phenotypic analysis in future HCS campaigns.

Item Type: Article
Keywords: High Content Screening, Phenotypic Screening, Similarity Metrics, HCS-fingerprints
Date Deposited: 14 Jun 2016 23:45
Last Modified: 06 Jul 2016 23:45
URI: https://oak.novartis.com/id/eprint/9492

Search

Email Alerts

Register with OAK to receive email alerts for saved searches.