The rapid and accurate detection of emerging viruses is paramount for effective disease management. Reverse transcription-polymerase chain reaction (RT-PCR) is the current gold standard for SARS-CoV-2 diagnosis, but it's a time-consuming process that requires RNA extraction, increasing the risk of infection for healthcare workers. This necessitates the development of faster, higher-throughput diagnostic methods. Nanopores, which are sub-micrometer-sized holes in a substrate, offer a potential solution. When a virus translocates through a nanopore, it causes a measurable change in ionic current, creating a characteristic waveform. Previous research has shown that artificial intelligence (AI) can analyze these waveforms to identify viruses without genome extraction. However, challenges remain in the precise fabrication of nanopore devices with high yield and the development of a robust and reliable measurement system for acquiring high-quality data suitable for AI analysis. This study addresses these challenges by developing a novel AI-nanopore platform for accurate and rapid coronavirus detection.

Several studies have explored the use of nanopores for virus detection. Solid-state nanopores have been used to detect various viruses and bacteria by analyzing the changes in ionic current during their passage through the nanopore. AI-driven classification of the resulting waveforms has demonstrated the potential for highly accurate single-virus identification. These studies, however, often faced limitations in the reproducibility and scalability of nanopore fabrication and the precision and stability of the measurement systems. The current research builds upon this prior work, leveraging advancements in nanopore fabrication and machine learning to overcome these limitations.

The researchers developed a novel AI-nanopore platform consisting of three main components: a scalable, cost-effective semiconducting nanopore module, a portable high-speed current measuring instrument, and machine learning software. The nanopore module was fabricated using microfabrication techniques, achieving high accuracy (diameter error ±10 nm) and yield (90%). The nanopores were designed with diameters of approximately 300 nm, comparable to the size of coronaviruses. Initially, the platform's performance was evaluated using polystyrene nanoparticles of known sizes. Ionic current-time waveforms were recorded and analyzed using a custom-developed machine learning algorithm that extracted key features from the waveforms and trained a classifier to distinguish between the nanoparticles with high accuracy (97%). This algorithm utilized various features including peak current (*I*_p), current duration (*t*_a), current vectors, and time vectors, to enhance the classification performance. Subsequently, the platform was used to analyze cultivated coronaviruses (HCoV-229E, SARS-CoV, MERS-CoV, and SARS-CoV-2). The ionic current waveforms generated by these viruses were analyzed using the same machine learning algorithm. Finally, the platform was tested on clinical saliva samples, comparing results against RT-PCR testing. The machine learning model was retrained using data from PCR-positive and PCR-negative samples. The positive or negative diagnosis of a sample was determined based on the positive ratio (the ratio of the number of waveforms of the novel coronavirus to the total number of waveforms). The sensitivity and specificity were evaluated at different measurement times using confusion matrices.

The AI-nanopore platform demonstrated high accuracy in distinguishing between polystyrene nanoparticles with diameters differing by only 20 nm. The platform successfully identified four different types of coronaviruses (HCoV-229E, SARS-CoV, MERS-CoV, and SARS-CoV-2) with high accuracy in controlled experiments. The platform achieved a detection sensitivity of 90% and a specificity of 96% for SARS-CoV-2 in clinical saliva samples within 5 minutes of measurement. This was achieved without RNA extraction. The machine learning algorithm played a crucial role in achieving high accuracy, especially when dealing with overlapping histograms of I_t and t_d in clinical samples. The analysis of waveforms in both the training and independent testing datasets showed time-dependent sensitivity and specificity, indicating the potential for faster and more efficient testing by adjusting measurement times.

The results demonstrate the feasibility of a rapid and accurate coronavirus detection method using an AI-nanopore platform. The elimination of the RNA extraction step significantly simplifies the testing procedure and reduces the risk of infection to healthcare workers. The high sensitivity and specificity achieved in the clinical saliva samples indicate the platform's potential for use in large-scale screening or point-of-care diagnostics. The platform's versatility is also highlighted by its adaptability to other viruses. The high F-value obtained in differentiating cultured SARS-CoV-2 and influenza A virus suggests the potential of this platform for simultaneous detection of multiple respiratory viruses. Further investigation and refinement of the machine learning algorithms and the nanopore design may further improve the sensitivity, specificity, and speed of the assay.

This study introduces a novel AI-nanopore platform for rapid and accurate coronavirus detection. The platform successfully identified four different coronaviruses and demonstrated high sensitivity and specificity for SARS-CoV-2 in clinical saliva samples without requiring RNA extraction. The platform's versatility and adaptability to other viruses, such as influenza A, offer significant potential for future applications in diagnostics and disease surveillance. Further research could focus on optimizing the platform for even faster detection, higher throughput, and integration with point-of-care devices.

The study's limitations include the relatively small sample size used for the clinical validation of the saliva tests, which could lead to overfitting and limit the generalizability of the findings. Also, the study relied on saliva samples filtered through a 0.45-μm membrane filter, which could potentially affect the virus concentration and its interaction with the nanopore. Further studies with larger, more diverse clinical samples and unfiltered specimens are needed to validate the platform's robustness and reliability. The study also involved handling multiple coronavirus strains with different biosafety levels (BSL-2 and BSL-3), which posed logistical and safety challenges. Direct comparison of multiple strains within the same device was not possible due to these biosafety concerns.