'All In One' SARS-CoV-2 variant recognition platform: Machine learning-enabled point of care diagnostics

Abstract

Point of care (PoC) devices are highly demanding to control current pandemic, originated from severe acute respiratory syndrome Coronavirus 2 (SARS-CoV-2). Though nucleic acid-based methods such as RT-PCR are widely available, they require sample preparation and long processing time. PoC diagnostic devices provide relatively faster and stable results. However they require further investigation to provide high accuracy and be adaptable for the new variants. In this study, laser-scribed graphene (LSG) sensors are coupled with gold nanoparticles (AuNPs) as stable promising biosensing platforms. Angiotensin Converting Enzyme 2 (ACE2), an enzymatic receptor, is chosen to be the biorecognition unit due to its high binding affinity towards spike proteins as a key-lock model. The sensor was integrated to a homemade and portable potentistat device, wirelessly connected to a smartphone having a customized application for easy operation. LODs of 5.14 and 2.09 ng/mL was achieved for S1 and S2 protein in the linear range of 1.0-200 ng/mL, respectively. Clinical study has been conducted with nasopharyngeal swabs from 63 patients having alpha (B.1.1.7), beta (B.1.351), delta (B.1.617.2) variants, patients without mutation and negative patients. A machine learning model was developed with accuracy of 99.37% for the identification of the SARS-Cov-2 variants under 1 min. With the increasing need for rapid and improved disease diagnosis and monitoring, the PoC platform proved its potential for real time monitoring by providing accurate and fast variant identification without any expertise and pre sample preparation, which is exactly what societies need in this time of pandemic.

Keywords: COVID-19; Laser-scribed graphene; Machine learning; Point-of-care; SARS-CoV-2; Sensor.

Graphical abstract

Fig. 1

Representation of fabrication step of AuNPs-LSG sensor including (a–b) Scanning electron microscopy (SEM) images with corresponding elemental mapping images at a scale bar of 5 μm and 10 μm, (c) X-ray photoelectron spectroscopy (XPS) data containing C1s, O 1s, N 1s, S 2p, and Au 4f7 spectra of AuNPs-LSG sensor.

Fig. 2

Oxidation response of the AuNPs-LSG sensor with magnified DPV peaks corresponding (a, c) SARS-CoV-2 S1 and (b, d) S2 protein various concentrations with respect to LSG RE. The insets indicate the logarithmic relation of the oxidation current difference (ΔIox) and S1 and S2 protein concentration (logCS1 and logCS2). Histograms showing (e) the response of the AuNPs-LSG sensor towards different viruses including Influenza B (INFB) (5.59 %RSD), Human Rhinovirus (HRV) (10.5 %RSD), Respiratory Syncytial Virus (RSV) (4.93 %RSD), Influenza A (H1N1) (INFA-H1N1) (0.85 %RSD), Human Coronavirus 229E/NL63 (HCoV-229E/NL63) (1.02 %RSD), Adenovirus/Rhinovirus, (HAdV/HRV) (1.97 %RSD), SARS-CoV-2 B.1.1.7 (2.59 %RSD), B.1.351 variants (0.46 %RSD), SARS-CoV-2 without mutation (0.58 %RSD) and control (negative) sample (18.7 %RSD),; (f) the change in oxidation current for nasopharyngeal swab dilution percentages 25% (0.65 %RSD), 50% (1.03 %RSD), 75% (0.47 %RSD), 100% (4.76 %RSD). (Error bars: ± SD for n = 3 for different sensors). 5.0 mM [Fe (CN)6]3–/4– containing 0.1 M PBS and 0.1 M KCl was used as redox buffer at 50 mV/s scan rate.

Fig. 3

Histogram showing the change in oxidation current values (ΔIox) of 63 nasopharyngeal swabs collected from control (negative) patients (orange), COVID-19 positive patients without any mutation (blue), with B.1.351 variant (red), B1.1.7 variant (purple) and B.1.617.2 variant (green). (Error bars: ± SD for n = 3 for different sensors). 5.0 mM [Fe (CN)6]3–/4– containing 0.1 M PBS and 0.1 M KCl was used as redox buffer at 50 mV/s scan rate. RSD values are given in Table S6 for n = 3. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4

(a) The portable PoC device with a smartphone connection by a USB-C connection including sensor attachment. (b) DPVs of the AuNPs-LSG sensor showing ΔIox (the oxidation current change) after each modification and after detecting 200 ng/mL of SARS-CoV-2 S1 and S2 antigens. (c) DPVs showing AuNPs-LSG sensor response for the swabs obtained from COVID-19 positive (+) and negative (−) patients by (i) the portable PoC device and (ii) the commercial potentiostat. (Error bars: ± SD for n = 3 for different sensors). 5.0 mM [Fe (CN)₆]^3–/4– containing 0.1 M PBS and 0.1 M KCl was used as redox buffer at 50 mV/s scan rate.

Fig. 5

(a) The Dense Neural Network (DNN) architecture prepared using Edge Impulse IDE with EON Compiler. (b) Spatial representation of the dataset collected by measuring nasopharyngeal swabs of COVID-19 positive and negative patients.