Nano‐SAR Development for Bioactivity of Nanoparticles with Considerations of Decision Boundaries. Issue 9 (19th February 2013)

• Record Type: Journal Article
• Authors: Liu, Rong; Rallo, Robert; Weissleder, Ralph; Tassa, Carlos; Shaw, Stanley; Cohen, Yoram
• Is Part Of: Small. Volume 9:Issue 9/10(2013)
• Page Start: 1842
• Page End: 1852
• Publication Date: 2013-02-19
• Subjects: Nanotechnology -- Periodicals; Nanoparticles -- Periodicals; Microtechnology -- Periodicals; 620.5
• DOI: 10.1002/smll.201201903
• Languages: English
• ISSNs: 1613-6810

Abstract

<abstract abstract-type="main" xml:lang="en"> <title>Abstract</title> <p>The development of classification nano‐structure–activity Relationships (nano‐SARs) of nanoparticle (NP) bioactivity is presented with the aim of demonstrating the integration of multiparametric toxicity/bioactivity assays to arrive at statistically meaningful class definitions (i.e., bioactivity/inactivity endpoints), as well as the implications of nano‐SAR applicability domains and decision boundaries. Nano‐SARs are constructed based on a dataset of 44 iron oxide core nanoparticles (NPs), used in molecular imaging and nano‐sensing, containing bioactivity profiles for four cell types and four different assays. Class definitions are developed on the basis of 'hit' (i.e., significant bioactivity) identification analysis and self‐organizing map based consensus clustering; these class definitions enable construction of nano‐SARs of a high classification accuracy (>78%) with different NP descriptor combinations that include primary size, spin‐lattice and spin‐spin relaxivities, and zeta potentials. Analysis of the nano‐SAR performance for different class definitions suggests that H4 (i.e., class with at least four hits) is a reasonable endpoint (from a 'regulatory' viewpoint) for keeping the level of false negatives (i.e., incorrect labeling of bioactive NPs as inactive) low. The establishment of a quantitative nano‐SAR applicability domain is demonstrated, making use of a probability density with the H4 class definition and naive Bayesian classifier (NBC) model (with spin‐lattice relaxivity and zeta potential as descriptors). Decision boundaries are determined for the above H4/NBC nano‐SAR for different acceptance levels of false negative to false positive predictions, illustrating a practical approach that may assist in regulatory decision making with a consideration of reducing the likelihood of identifying bioactive NPs as being inactive.</p> </abstract>