Biomedical Engineering Reference
In-Depth Information
analysis. Jurs et al. (10) presented a comprehensive review on computational methods for
the analysis of chemical sensor array data covering the basis of each approach, its suit-
ability and applications, and a critical discussion of each method.
Data processing methods include chemometric techniques, ANNs, pattern recognition,
or a combination of these (8). The selection of an appropriate method for data analysis is
strongly dependent upon a number of experimental variables as well as the nature and the
type of information under consideration (10). Depending on the final application and
required performance of the system, computational methods in conjunction with multiar-
ray sensors can provide qualitative or quantitative analysis or both. Typically, the data
processed using computational methods are graphically presented. However, when the
amount of data is very high, graphical presentation may be difficult, and therefore addi-
tional methods such as principal component analysis are needed to reduce the dimen-
sionality of datasets (10). In some cases, two or more computational methods are used and
compared so as to achieve a higher effectiveness. For instance, in a recent study, we have
compared the performance of two computational methods when investigating detection
and classification of OP nerve agents: ANNs and a novel learning scheme, SVMs. For this
specific application, three SVM kernels were tested: the Scholkopf kernel, the polynomial,
and the Gaussian radial basis function. The use of SVM methods is a recent development,
almost unknown from analytical chemistry perspective. Experimental results showed that
SVMs provide a more accurate classification of OP than the traditional ANN as machine
learning and recognition programs (4). The sensor design utilized arrays of conducting
polymers from a commercial electronic nose instrument to generate molecular patterns for
the six different OPs: parathion, malathion, dichlorvos, trichlorfon, paraoxon, and diazi-
non. In terms of classification, dichlorvos, trichlorfon, diazinon, parathion, and paraoxon
were classified perfectly. This is remarkable considering the resemblances in the chemical
structures of paraoxon and parathion with only a single atom difference. Malathion was
detected at 95%. When the ANN was used, the positive predictions were inconsistent for
the range of chemicals tested.
19.6
Conclusion and Future Trends
The current trends in the development of biosensors for toxicity monitoring and screening
are conducted through small and fast multiarray sensors that will allow detection of mul-
tiple samples simultaneously, with substantial decrease in costs per sample throughput.
These devices are expected to function as alarm systems by sensing the presence of toxi-
cants up to a fixed threshold value. Ideally, multiarray biosensors should generate a pat-
tern of known analytical signals with distinct characteristics for samples or analytes that
could be subsequently used to correctly classify and differentiate between unknown sam-
ples possessing various degrees of toxicity. The information could be qualitative or semi-
quantitative and extremely useful for an initial screening of toxicants in the field, before
more sophisticated investigations could be performed in a specialized analytical labora-
tory. The real success in the development of a reliable multiarray biosensor depends on
several parameters including the type and the number of electrodes composing the array,
the sophistication of the transductor system, which may facilitate efficient signal conver-
sion on multiple channels, as well as the availability of a powerful data processing and
quantification program. This chapter provided a general overview of current multiarray
biosensors including details regarding their design, signal transduction data treatment,
and applications for the toxicity monitoring and bacterial pathogens.
