Biomedical Engineering Reference
In-Depth Information
TABLE 10.4
Classification of Dense Cell Membrane Samples Versus
Disrupted Cell Membranes of TiO
2
Features
ε
resub
x
1
x
4
x
5
0
x
0
x
1
x
2
0.067
x
1
x
2
x
4
0.067
x
1
x
2
x
5
0.067
x
2
x
4
x
5
0.067
Source:
Reprinted with permission from Sayes, C. and Ivanov, I. 2010.
Risk.
Anal
. 30, 1723-1734.
Note:
Triplet-wise LDA classifiers are shown. ε
resub
denotes the resubstitution
error for the respective classifier. For the legend of the labeling of the
features, see Table 10.1. With permission.
10.5.4 a
rtIfIcIal
I
NtellIgeNce
A new avenue has opened to QSAR models recently through the use of artificial intelligence.
One group of scientists applied neural and fuzzy-neural networks with the QSAR approach
(Mazzatorta et al., 2003). This study was conducted on 562 organic compounds in order to estab-
lish models for predicting the acute toxicities in lish In the discussion section, they stressed the
problems in neural network training, such as over fitting, and illustrated the difficulties of model-
ing the toxicity of chemicals. Although introducing artificial intelligence into QSAR approaches
has been criticized, when the problems caused by parameters are settled, it would eventually be
a powerful method.
10.5.5 N
aNoINforMatIcs
Nanoinformatics is a relatively new field that has emerged over the last few years. Its purpose is to
integrate information relevant to the nanoscale science and engineering communities to develop
and implement effective mechanisms for collecting, validating, storing, sharing, analyzing, model-
ing, and applying that information. Nanoinformatics can be applied in five major areas: Delivery
systems, implantable devices, diagnosis and prevention, therapeutics, and materials (Maojo et al.,
2012). All of the
in silico
methods are the basis of nanoinformatics. However, it still faces a num-
ber of challenges, mainly due to the rapid emergence of new and revised nanoparticles, data from
uncertain and inaccurate methods and protocols, and various types of risks.
10.6 CONCLUSION
Current research on nanotoxicity is very dispensing. Various cell lines and animal models are
being used, and different mechanisms of action are being tested. Moreover, the inconsistencies in
nanoparticle preparation and dosimetry make it even harder to draw any conclusions from present
data. Thus, there is an urgent need to standardize nanotoxicity studies in order to fairly assess the
potential risk of new and existing nanoparticles.
Toxicogenomic approaches with proteomics and metabolomics provide a comprehensive assess-
ment of gene expression, protein expression, or metabolite generation in a particular tissue, organ, or
