Stability and performance of titanium dioxide nanoparticle biosensor platforms for rapid detection of human serum albumin biosensor

Abstract

The development of electrical biosensors for serum albumin detection has gained significant attention due to their high sensitivity, low cost, and simplicity. However, studies addressing the structural stability and platform regeneration of such sensors remain limited. This study presents a nanoscale titanium dioxide (TiO₂) nanoparticle-based interdigitated electrode (IDE) amperometric biosensor for the rapid detection of human serum albumin (HSA). Molecular docking simulations were first employed to investigate the structural and electrostatic characteristics of the antibody–antigen interface. The simulations confirmed that APTES-functionalized nanostructured TiO₂ IDE surfaces enhance antibody binding stability at pH 7 under the experimental conditions, providing molecular-level validation of the robustness of antibody immobilization on TiO₂ surface. The biosensor's performance was evaluated through electrical current response resulting from the antibody–antigen interaction on a TiO₂ nanoparticle thin-film platform, achieving detection of HSA concentrations ranging from 10 µg mL⁻¹ down to 1 pg mL⁻¹, with a coefficient of determination (R²) of 0.98 and a low limit of detection (LOD) of 0.33 pg mL⁻¹. Sensor reproducibility was examined using five independently prepared samples, yielding consistent current–voltage (I–V) profiles and relative standard deviation of 9.25% and 12.26% at 1 ng mL⁻¹ and 1 µg mL⁻¹ HSA concentrations, respectively. The structural robustness of the sensor platform was further evaluated through controlled surface cleaning. While regeneration has been explored in other biosensor types, systematic electrical evaluation of reused TiO₂ nanoparticle IDE platforms has not been well established. The regenerated devices maintained stable I–V trends and functional detection capability. These findings highlight the promise of TiO₂ nanoparticle IDE systems for cost-effective and resource-limited diagnostic applications.

A TiO₂ nanoparticle biosensor enables ultrasensitive human serum albumin detection through antibody–antigen recognition on a functionalized IDE surface, demonstrating high stability, reproducibility, and cleaning-based regeneration.

1. Introduction

Human serum albumin (HSA) is the most abundant protein in blood plasma and plays a vital role in maintaining physiological homeostasis. It regulates fluid distribution and contributes significantly to plasma oncotic pressure. In addition, HSA exhibits remarkable ligand-binding capacity, interacting with various of substances, including free metal ions, hormones, and pharmaceutical drugs.^1,2 As noted by Kim (2017), the predominance of HSA in plasma, along with its functional versatility, renders it a clinically significant biomarker particularly for assessing kidney function. Normal HSA concentrations in the vascular system range from 35 g L⁻¹ to 50 g L⁻¹.³ Clinically, HSA levels are often evaluated through urine analysis, as healthy kidneys effectively prevent albumin from passing into the urinary tract. Under normal circumstances, the presence of HSA in the urinary system should be negligible or zero (proteinuria < 4 mg m⁻² h⁻¹) (microalbuminuria < 30 mg L⁻¹), due to selective filtration by the glomerular filtration barrier, which permits only water and small solutes (<5 kDA) to pass.⁴ Elevated urinary albumin levels, referred to as albuminuria or microalbuminuria, are an early marker of kidney disease. This highlights the importance of HSA as a critical biomarker for diagnosing kidney diseases and emphasizes the need for highly sensitive sensors capable of detecting HSA at clinically relevant concentrations.

According to Xu (2020), early research on HSA detection relied primarily on precipitation-based methods, which were later deemed inefficient and time-consuming. Over time, a range of analytical techniques have been developed for HSA detection, including dye-binding methods,⁵ optical methods,⁶ electrochemical methods,⁷ and immunochemical approaches.⁸ While these methods have improved the detection of HSA, comprehensive reviews assessing their relative strengths and weaknesses remain scarce.⁹ Moreover, these methods often require costly instrumentation and substantial amounts of chemical reagents. To address these limitations, microdevices based on electrical biosensors have been developed as a more efficient means of detecting HSA.

A biosensor is a device that detects biological responses via physical, chemical, or biological changes, and converts them into measurable signals for quantitative analysis.¹⁰ Among the various biosensor types, electrical-based biosensors are renowned for their high sensitivity, rapid response time, label-free approaches, high specificity, and low manufacturing cost due to ease of miniaturization, making them suitable for various applications and research.¹¹ These biosensors detect target biomolecules, or analytes, by measuring electrical signals such as current, voltage, conductance or impedance.¹² This study focuses on an amperometric biosensor platform, which monitor changes in electrical current to assess the presence and concentration of the analyte. This measurement is made possible by the interaction between the analyte and the functionalized electrode surface.¹³ For electrical readout, a direct current (DC) picoammeter is utilized due to its portability and capability to detect small changes in currents.¹⁴ In this work, a biosensor incorporating a titanium dioxide (TiO₂)-based interdigitated electrode (IDE) is used to transduce the antibody–antigen interaction of HSA into a measurable electrical current.

IDE feature a closely spaced, repeated finger-like shaped design of electrodes that can operate on a micro-scale, also known as a microelectrode. IDE are commonly used in sensor structures requiring higher sensitivity and signal detection capabilities. Their effectiveness is largely attributed to a low ohmic drop and high signal-to-noise ratio, both of which result from the minimal spacing between adjacent electrode fingers. This geometry increases the effective surface area, thereby enhancing current output. The output signal strength is also directly influenced by the IDE's structural parameters, such as finger width, spacing, and length.¹⁵ While there have been limited experiments on HSA detection using IDE-based biosensors, Tsai (2016) reported successful detection within a concentration range of 10–300 µg mL⁻¹, with a detection limit of 9.7 µg mL⁻¹, using a biosensor featuring a standard screen-printed electrode.¹⁶

In this research, the IDE plays a critical role in converting the biological response of HSA into a current signal. To facilitate this transduction process, the IDE is coated with metal oxide nanoparticles, specifically titanium dioxide (TiO₂), which functions as a semiconductor within the biosensor device. During sensing process, the conductivity changes of the metal oxide play a pivotal role for current modulation. Among various metal oxide used in IDE coatings, TiO₂ has garnered significant attention from researchers due to its proven efficacy across a wide range of applications.¹⁷ TiO₂ offers an exceptional combination of biocompatibility, non-toxicity, cost-effectiveness, and, most importantly, chemical stability at the nanoscale.¹⁸ Its strong resistance to chemical degradation, even in harsh or biological environments, sets it apart from many other oxide materials. This stability ensures long-term sensor performance and compatibility with living systems, which is essential for biomedical applications. Additionally, TiO₂'s low-cost synthesis and proven structural integrity in physiological conditions make it a highly reliable and scalable material for durable biosensor development.¹⁹ In this study, TiO₂ is applied in nanoparticle form to increase the surface area of the electrode, thereby enhancing its sensitivity. This nanostructured TiO₂ provides a large, reactive interface that is particularly advantageous for amperometric biosensing.²⁰

Accordingly, this study presents the development of a nanostructured TiO₂ nanoparticle-based IDE biosensor for highly sensitive detection of HSA. The biosensor utilizes a titanium dioxide (TiO₂) nanoparticle platform to transduce the antibody–antigen interaction into a measurable electrical current. The performance of the biosensor is evaluated through experiments assessing the detection range, sensitivity, and limit of detection (LOD). The sensor's effectiveness is demonstrated by detecting ultra-low concentrations of HSA, highlighting its potential for practical applications, particularly in point-of-care (POC) testing and mass production. Overall, this work aims to establish the TiO₂ nanoparticle-based IDE biosensor as a reliable and cost-effective tool for detecting HSA, offering a promising solution for improving early diagnosis and monitoring of kidney diseases.

2. Methodology

2.1. Materials

Titanium isopropoxide (Ti(OCH(CH₃)₂)₄, 97%), acetic acid (CH₃COOH, 99.7%), nitric acid (HNO₃, 68%), and hydrochloric acid (HCl, 37%), were obtained from Sigma-Aldrich. Anti-Human Albumin Antibody produced in rabbit (A3293) and Albumin from human serum (A3782) were also obtained from Sigma Aldrich. Ethanol absolute was obtained from HmbG Chemicals. AZ 1518 Photoresist and AZ 326 MIF Developer were obtained from Merck Performance Materials GmbH.

2.2. Instrumentation

The structural properties were observed using a Bruker D8 Advance Pro X-ray diffractometer (XRD) with CuKα radiation (45 kV, 20 mA, λ = 0.15406 nm) and (20° < 2θ < 60°). The morphological properties were observed using MERLIN (Carl Zeiss) field emission scanning electron microscopy (FESEM). FESEM operated at 3 kV and had a scanning magnification of 100 kX. The functional groups and chemical bonds of the sample were examined using a PerkinElmer Spectrum 400 Fourier Transform Infrared Spectrometer (FTIR). Current–voltage (I–V) and real-time measurements were performed by Keithley 236 Source-Measure Units (SMU).

2.3. TiO₂ nanoparticle synthesis

A TiO₂ solution was prepared by mixing the TiO₂ precursor with a solvent and a stabilizer. Titanium isopropoxide (Ti(OCH(CH₃)₂)₄) was used as the precursor, ethanol (C₂H₅OH) as the solvent, and acetic acid (CH₃COOH) as the stabilizer. These three solutions were mixed in a volumetric ratio of 1 : 9 : 0.1, respectively. The solution was stirred using a magnetic stirrer at 75 °C for 1 hour.²¹ A p-type silicon nitride (Si₃N₄) wafer was utilized as the substrate. Si₃N₄ wafer was prioritized over common alternatives like SiO₂ due to its superior mechanical hardness and chemical resistance. It acts as a high-density diffusion barrier that maintains high dielectric strength and structural integrity, which is essential for a biosensor platform.^22,23 The Si₃N₄ wafer was cleaved using a diamond cutter to achieve the size required for the IDE. Subsequently, it was rinsed with isopropyl alcohol followed by deionized (DI) water, and dried using a nitrogen (N₂) gun to ensure a dry and clean wafer devoid of watermarks and foreign particles.

2.3.1. IDE fabrication

IDEs were fabricated using photolithography process. Fig. 1 shows the fabrication process of the IDE. Initially, a 5 nm of chromium (Cr) film was deposited by an electron beam thermal evaporator (MODEL: VCP-1100) with a deposition rate of 0.5 Å s⁻¹. This was followed by a deposition of a 45 nm of gold (Au) film at a rate of 0.3 Å s⁻¹. Gold was selected as the primary electrode material due to its high electrical conductivity and chemical inertness. Unlike more reactive metals such as aluminum or silver, gold remains stable in varying ionic environments and resists oxidation, ensuring the integrity of the electrodes during surface functionalization and repeated sensing cycles.^24,25 A layer of positive photoresist (AZ 1518 Photoresist) was applied onto the gold-coated substrate using resist spin-coater (MODEL: Mikasa SpinCoater 1H-D3) with a spin speed of 3000 rpm for 1 minute. The coated substrate was soft-baked on a hot plate at 90 °C for 1 minute. Then, the IDE pattern was transferred from a photomask of IDE onto the deposited photoresist using mask aligner (Model: Karl Suss MJB) under UV exposure for 1 min. Subsequently, the patterned substrate was immersed in a developer solution (AZ 326 MIF Developer) to dissolve the unexposed areas of the photoresist, revealing the patterned areas where gold will be etched. After that, the patterned substrate was submerged in the aqua regia (a mixture of HNO₃ and HCl). Aqua regia acts as the Au and Cr etchant solution to dissolve the exposed areas, leaving behind the patterned gold structures protected by the developed photoresist. Finally, the remaining photoresist was stripped using acetone, rinsed with DI water, and blown dry with N₂ gun. The resulting IDE features 9 µm electrode gap, a geometry selected based on prior optimization to ensure a high signal-to-noise ratio. This specific spacing prevents the electrical instability and short-circuiting risks associated with smaller gaps (less than 2 µm) while avoiding the high resistance and weakened electric fields observed in larger gaps (more than 50 µm), thereby ensuring a stable and reproducible baseline.^17,26 Once the IDE was patterned onto the wafer, the TiO₂ solution was spin-coated on top of the IDE using a spin coater at a spin speed of 3000 rpm for 30 seconds. The TiO₂ thin films formed were subsequently dried on a hotplate at 175 °C for 10 minutes. The TiO₂ coating process was repeated three times to achieve the desired film thickness.¹⁷ Finally, the TiO₂ thin films were annealed at 600 °C for 1 hour using box furnace.

Fig. 1. Schematic representation for IDE fabrication with TiO₂ nanoparticles thin film. (a) Bare Si₃N₄ wafer as the substrate of the device. (b) 5 nm Cr layer was deposited on the substrate to act as an adhesive layer between the Au and the substrate. (c) 45 nm Au layer was then deposited on top of the Cr layer to serve as the electrode material. (d) Photoresist was spin-coated onto the Au layer to function as a pattern-transfer medium. (e) The IDE pattern was transferred from a chrome mask onto the substrate. (f) The IDE pattern was developed, rinsed, and dried. (g) Unprotected Au regions are etched away to define the final interdigitated electrode geometry. (h) TiO₂ thin films were deposited onto the interdigitated area to function as the sensing area. (i) Surface functionalization with APTES introduces amine groups (NH₂) that enable covalent immobilization of anti-HSA antibodies. (j) Subsequent binding of HSA antigens completes the sensing interface, where the antigen–antibody interaction is transduced into an electrical signal by the TiO₂ IDE platform.

2.4. Surface functionalization and characterization

To facilitate the interaction between the inorganic TiO₂ surface and the biological sample, the surface of the TiO₂-based IDE was functionalized with (3-aminopropyl)triethoxysilane (APTES), introducing amine groups for effective antibody immobilization. A 24% APTES solution was prepared by mixing 60 µL of APTES (99%) with 190 µL of deionized water, followed by vortex mixing to ensure homogeneity. A 10 µL aliquot of this solution was then drop-cast onto the sensing area of the device and left for 30 min at room temperature to allow proper surface modification.

Each stage of surface modification and bio-recognition was verified through FTIR spectroscopy by identifying the characteristic functional groups. The structural and morphological properties of the TiO₂ layer were also characterized using XRD and FESEM, respectively, to ensure surface quality and uniformity prior to functionalization.

2.5. HSA detection via immunosensing and electrical characterization

HSA detection was performed through electrical analysis using the TiO₂ nanoparticle-based IDE biosensor, with I–V measurements recorded via a Keithley 236 Source-Measure Units (SMU). The sensing mechanism utilized in this study is based on immunosensing, specifically the binding interaction between immobilized HSA antibodies and their corresponding HSA antigens. This antibody–antigen interaction induces changes in the electrical properties of the sensor, which were quantified through I–V analysis.

For antibody immobilization, 10 µL of 5 mg mL⁻¹ HSA antibody was deposited onto the APTES-functionalized nanostructured TiO₂ surface and incubated at room temperature. The HSA antigen stock was prepared by dissolving 5 mg of crystallized human serum albumin powder in 100 µL of deionized water. Antigen solutions of varying concentrations were prepared by serial dilution from HSA antigen stock using deionized water and mixed with a vortex, spanning a wide concentration ranging from 10 µg mL⁻¹ down to 1 pg mL⁻¹. To allow antigen–antibody binding, 10 µL of the antigen solution was drop-cast onto the sensing area.

For reusability testing, the used sensor surface was washed with heated deionized water (90 °C), and the water was blown off before drying on a hotplate at 150 °C for 20 min. This cleaning cycle was repeated three times to regenerate the surface prior to reuse in subsequent HSA detection experiments.

2.5.1. Design rationale and sensing mechanism

The structural architecture and the molecular sensing mechanism of the developed platform are summarized in Fig. 2. The choice of a Si₃N₄ substrate provides a chemically resistant and high-density diffusion barrier, ensuring device longevity during repeated cleaning cycles. The gold IDEs offer high electrical conductivity and chemical inertness, maintaining stable performance in ionic environments.

Fig. 2. Schematic of the TiO₂-IDE surface biosensor architecture and molecular binding interface. The diagram illustrates the layered configuration from the Si₃N₄ substrate and gold IDEs to the functionalized TiO₂ thin film. The upper section highlights the sequential surface modification (APTES and HSA-antibody) and binding of HSA antigens.

The detection mechanism of the TiO₂-IDE biosensor is based on the sequential functionalization of the TiO₂ surface to detect HSA capture, as illustrated in Fig. 2. The process occurs in four primary stages:

2.5.2. Surface hydroxylation

Upon exposure to moisture or ambient conditions from the TiO₂ solution, the TiO₂ nanoparticle surface is populated with hydroxyl groups (–OH), referred to as titanol groups (Ti–OH). These groups serve as the active sites for subsequent silanization.

2.5.3. Silanization via APTES

The TiO₂ surface is functionalized with APTES. During this stage, the triethoxysilane groups of APTES undergo hydrolysis and a subsequent condensation reaction with the Ti–OH groups, forming stable Si–O–Ti covalent bonds.²⁷ This leaves the terminal amino groups (–NH₂) of the APTES molecule free and oriented away from the surface.

2.5.4. Antibody immobilization

The HSA antibodies are immobilized onto the APTES-modified surface. This occurs primarily through the formation of covalent amide bonds (–CO–NH–) between the carboxyl groups (–COOH) of the antibody and the free amine groups (–NH₂) of the APTES.²⁸ Additionally, secondary interactions such as electrostatic attraction and hydrogen bonding between the amine groups and the antibody's hydroxyl groups further stabilize the interface.

2.5.5. Antigen–antibody interaction

The specific detection of HSA occurs through the formation of a stable immunocomplex at the TiO₂–Ab interface. The HSA antigens bind to the immobilized antibodies via highly specific non-covalent interactions, including hydrogen bonding, van der Waals forces, and hydrophobic interactions.²⁹ The molecular details of these interactions, specifically the key residues involved across different pH levels (0, 7, and 14), are systematically documented in Tables 3, 4, and S1. These interactions induce a localized change in charge density, which serves as the primary mechanism for the observed electrical transduction. This binding event alters the charge distribution at the TiO₂ interface, resulting in a measurable change in the electrical I–V characteristics of the IDE platform.

Table 3. Current responses of the TiO₂ IDE biosensor at 1.8 V for fresh and reused measurements at 1 ng mL⁻¹ and 1 µg mL⁻¹ HSA, showing percentage signal loss after regeneration.

Sample	Current (µA)		Reuse current (µA)		Percentage current drop (%)
	1 ng mL−1	1 µg mL−1	1 ng mL−1	1 µg mL−1	1 ng mL−1	1 µg mL−1
1	16.05	13.3	0.22	0.19	98.69	98.57
2	18.86	15.3	4.59	4.25	75.69	72.22
3	18.51	16.36	5.76	4.9	68.88	70.05
4	16.37	14.76	5.89	4.96	64.02	66.4
5	18.51	16.14	5.57	5.39	69.91	66.6
Mean	17.66	15.17	4.41	3.94	75.44	74.77

Amino acid residues involved in the antibody–antigen binding interface and the corresponding binding energies obtained from HDOCK and interaction energy (IE) analyses (kcal mol⁻¹).

Binding residues in 3 Å interface
Antibody
Heavy chain (VH): ARG30, SER52, SER53, SER54, LYS105
Light chain (VL): ASP70, TYR92, PHE94
Albumin
LYS4, GLU227, GLU230, TYR263, GLU266, ASN267, ASP308

pH	Energy (kcal mol−1)
	HDOCK	IE
0	−160.94 ± 0.64	−168.94
7	−177.06 ± 0.63	−180.90
14	−175.94 ± 0.64	−176.66

3. Results and discussions

3.1. Surface characterization of TiO₂ film

Field Emission Scanning Electron Microscopy (FESEM) analysis was conducted to examine the surface morphology of the TiO₂ film and to estimate the size distribution of the TiO₂ nanoparticles.³⁰Fig. 3a presents the FESEM image of the TiO₂ film after annealing at 600 °C. The image reveals the formation of uniformly distributed spherical-shaped particles across a nanocrystalline grain structure, a typical characteristic of TiO₂ nanoparticles under the given annealing conditions.²⁶ The uniform distribution of spherical particles is attributed to the triple-layer spin coating process employed to achieve a consistent and homogeneous surface. This multi-layer coating approach was adopted to increase the film thickness and ensure smooth electron flow. The enhanced electron flow is critical for biosensing applications, as it facilitates efficient charge transfer during biosensing reactions, which is the primary focus of this study.³¹

Fig. 3. Comprehensive characterization of the TiO₂-based IDE biosensor. (a) FESEM image of the annealed TiO₂ thin film, showing uniformly distributed spherical nanoparticles with nanocrystalline features. (b) XRD pattern confirming the anatase phase of TiO₂ with characteristic diffraction peaks. (c) FTIR spectra representing progressive surface modification and biomolecular interaction: (i) bare TiO₂ surface, (ii) APTES-functionalized TiO₂, (iii) after antibody (AB) immobilization, and (iv) after antigen (AG) binding. These results confirm the successful chemical modification of the sensor surface and effective bio-recognition binding.

As reported by Tonny (2019), an increase in film thickness correlates with a reduction in sheet resistance, thereby enhancing the overall conductivity of the thin film.³² Based on the FESEM analysis, the average size of TiO₂ nanoparticles was estimated to be approximately 20 nm. This nanoscale size is highly advantageous for biosensor applications, as it provides a high surface-to-volume ratio, which promotes effective biomolecule immobilization and enhances sensor sensitivity.³³

3.2. X-ray diffraction (XRD) analysis on TiO₂ film

XRD was conducted to analyze the crystal structure and estimate the crystallite size of the TiO₂ thin film.³⁴ The measurements of the samples were conducted using a Bruker D8 Advance X-ray diffractometer with CuKα radiation (45 kV, 20 mA, k = 0.15406 nm and (20 < 2θ < 60)). Fig. 3b illustrates the XRD pattern of the TiO₂ wafer annealed at 600 °C for 1 hour. In the spectrum, peaks are labeled as “A” for anatase structure, “R” for rutile structure, and “Si” for silicon substrate.

Based on the XRD results, five diffraction peaks were observed at 25.3°, 37.8°, 48.1°, 54.1°, and 55.1°, corresponding to the (101), (103), (200), and (220) planes. The XRD patterns were compared with the JCPDS reference data of anatase and rutile (JCPDS cards 21-1272 and 001-1292, respectively). The observed anatase phase (tetragonal) shows peaks that correspond with the standard XRD pattern values for TiO₂ (JCPDS data card (21-1272)).^17,35 Notably, the anatase phase, particularly with the (101) orientation, exhibits as the dominant peaks for TiO₂. This is consistent with previous studies reporting anatase as the preferred phase for biosensing applications due to its higher surface area and photocatalytic properties.

In addition to determining the crystal structure of the TiO₂ wafer, XRD analysis was used to estimate the average crystallite size (particle diameter) by using the Scherrer formula:

(D = 0.9λ/B cos θ) 1

where D is the particle diameter, λ is the wavelength of X-ray radiation at 1.5406 Å, B is the full width at half maxima of the diffraction peak and θ is the diffraction angle.³⁶

Based on this calculation, the estimated average particle diameter of TiO₂ nanoparticles is found to be 16 nm. This result aligns well with the nanoparticle size observed in the FESEM analysis, further validating the nanocrystalline nature of TiO₂ film. The small crystallite size is beneficial for biosensor performance, as it increases surface-to-volume ratio and promotes efficient charge transfer, both of which are critical for enhancing sensitivity and response time in electrical biosensing.³³

3.3. Fourier transform infrared (FTIR) analysis

After preparing the TiO₂ thin film, surface functionalization was carried out using (3-aminopropyl)triethoxysilane (APTES) to introduce a reactive contact layer onto the TiO₂ surface. This process incorporates free amino groups (NH₂) from APTES, which facilitates the immobilization of biological components such as HSA.³⁷ Following this, the sample underwent FTIR characterization to identify the presence of functional groups and their influence on the surface properties of the TiO₂ wafer. Fig. 3c illustrates the FTIR spectra of the various sample tested:

(i) Bare TiO₂ – to establish the baseline spectrum and identify inherent TiO₂ vibrational modes.

(ii) TiO₂ functionalized with APTES (TiO₂ + APTES) – to detect silane coupling and confirm amine group introduction.

(iii) TiO₂ functionalized with APTES and antibody (TiO₂ + APTES + AB) – to verify successful antibody immobilization.

(iv) TiO₂ functionalized with APTES, antibody, and antigen (TiO₂ + APTES + AB + AG) – to observe additional shifts indicating antibody–antigen complex formation.

In panel (i), the FTIR spectrum of bare TiO₂ shows a prominent peak around 1000 cm⁻¹, which is attributed to the stretching modes of Ti–O bonds.³⁸ Additionally, a peak around 2300 cm⁻¹ is observed, suggesting hydrogen bonding associated with oxygen-containing groups on the TiO₂ surface.³⁹

In panel (ii) (TiO₂ + APTES), the spectrum shows new peaks associated with the APTES molecule. The Ti–O–Si stretching vibration is still evident near 1000 cm⁻¹ due to silane bonding to the oxide surface.⁴⁰ A new peak at 2900 cm⁻¹ corresponds to asymmetric –CH₂ stretching from the propyl chain of APTES.¹⁷ Additionally, two distinct peaks around 1400 cm⁻¹ and 1500 cm⁻¹ belong to the NH₂ scissor vibrations, indicating the presence of the NH₂ terminal group of APTES.⁴¹

In panel (iii), the FTIR spectrum of the TiO₂ surface functionalized with APTES and HSA antibody exhibits new absorption bands in the range of 1500–1650 cm⁻¹, which can be attributed to protein-specific vibrations. These include the Amide I band (primarily C O stretching, around 1650 cm⁻¹) and amide II band (C–N stretching and N–H bending, around 1550 cm⁻¹), indicating successful antibody immobilization.

Upon the addition of the HSA antigen in panel (iv), these protein-related bands become more pronounced, particularly the Amide I band at 1651 cm⁻¹. This increase in intensity confirms further protein binding, consistent with the formation of an antibody–antigen complex on the TiO₂ surface. The presence and amplification of these characteristic amide peaks across panels (iii) and (iv) clearly demonstrate the progressive biofunctionalization of the sensor surface.⁴² This observation aligns with Usoltsev and team, who also identified a peak at 1650 cm⁻¹.⁴³ The appearance of these peaks confirms the successful immobilization of APTES, HSA antibody, and HSA antigen on the TiO₂ surface, as evidenced by the detection of their respective functional groups.

In contrast, panel (v) shows the FTIR spectrum of the regenerated TiO₂ electrode after the cleaning process, which closely resembles that of bare TiO₂. The disappearance of amide-related bands, particularly the C O stretching vibration of Amide I around 1650 cm⁻¹ and the N–H bending/C–N stretching vibrations of Amide II around 1550 cm⁻¹, indicates effective removal of protein residues, confirming that the surface was reset to its initial state. This restoration demonstrates that the cleaning protocol successfully eliminates biomolecular contaminants while preserving the integrity of the TiO₂ substrate, thereby supporting the sensor's reusability.

3.4. Electrical measurement of TiO₂ nanoparticle-based IDE

This TiO₂ nanoparticle-based biosensor operates as an amperometric immunosensor, relying on the specific binding interaction between HSA antibodies and their target antigens. These binding events induce measurable changes in current, with the magnitude of the response correlating directly to the concentration of the target analyte.⁴⁴

Fig. 4 shows the current–voltage (I–V) measurements recorded using Keithley 236 Source-Measure Units (SMU). Amperometric detection was conducted for HSA concentrations ranging from 10 µg mL⁻¹ down to 1 pg mL⁻¹, under an applied voltage sweep of 0 V to 3 V. As depicted in Fig. 4, the current begins to increase gradually as the voltage rises. Notably, between 1.5 V to 2.5 V range, there is a marked shift in current, indicating that the response for each concentration of HSA becomes evident. This voltage range is critical, as it demonstrates the sensor's ability to differentiate varying HSA concentrations. The voltage of 1.8 V was selected as the optimal reference point due to the consistent and stable current responses across different HSA concentrations using five replicates, enabling effective sensitivity analysis.

Fig. 4. I–V and calibration characteristics of the TiO₂-based IDE biosensor for HSA detection using Sample 1. (a) I–V curves recorded at different HSA concentrations showing a concentration-dependent decrease in current, where the current begins to increase gradually as the voltage rises. (b) Calibration curve showing the change in current (ΔI) at a fixed voltage of 1.8 V versus HSA concentration, demonstrating a linear increase with sensitivity of 0.661 µA per decade. (c) Relative change in current (ΔI/I₀) versus HSA concentration across five samples, used to calculate the limit of detection (LOD). The biosensor exhibited a strong linear correlation (R² = 0.98), with an LOD of 0.33 pg mL⁻¹ based on five replicate measurements. (d) Reproducibility test for two HSA concentrations (1 pg mL⁻¹ and 1 ng mL⁻¹) across five independent measurements, showing consistent responses with low variation across five independently fabricated devices.

As shown in Fig. 4a, the current responses exhibit an inverse correlation with HSA concentrations. This trend is attributed to steric hindrance and electrostatic repulsion introduced by high concentrations of HSA molecules. As more HSA molecules bind to the surface, molecular crowding forms a physical barrier that limits charge transfer across the TiO₂ interface. At the same time, the accumulation of negatively charged HSA molecules generates electrostatic repulsion that further suppresses current flow. At elevated concentrations, saturation of antibody binding sites contributes to a plateau in the response. These effects align with established findings in immunosensing systems, where surface saturation and electrostatic interactions can diminish the signal at elevated analyte levels.⁴⁵

The inverse relationship between current and HSA concentration can be explained by a combination of steric hindrance, electrostatic effects, and site-saturation phenomena. At higher antigen coverage, bound protein layers create physical/steric barriers that impede charge transfer at the transducer surface. This steric-hindrance sensing mechanism is well documented in electrochemical protein assays.⁴⁶ Additionally, adsorption of charged proteins alters the local surface potential and screening environment, which modulates the interfacial electric field and further suppresses current; exploiting or accounting for these electrostatic interactions is a common design consideration in sensitive biosensors.⁴⁷ Besides that, at sufficiently high analyte concentrations antibody binding sites become saturated and additional antigen produces progressively smaller changes in signal, producing the characteristic plateau in the dose–response curve. This saturation behavior and its effect on signal linearity are described in immunoassay and biosensor literature.⁴⁸

The ability of the biosensor to detect and respond to HSA concentrations across a wide dynamic range, from picograms to micrograms per milliliter, underscores its high sensitivity. The observed decrease in current with increasing HSA concentration not only demonstrates the sensor's responsiveness to biomolecular interactions but also confirms its precision in quantifying analyte levels. This performance highlights the biosensor's suitability for applications requiring low detection limits and accurate quantification of biomarkers, making it a promising tool for sensitive and reliable immunosensing.

Fig. 4b displays the calibration curve of the biosensor at various HSA concentrations, measured at a fixed voltage of 1.8 V. A linear increase in the change in current (ΔI = I_o – I) was observed with rising HSA concentration, with a sensitivity of 0.661 µA per decade determined from the slope of the calibration curve. This linearity and sensitivity highlight the sensor's ability to provide reliable, concentration-dependent responses, which are competitive with or superior to those of similar devices reported in the literature.⁴⁹

To account for inherent device-to-device baseline variation and common-mode noise, the analytical performance was further evaluated using the relative change in current (ΔI/I_o) at various HSA concentrations, as shown in Fig. 4c. Normalizing the response to each individual sensor's baseline is a standard practice in sensing platforms to ensure reproducibility across different sensor chips, as exemplified by the normalized current measurements reported in ref. 50. The biosensor exhibited a linear response across the tested concentration range, with a coefficient of determination (R²) of 0.98, indicating a strong correlation between signal and analyte concentration. The LOD, calculated using five replicates from the calibration curve, was determined to be 0.33 pg mL⁻¹.⁵¹ Relative to other studies, the LOD and sensitivity range of this device demonstrate favorable performance, falling within a promising range for practical applications. The table below provides a comparative analysis of the performance of this immunosensor with previous studies on HSA detection (Table 1).

Table 1. Performance evaluation of HSA detection in comparison to existing immunosensor platforms.

HSA biosensor	Measurement range (g mL−1)	LOD (g mL−1)	Key limitation and features	Source
ZnO nanorods immune-FET	0.01 – 100 µ	9.81 × 10−9	FET-based transduction; requires gate control; regeneration not reported	49
SWCNT network immune-FET	1 f – 10 p	0.47 × 10−15	Ultra-low LOD; complex biasing requirements; regeneration not reported	52
CNT-based immuno-FET	0.07 – 70 µ	18.6 × 10−9	FET-based architecture; modified with BCG; regeneration not reported	53
Screen-printed carbon electrodes	2.5 – 500 µ	1.55 × 10−6	Uses chemical linkers (EDC-NHS) functionalization; high LOD; regeneration not reported	54
TiO2 nanoparticle IDE	1 p – 10 µ	0.33 × 10−12	Simple I–V transduction; cleaning-based regeneration demonstrated	This work

To evaluate reproducibility, the performance consistency of the TiO₂-based IDE biosensor was assessed using four additional independently prepared samples, each tested under identical conditions. As shown in Fig. 5, all samples exhibited similar I–V profiles, with current values consistently decreasing with rising HSA concentration. This trend was observed across the entire concentration range from 1 pg mL⁻¹ to 10 µg mL⁻¹, confirming that the biosensor responds reliably regardless of sample variation. As shown in the calculated RSD values (Table 2), the sensor exhibits higher variation (54.79% and 59.81%) at 1 pg mL⁻¹ and 10 pg mL⁻¹, respectively. This is expected at concentrations near the noise level, as the ΔI (measured current minus blank current) is highly sensitive to minor baseline fluctuations in this region.⁵⁵ However, these low-concentration signals remain consistent with the overall linear regression trend (R² = 0.98), confirming that the sensor maintains its detection capability at the picogram level.

Fig. 5. I–V curves for consistency testing of the TiO₂-based IDE biosensor across four independently fabricated samples: (a) Sample 2, (b) Sample 3, (c) Sample 4, and (d) Sample 5. All devices displayed the same trend of decreasing current with increasing HSA concentration, highlighting the reproducibility of the fabrication process and robustness of sensor performance.

Table 2. Comprehensive reproducibility data and RSD analysis (n = 5).

HSA concentrations	ΔI (Sample 1) (µA)	ΔI (Sample 2) (µA)	ΔI (Sample 3) (µA)	ΔI (Sample 4) (µA)	ΔI (Sample 5) (µA)	Average ΔI (µA)	Standard deviation (µA)	RSD (%)
1 pg mL−1	0.16	0.82	0.49	0.5	0.3	0.454	0.24876	54.79
10 pg mL−1	0.19	0.95	1.47	1.06	0.5	0.834	0.49883	59.81
1 ng mL−1	2.02	1.77	2.26	2.2	2.05	2.06	0.19066	9.25
10 ng mL−1	2.23	3.2	2.51	2.34	3.71	2.798	0.63417	22.67
1 µg mL−1	4.77	5.33	4.41	3.81	4.42	4.548	0.55697	12.25
10 µg mL−1	5.16	4.96	4.72	5.03	5.82	5.138	0.41342	8.05

To assess the sensor's performance consistency, we have emphasized the sensor's primary quantitative range at 1 ng mL⁻¹, where the RSD remains within 8.15% to 22.80%. Within this range, reproducibility was specifically examined at 1 ng mL⁻¹ and 1 µg mL⁻¹ HSA concentrations, as shown in Fig. 4d. The calculated relative standard deviation (RSD) values were 9.25 ± 0.25% and 12.26 ± 0.56%, respectively, indicating acceptable reproducibility for antibody–antigen assays and are within the typical range reported for immunoassay.⁵⁶ This confirms that the platform is highly reproducible within its primary quantitative range while maintaining a high degree of sensitivity.

The minimal variation in current values across the four additional samples highlights the robustness of the biosensor's design and fabrication process, suggesting consistent sensor surface quality and functionalization. Although minor fluctuations in current were observed, they remained within acceptable limits, indicating strong repeatability and reliable response across the entire HSA concentration range. This reproducibility is essential for clinical and point-of-care applications, where accurate, repeatable measurements are critical. Overall, the results confirm that the TiO₂-based IDE biosensor delivers not only high sensitivity but also consistent and reliable performance.

The regeneration of the device was assessed by rinsing it with heated deionized water at 90 °C and drying it on a hot stirrer at 150 °C for approximately 20 minutes. While these temperatures denature the biological layer, the high thermal stability of the TiO₂ nanoparticle film and Si₃N₄ substrate allows the inorganic platform to be cleaned and re-functionalized without structural degradation. As illustrated in Fig. 6, the current response of the bare TiO₂ sensor consistently adhered to the expected trend even after reuse. In the reuse sample 1, the biosensor maintained a linear detection ranges from 1 pg mL⁻¹ to 10 µ g mL⁻¹, with a linearly decreasing current response observed as HSA concentration increased, consistent with the trend seen in the fresh sample. Additionally, the I–V profiles also followed the same trend, showing a linear increase in current with increasing applied voltage. To further evaluate platform stability, the device was tested using five reuse samples, all of which exhibited similar I–V profiles and concentration-dependent current responses, as depicted in Fig. 6.

Fig. 6. I–V curves for reusability testing of the TiO₂-based IDE biosensor across five reuse samples: (a) reuse Sample 1, (b) reuse Sample 2, (c) reuse Sample 3, (d) reuse Sample 4, and (e) reuse Sample 5. The regenerated sensors consistently followed the expected trend of current reduction with rising HSA concentration, confirming functional reusability despite some reduction in absolute current values.

To provide quantitative insight, current responses were compared at a fixed voltage of 1.8 V for both fresh and reused devices. At 1 ng mL⁻¹ HSA, the mean current decreased from 17.66 ± 1.34 µA (fresh) to 4.41 ± 2.40 µA (reuse), while at 1 µg mL⁻¹ it decreased from 15.17 ± 1.23 µA to 3.94 ± 2.13 µA. This corresponds to an average performance decline of ∼75% (1 ng mL⁻¹: 75.44% ± 12.2%; 1 µg mL⁻¹: 74.77% ± 12.1%), as summarized in Table 3. Although a reduction in current magnitude is observed, the drop remains within the expected operational range of regenerated biosensors and does not compromise sensing performance. The preserved linear detection trend, wide concentration range, and consistent I–V characteristics confirm that the device continues to function reliably. These findings underscore the biosensor's structural durability and its ability to maintain practical platform regeneration for HSA detection.

3.5. Prediction of antibody–antigen binding complex interaction at different pH

Molecular docking using HDOCK⁵⁷ was performed to support and interpret the I–V data, providing molecular-level insights into the stability and binding interactions between the antibody and its antigen. Crystal structure of a complex between human serum albumin and the antibody Fab SL335 (PDB ID: 6M58) was retrieved from the Protein Data Bank.⁵⁸ Prior to dock the antibody and the antigen, the structures were pre-processed by removing crystallographic water molecules, adding missing residues, and performing structural cleanup using Discovery Studio. Hydrogens were added, and charges were assigned using the AM1-BCC method in UCSF Chimera,⁵⁹ ensuring accurate electrostatic potential representation particularly important for simulating pH-dependent effects. Minimization was then performed using the Amber ff14SB force field, initially with the steepest descent method (0.02 angstroms for 100 steps), followed by further refinement using the Conjugate Gradient method.

The binding interactions between antibody and antigen are highly sensitive to the protonation states of key amino acid residues, which are governed by the surrounding pH environment. Our computational analysis predicts distinct variations in antibody–antigen complex stability across acidic (pH 0), physiological (pH 7), and basic (pH 14) conditions, primarily driven by changes in electrostatic charge and molecular complementarity. The protonation states of key amino acid residues were determined using PROPKA, which predicts pK_a values based on protein structure and environment.⁶⁰ The albumin were re-docked to the antibody and evaluated their binding affinity at pH 7, and the resulting binding pattern, along with the associated interactions, closely resembled those observed in the crystal structure. The best-docked structure, exhibiting the lowest binding energy, was superimposed with the X-ray complex. The close alignment of key interacting residues confirmed the structural accuracy of the docking model and supported the reliability of the predicted antibody–antigen interface as in Fig. 7. Protein–protein interactions of the best-docked structure in different pH were analysed by calculating the interaction energy using the formula: interaction energy (IE) = energy of complex − energy of antibody − energy of albumin, where all energy terms were obtained using the AMBERff14SB force field. “Albumin here refers to the antigen (HSA) involved in the antibody–antigen complex. The residues involved in the binding interfaces within 3.0 Å were listed in Table 4.

Fig. 7. Superimposition of docked (yellow) and the crystal structures (blue) of human serum albumin and the antibody.

The protonation states of the listed amino acids in antibody–albumin complex at the binding interface vary across pH 0, 7, and 14, which in turn influence their binding interactions. At pH 0 (highly acidic), basic residues such as ARG30 and LYS105 in the heavy chain of antibody, and LYS4 in albumin, are fully protonated and positively charged, enhancing electrostatic attraction with negatively charged groups. Acidic residues like ASP70 in light chain of antibody, and GLU227, GLU230, GLU266, and ASP308 in albumin are protonated and neutral, reducing repulsion. Tyrosine residues (TYR92 in antibody, TYR263 in albumin), as well as serine (SER52, SER53, SER54) and asparagine (ASN267) in antibody, remain mostly neutral due to their higher pKa values. At physiological pH 7, acidic residues are generally deprotonated and negatively charged, while basic residues remain protonated and positively charged, allowing strong electrostatic complementarity that stabilizes antibody–antigen binding. Neutral residues like serine, asparagine, and phenylalanine remain unchanged. At pH 14 (highly basic), acidic residues remain negatively charged, but basic residues tend to lose protons and become neutral, which can weaken electrostatic interactions. Tyrosine side chains may also deprotonate, gaining negative charges.

The simulation was first performed at pH 7, which closely approaches real physiological conditions and reflects our experimental environment (HSA diluted in deionized water, pH ∼7). At this pH, the antibody–antigen complex exhibited the strongest binding energy of −177.06 ± 0.63 kcal mol⁻¹. Additional simulations were conducted at both pH 0 and pH 14, yielding slightly different binding energies of −160.94 ± 0.64 kcal mol⁻¹ and −175.94 ± 0.64 kcal mol⁻¹, respectively. The HDOCK and interaction energy (IE) scores show a consistent trend, supporting the reliability of the predicted binding interactions. Unlike the HDOCK score, which is based on a hybrid docking algorithm and empirical scoring, the IE reflects the sum of van der Waals, electrostatic, and solvation energies between the antibody and antigen in their bound state. This energy is expressed in kcal mol⁻¹, with more negative values indicating stronger and more stable binding interactions. Together, the HDOCK and IE values offer complementary insights, with the IE providing a more rigorous, physics-based evaluation of the binding complex.

The variation in antibody–antigen binding interaction across different pH values remains within a narrow range and may not be statistically significant. These pH-dependent protonation changes affect the strength and nature of binding interactions, with optimal stability and binding typically observed near physiological pH due to balanced charges promoting strong molecular recognition between antibody and antigen. In fact, highly acidic conditions (e.g., pH 0) may superficially enhance binding due to increased protonation and electrostatic complementarity. However, such environments typically induce protein denaturation. Excessive protonation of acidic residues disrupts native electrostatic interactions and tertiary structure, ultimately compromising the stability of antibody–antigen complexes.^61–63 Importantly, these docking results directly support the experimental observations. The stronger and more stable binding predicted at pH 7 aligns with the robust and reproducible sensor responses were obtained when operating in deionized water (pH ∼7) This computational–experimental alignment follows similar strategies reported in biosensor studies, where docking analyses were used to rationalize binding conditions and guide antibody immobilization approaches.⁶⁴

3.6. Binding of antibody–albumin complex on APTES-functionalized TiO₂ IDE electrode sensor surface

To construct a biosensor model exactly meet our experimental model, the TiO₂ substrate bottom platform was introduced. TiO₂ in anatase phase (tetragonal, space group I4₁/amd) with titanium and oxygen atoms occupying 4b and 8e Wyckoff positions, respectively was used. The Ti–O octahedral arrangement forms a puckered structure along the [001] direction, increasing surface area and providing active sites for molecular attachment. This TiO₂ layer was integrated with a gold (Au) (111) layer and an underlying silicon nitride (Si₃N₄) base, forming a (15 × 15 × 1) supercell representing the interdigitated electrode (IDE) surface, as illustrated in Fig. 8: (a) side view and (b) top view. The Au and Si₃N₄ structures were modelled using their respective crystallographic parameters.^65,66

Fig. 8. The schematic functionalised TiO₂ IDE with APTES (a) top view and (b) side view.

The surface was functionalized with an APTES (3-aminopropyltriethoxysilane) linker molecule, which was geometry-optimized using the B3LYP/6-311++G (d,p) level of theory in Gaussian.⁶⁷ This ensured accurate charge distribution and bonding parameters for subsequent docking. The antibody structure was similarly prepared, and both systems were assigned partial atomic charges using the AM1 semi-empirical method to reflect local electrostatics during interaction.

The optimized APTES molecule was positioned atop the TiO₂ IDE surface, with its terminal oxygen atoms oriented toward the underlying silicon nitride layer, as illustrated in Fig. 8. This APTES-functionalized TiO₂ IDE model served as the structural basis for molecular docking simulations with the antibody. The model accurately represents the biosensor's configuration, incorporating both the APTES linker and the TiO₂ IDE surface. The docking simulations were carried out using the HDOCK server, with all interactions modelled under neutral pH conditions to reflect the experimental environment. Further calculation on interaction energy result was tabulated as shown in Table 5.

Table 5. HDOCK binding scores and interaction energy (IE) values (kcal mol⁻¹) for the antibody–albumin complex on TiO₂/Au/Si₃N₄ surface with and without APTES functionalization at pH 7.

Model		Energy (kcal mol−1)
Complex	Surface	HDock	IE
Antibody–albumin	TiO2/gold/Si3N4	−682.74 ± 1.00	−690.65
Antibody–albumin	APTES/TiO2/gold/Si3N4	−696.87 ± 1.00	−720.02

As shown in Table 5, the antibody–albumin complex demonstrates stronger binding energy on the APTES-functionalized TiO₂ IDE surface compared to the unmodified surface, with an HDOCK score of −14.13 ± 1.00 and an interaction energy (IE) of −29.37 kcal mol⁻¹. These findings indicate favorable antibody immobilization under physiological conditions, supporting the sensor's robustness and operational reliability. More negative binding free energies are widely associated with enhanced molecular stability, reflecting stronger and more favorable interactions. In a comparable case, studies on SARS-CoV-2 antibodies revealed that glycosylation of the receptor-binding domain (RBD) and antibody S309 led to increased binding stability, as evidenced by a more negative binding free energy.⁶⁸

Additionally, the binding energy between the antibody and antigen was favorable, further confirming the strength of their interaction. This dual enhancement at both the surface interface and molecular recognition levels ensures reliable biosensor performance by reinforcing stability and sensitivity under physiological conditions. Notably, the TiO₂ IDE surface significantly improves antibody binding stability through enhanced electrostatic compatibility, especially at pH 7, where surface charge balance is achieved. The TiO₂ layer, functionalized with APTES, creates a conducive environment for antibody immobilization, strengthening molecular interactions and maintaining the structural integrity of the antibody–antigen complex.

This study underscores the stability of antibody interactions with both biological targets and engineered surfaces under physiological pH. Molecular docking simulations demonstrated that at pH 7, the APTES-modified TiO₂ IDE surface supports strong and stable antibody binding due to improved electrostatic complementarity. Surface modelling further confirmed the feasibility of robust antibody attachment, highlighting the critical importance of stable antibody–antigen interactions in the design of effective, real-world biosensing applications.

4. Conclusion

In conclusion, the nanostructured TiO₂-based IDE biosensor was successfully characterized in terms of surface morphology, functional group identification, surface functionalization, and electrical conductivity. The biosensor exhibited a consistent and sensitive electrical response to varying concentrations of human serum albumin (HSA), ranging from 10 µg mL⁻¹ down to 1 pg mL⁻¹, as indicated by measurable current shifts in the I–V characteristics. This sensitivity is attributed to the nanoscale surface area and efficient electrical conductivity of the TiO₂ nanoparticle-based IDE, which enhance the antigen–antibody binding interaction. The sensor's performance was validated across five independently fabricated devices and further confirmed through post-cleaning measurements, which showed minimal variation – highlighting the biosensor's repeatability and reusability. Molecular docking simulations provided detailed insights into the antibody–antigen binding mechanism by considering the different protonation states across varying pH levels. The results demonstrate that the APTES-functionalized nanostructured TiO₂ surface significantly enhances antibody binding stability, likely due to improved electrostatic complementarity under physiological conditions. Additionally, the antibody–antigen binding energy was favourable, further confirming the strength of their interaction. This dual enhancement at both the sensor surface and molecular recognition levels ensures robust and reliable biosensor performance, supporting the sensor's stability and sensitivity in physiological environments. The TiO₂-based IDE biosensor developed in this study is cost-effective, miniaturized, and well-suited for point-of-care (POC) applications, aligning with current trends in portable diagnostic technologies.

Author contributions

Mohamad Nizar Hadi Mohamad Nassir (conceptualization: lead; data curation: lead; formal analysis: lead; investigation: lead; methodology: lead; project administration: lead; visualization: lead; writing – original draft: lead; writing – review & editing: lead). Sharipah Nadzirah Syed Ahmad Ayob (supervision: supporting; writing – review &editing: supporting). Azrul Azlan Hamzah (funding acquisition: lead; resources: equal; supervision: supporting). Ahmad Ghadafi Ismail (resources: equal; supervision: supporting). Hung Wei Yu (data curation: supporting; resources: equal). Edward Yi Chang (data curation: equal; resources: equal). Vannajan Sanghiran Lee (software: equal; writing – review & editing: supporting). Muhammad Zhafran Zakaria (software: equal). Keat Hoe Yeoh (data curation: supporting). Yee Hui Robin Chang (data curation: supporting). You-Chen Weng (data curation: supporting; resources: equal). Chang Fu Dee (resources: lead; supervision: lead; writing – review & editing: lead).

Conflicts of interest

There are no conflicts to declare.

Data availability

The data supporting the findings of this study are provided in the supplementary information (SI) file associated with this article. Supplementary information is available. See DOI: https://doi.org/10.1039/d5ra08974c.