*2.1. Experimental Setup*

As was described, nucleic acid quantification plays a major role in research and clinical study, ranging from the diagnosis of infectious diseases to food safety assurance and so on. Nucleic acids also have important biomarkers for biological studies and diagnosis [33]. In this experiment, a novel technique to identify DNA fragments is introduced. This technique identifies DNA fragments based on their frequency-dependent dielectric properties. In this experiment, DNA fragments which are coupled on micron-sized particles pass through a microfluidic channel made of polydimethylsiloxane (PDMS). The microfluidic PDMS channel is the first layer of the device. The second layer is a pair of electron beam-deposited reusable coplanar gold electrodes on a fused silica substrate. The microfluidic channel is 30 μm wide and 15 μm high, with a micron-sized electrode. The electrodes are 20 μm in width, and the gap between the two electrodes is 30 μm. We should point out that we experimentally verified that the sensitivity of the microfluidic channel increases as the width of the channel decreases and approaches the size of the bead. However, this increases the risk of clogging in the channel as it becomes too small. We designed the microchannel with the aforementioned configuration, which is large enough to minimize clogging and small enough to obtain sufficient sensitivity during measurements. Figure 1A represents the image of device which is made from PDMS and Figure 1B illustrates the microscopic image of the channel [5].

We compared our method to two commercially available technologies: gel electrophoresis and real-time PCR (also known as quantitative PCR or qPCR). Both of these are commonly used for DNA detection and sizing. The standard detection limit of gel electrophoresis using DNA bound to ethidium bromide is between 0.5 and 5.0 ng/band. However, with optimized gel electrophoresis technology, the Agilent Bioanalyzer can detect PCR products at concentrations as low as 0.1 ng/band and complete the analysis within 30 min. Real-time PCR has a detection limit of several copies of a DNA molecule per microliter or several fg/μL. However, it is relatively slow, with a sample processing time of over an hour, and has limitations in terms of DNA fragment size (e.g., amplicon size should be <200 bp). Furthermore, real-time PCR is costly and complex due to the need for simultaneous thermal cycling and fluorescence detection. It has limited multiplexing capabilities, making it difficult to miniaturize for portable applications. In contrast, our impedance sensor in combination with microfluidic technology has the potential for multiplexing and portability.

In this experiment, six different quantities and concentrations of DNA with a fixed length of 300 bp are integrated with a 2.8 μm paramagnetic bead and pass through a custom-made microfluidic chip. Three different types of magnetic beads (M270, M280, and C1) are tested. Based on the properties and the nature of our sensor, we chose to proceed with the M280 type (2.8 μm paramagnetic bead).

In this study, purified biotinylated DNA of a known quantity was serially diluted to obtain the desired concentrations. This DNA was then mixed with the beads to create DNA bound to the beads. The number of DNA molecules per bead is only an estimated average based on measurements of approximately 500 beads. This estimation was made after testing approximately 2000 beads per sample. The DNA-binding efficiency is determined by the very high binding affinity of the streptavidin–biotin interaction (Kd =10−15). The beads contain streptavidin, and the DNA is biotin-labeled. These beads have a binding capacity of 10 ug ds-DNA per mg of beads. This knowledge was used when combining various DNA amounts with the beads. In this study, the lower limit of detection identified is 0.0039 fmol, and the maximum DNA concentration is 0.19 fmol [5]. For testing the sensitivity of the sensor, we diluted a 1-microliter aliquot of the DNA-coated beads in 60 μL of phosphatebuffered saline (PBS) for detecting small amounts of DNA. PBS, which has a relatively

high salt concentration and high conductivity, has been shown to enhance the sensitivity of impedance measurements. Our sensor is capable of quantifying DNA fragments at high accuracy and precision at the femtomolar level and over a 100-fold dynamic range. Figure 1C represents the streptavidin–biotin linkage between DNA and beads. In this method, the target DNA was generated by using biotinylated DNA oligonucleotides and PCR (polymerase chain reaction) [5]. The procedures are as follows:


**Table 1.** Model outputs [5].


Multi-frequency impedance cytometry techniques have been performed to detect the impedance difference of beads integrated with different amounts of DNA. The impedance response was measured at 8 different frequencies simultaneously by using a multi-frequency lock-in amplifier (Zurich Instruments HF2A, Zurich, Switzerland). When an AC voltage is applied between electrodes, a flowing particle or cell perturbs the AC electric field, which results in a momentary increase in the impedance/decrease in the voltage.

In this experimental setup, the first electrode is excited with combination of 8 frequencies ranging from 100 kHz to 20 MHz, and the second electrode is connected to the transimpedance amplifier. Figure 1E shows representative multi-frequency time series data of bare magnetic beads in voltage. The voltage is normalized for a straightforward comparison. For testing of 300 bp DNA beads, 6 different concentrations of DNA were measured to study the effect of the different amounts of DNA on the frequency. To compare the impedance response from different DNA concentrations, the impedance of bare beads with no DNA was measured in the same experiment. Figure 1F shows representative time series data comparing bare magnetic beads to DNA at the highest concentration (500 kHz frequency). In this figure, as well, the voltage is normalized for better comparison.

Table 1 showed the different concentrations of DNA coupled with paramagnetic beads. To compare the impedance response of different concentrations of DNA integrated with paramagnetic beads, we performed the same experiment with bare beads. In this experiment, 2.8 μm paramagnetic beads with no DNA concentration passed through the microfluidic channel, and the impedance response of a bare bead is obtained.

**Figure 1.** Overview of the process. (**A**) Image of device. (**B**) Microscopic image of channel and electrodes. (**C**) The sample preparation after binding of biotinylated DNA to paramagnetic beads. (**D**) The schematic diagram of detection. (**E**) Representative data of bare paramagnetic beads. (**F**) Representative data of bare paramagnetic beads and beads integrated with most-concentrated DNA.

The results showed that there is positive relationship between DNA amounts per bead and the impedance peak response (IPR). As DNA concentration per bead increases, the IPR increases as well. These findings showed the positive correlation of DNA amounts attached to beads with IPR. In addition, increased DNA amounts resulted in a higher surface potential of the beads, which was associated with a larger impedance difference compared to the control bare bead. The details of the nucleic acid sample preparation and the impedance chip preparation, along with the experimental procedures, are those described in the work by Sui et al. [5].

As we described, in this experiment 6 different DNA concentrations coupled to paramagnetic beads are examined. In addition, it is very difficult to bind very small inputs of

DNA to beads. Given the need for testing small DNA amounts in many samples, there is a utility for novel machine learning approaches for accurate and high-throughput DNA quantification. Furthermore, by proposing a general regression model, we can predict unknown DNA concentrations with a fixed length of 300 bp coupled to a bead. The combination of microfluidics, which generates vast amounts of complex data, with machine learning methods represents an emerging opportunity in biotechnology. On the other hand, the development of microfluidic chips and experimental design is expensive and time-consuming, and the method is prone to bias by the user. In the next section, we propose a novel hybrid regression model to address this difficulty. All the electrical properties obtained from the Zurich Instruments tools (including frequency, imaginary and real part of peak intensity) are leveraged to identify correlations between these properties and the amount of DNA per bead. Machine learning tools are then used to develop a general model and platform for predicting nucleic acid concentration.
