'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.

Rapid Quantification of SARS-Cov-2 Spike Protein Enhanced with a Machine Learning Technique Integrated in a Smart and Portable Immunosensor

“…Moreover, if needed, the C code of the trained classifier can be generated and included in the device firmware to perform sample labelling directly on board. Compared to other solution presented in the literature [ 8 , 49 , 50 ], both the training and test phases can be processed autonomously (either on board or on the cloud platform), improving the ease of use and portability of the device, thus promoting its introduction in home and PoCT environments outside a clinical setting.…”

“…It is a portable wireless device that uses the Wi-Fi protocol to connect to a cloud environment for data processing and visualization. The main feature of this device is that, unlike the solutions presented in the literature [ 8 ], it does not need to rely on external devices (e.g., PC, tablets or smartphones) for operation. In addition to greatly simplifying the system architecture, this feature makes the device particularly suitable for PoCT applications.…”

“…The features offered by ML represent a great advantage in several fields, especially in relation to the management of infectious diseases [ 4 , 5 , 6 , 7 ]. A clear example can be found in the pandemic caused by the SARS-CoV-2 virus [ 8 , 9 , 10 , 11 ]. The rapid spread of the virus and the lack of effective treatments for the disease have led to millions of deaths worldwide, as well as severe socioeconomic consequences.…”

“…The platform was applied to 23 blood serum samples of COVID-19 patients for clinical validation, with a LOD of 2.9 ng/mL in buffer solution; however, even though possible interference from immunoglobulins was assessed, the responses generated by antigen proteins from other frequently occurring viruses were not evaluated. More recently, the same authors improved the performance of the developed device by including machine learning (ML) features and by replacing the antibody receptor with a human ACE2 enzyme [ 8 ], enabling the evaluation of the response induced by different variants of SARS-CoV-2 in nasopharyngeal swab samples. In this, case a LOD of 5.14 ng/mL was obtained for SARS-CoV-2 S1 with a sample incubation time of one hour; however, the developed biosensor showed moderate selectivity towards SARS-CoV-2 among other common respiratory viruses, such as influenza A H1N1.…”

“…By limiting the computational power and the time required during the test phase, we were able to implement the whole acquisition and processing actions on a low-cost, low power MCU platform with benefits in terms of flexibility (no need to select a particular platform) and a higher usability (longer battery life, ease of use). Even though a similar approach was applied recently by Beduk et al [ 8 ], that solution requires external devices for the training phase and a connection via a wired serial port, limiting portability. Moreover, that device is powered via a smartphone USB connection.…”