Information visualization techniques for sensing and biosensing

Abstract

The development of new methods and concepts to visualize massive amounts of data holds the promise to revolutionize the way scientific results are analyzed, especially when tasks such as classification and clustering are involved, as in the case of sensing and biosensing. In this paper we employ a suite of software tools, referred to as PEx-Sensors, through which projection techniques are used to analyze electrical impedance spectroscopy data in electronic tongues and related sensors. The possibility of treating high dimension datasets with PEx-Sensors is advantageous because the whole impedance vs. frequency curves obtained with various sensing units and for a variety of samples can be analyzed at once. It will be shown that non-linear projection techniques such as Sammon's Mapping or IDMAP provide higher distinction ability than linear methods for sensor arrays containing units capable of molecular recognition, apparently because these techniques are able to capture the cooperative response owing to specific interactions between the sensing unit material and the analyte. In addition to allowing for a higher sensitivity and selectivity, the use of PEx-Sensors permits the identification of the major contributors for the distinguishing ability of sensing units and of the optimized frequency range. The latter will be illustrated with sensing units made with layer-by-layer (LbL) films to detect phytic acid, whose capacitance data were visualized with Parallel Coordinates. Significantly, the implementation of PEx-Sensors was conceived so as to handle any type of sensor based on any type of principle of detection, representing therefore a generic platform for treating large amounts of data for sensors and biosensors.

Information Visualization to Enhance Sensitivity and Selectivity in Biosensing

“…Feature selection is essential in many data analysis problems, including biosensor optimization. The work by Paulovich et al [29], for instance, deals with feature selection in the context of seeking to optimize sensor performance (this is further discussed in Sect. 4).…”

“…In addition to the linear techniques, such as PCA, non-linear methods can be employed, as we shall comment upon below. The third advantage is the possible optimization of sensing performance that goes beyond exploiting the whole data, for instance employing feature selection strategies to maximize inter-cluster distances while minimizing intracluster distances [29].…”

“…Once again, the authors tried several projection methods available in a free platform called PEx-Sensors (see below) [29], and IDMAP provided the best classification results. the buffer, and the so-called synthetic samples made with the buffer to which anti-Leishmania, anti-T. Cruzi and negative antibodies were added.…”

“…[92] With regard to the third advantage of information visualization methods, one may mention the optimization of biosensor performance using feature selection coupled with visualizations obtained with projection techniques. Paulovich et al [29] used Parallel Coordinates (PC) visualizations [21] of capacitance data of a PAH/PVS sensing unit, obtained much in the same way as the aforementioned measurements, for aqueous solutions containing the analyte phytic acid to be detected. Owing to the lack of specific interaction, the distinguishing ability of this sensing unit was expected to be poor.…”

“…Choosing the most suitable frequencies for distinguishing the sample amounts to feature selection, which can be done quantitatively using the silhouette coefficients. Paulovich et al [29] employed a genetic algorithm to scan the whole data space of cluster silhouettes and automatically identify the best frequencies for distinction. Figure 13 depicts a parallel coordinates visualization for the 10 best frequencies selected, where a better distinction capability is readily observed in comparison with Fig.…”