Optical imaging spectroscopy for rapid, primary screening of SARS-CoV-2: a proof of concept

1 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports Optical imaging spectroscopy for rapid, primary screening of SARS‑CoV‑2: a proof of concept Emilio Gomez‑Gonzalez1,2*, Alejandro Barriga‑Rivera1,3, Beatriz Fernandez‑Muñoz4, Jose Manuel Navas‑Garcia5, Isabel Fernandez‑Lizaranzu1,2, Francisco Javier Munoz‑Gonzalez1, Ruben Parrilla‑Giraldez6, Desiree Requena‑Lancharro1, Pedro Gil‑Gamboa1, Cristina Rosell‑Valle2,4, Carmen Gomez‑Gonzalez7,8, Maria Jose Mayorga‑Buiza2,9,10, Maria Martin‑Lopez2,4, Olga Muñoz11, Juan Carlos Gomez‑Martin11, Maria Isabel Relimpio‑Lopez10,12,13, Jesus Aceituno‑Castro11,14, Manuel A. Perales‑Esteve15, Antonio Puppo‑Moreno7,8, Francisco Jose Garcia‑Cozar16, Lucia Olvera‑Collantes17, Raquel Gomez‑Diaz2, Silvia de los Santos‑Trigo18, Monserrat Huguet‑Carrasco19, Manuel Rey20, Emilia Gomez21, Rosario Sanchez‑Pernaute4, Javier Padillo‑Ruiz2,10,22 & Javier Marquez‑Rivas2,10,23,24 Effective testing is essential to control the coronavirus disease 2019 (COVID‑19) transmission. Here we report a‑proof‑of‑concept study on hyperspectral image analysis in the visible and near‑infrared range for primary screening at the point‑of‑care of SARS‑CoV‑2. We apply spectral feature descriptors, partial least square‑discriminant analysis, and artificial intelligence to extract information from optical diffuse reflectance measurements from 5 µL fluid samples at pixel, droplet, and patient levels. We discern preparations of engineered lentiviral particles pseudotyped with the spike protein of the SARS‑CoV‑2 from those with the G protein of the vesicular stomatitis virus in saline solution and artificial saliva. We report a quantitative analysis of 72 samples of nasopharyngeal exudate in a range of SARS‑CoV‑2 viral loads, and a descriptive study of another 32 fresh human saliva samples. Sensitivity for classification of exudates was 100% with peak specificity of 87.5% for discernment from PCR‑negative but symptomatic cases. Proposed technology is reagent‑free, fast, and scalable, and could substantially reduce the number of molecular tests currently required for COVID‑19 mass screening strategies even in resource‑limited settings. OPEN 1Department of Applied Physics III, ETSI School of Engineering, Universidad de Sevilla, Camino de los Descubrimientos s/n, 41092 Sevilla, Spain. 2Institute of Biomedicine of Seville (IBIS), 41013 Sevilla, Spain. 3School of Biomedical Engineering, The University of Sydney, Sydney, NSW 2006, Australia. 4Unidad de Producción y Reprogramación Celular (UPRC), Red Andaluza de Diseño y Traslación de Terapias Avanzadas, Consejería de Salud y Familias, Junta de Andalucía, 41092 Sevilla, Spain. 5EOD-CBRN Group, Spanish National Police, 41011 Sevilla, Spain. 6Technology and Innovation Centre, Universidad de Sevilla, 41012 Sevilla, Spain. 7Service of Intensive Care, University Hospital ‘Virgen del Rocio’, 41013 Sevilla, Spain. 8Department of Medicine, College of Medicine, Universidad de Sevilla, 41009 Seville, Spain. 9Service of Anesthesiology, University Hospital ‘Virgen del Rocio’, 41013 Sevilla, Spain. 10Department of Surgery, College of Medicine, Universidad de Sevilla, 41009 Seville, Spain. 11Instituto de Astrofísica de Andalucía, CSIC, 18008 Granada, Spain. 12Department of Ophthalmology, University Hospital ‘Virgen Macarena’, 41009 Sevilla, Spain. 13OftaRed, Institute of Health ‘Carlos III’, 28029 Madrid, Spain. 14Centro Astronomico Hispano Alemán, 04550 Almeria, Spain. 15Department of Electronic Engineering, ETSI School of Engineering, Universidad de Sevilla, 41092 Sevilla, Spain. 16Department of Biomedicine, Biotechnology and Public Health, University of Cadiz, 11003 Cadiz, Spain. 17Instituto de Investigación e Innovación Biomedica de Cádiz (INIBICA), 11009 Cadiz, Spain. 18Corporación Tecnológica de Andalucía, 41092 Sevilla, Spain. 19CER ‘Dr. Gregorio Medina Blanco’, 41807 Espartinas, Sevilla, Spain. 20CAMBRICO BIOTECH, 41015 Sevilla, Spain. 21Joint Research Centre, European Commission, 41092 Sevilla, Spain. 22Department of General Surgery, University Hospital ‘Virgen del Rocío’, 41013 Sevilla, Spain. 23Service of Neurosurgery, University Hospital ‘Virgen del Rocío’, 41013 Sevilla, Spain. 24Centre for Advanced Neurology, 41013 Sevilla, Spain. *email: [email protected] 2 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ The widespread community transmission of the coronavirus disease (COVID-19) has forced nations around the globe to impose severe measures including lockdown periods1, border closures2, and mass screening3 to name a few. The vast geographical spread of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and the large number of asymptomatic carriers are among the causes hampering the eradication of the disease. Fortunately, identification and isolation of confirmed cases has demonstrated efficacy in controlling local outbreaks while reducing its propagation4 until the achievement of herd immunity allows a return to normality5. In an unprecedented scenario in our lifetime, the COVID-19 pandemic has stimulated biomedical research to find technological solutions to manage such a global threat. The development of reliable, fast and scalable detection tools is among the top priorities in fighting the spread of the virus, as an increased screening and identification capacity becomes critical in containing outbreaks6, and effective and extended testing procedures are required to improve epidemiological models7 and support public health decisions. Screening and diagnostic tests. From the perspective of disease management, two main categories of examinations are usually considered, namely screening and diagnostic tests8. Screening exams are intended to detect early disease or asymptomatic carriers, i.e., to identify individuals who have the target disease, but which may not have any sign or symptom of it. In a viral infection scenario, the goal of screening tests is the identification of infected individuals to avoid that they may propagate the disease, and they are usually given to asymptomatic individuals9 or to those without known exposure to the pathogen10. To be effective, screening tests should detect infectious subjects before the onset of symptoms and be as broadly applied as possible within the population under analysis. They should not be invasive preferably and be easy to use. Screening tests document ‘an estimate of the level of risk and determine whether a diagnostic test is justified’11. False positives are considered acceptable for high-sensitivity screening tests, ‘particularly if they are not harmful nor expensive’11. Diagnostic tests are intended to diagnose the target disease or condition on the evaluated individual, so that proper clinical measures can be taken. They are required to provide diagnostic precision and accuracy, even at high cost and discomfort for patients. They are also applied after positive results from a screening test8. In a viral infection scenario, diagnostic tests are performed when there are reasons to suspect that an individual may be infected, for example with symptoms, or having been exposed to the pathogen (e.g., close contacts of confirmed cases), in a high-risk group (e.g., health care personnel), or to determine the stage and evolution of the infection in a sick subject. From an operational perspective, diagnostic tests for COVID-19 can be grouped in three major categories: molecular tests (based on detection of the genetic material of the virus, mainly performed on nasal and throat swabs), antigen tests (which detect specific proteins of the SARS-CoV-2 virus, also performed on throat swabs), and serology tests (immunoassays on blood samples that detect certain specific viral antibodies, quantify the immune response and assess the level of immunity). The leading diagnostic test is laboratory-based polymerase chain reaction (PCR), which is a nucleic acid amplification test, i.e., a molecular exam, that has become the common ‘gold standard’ for the detection of SARS-CoV-2, providing excellent sensitivity and specificity while relatively inexpensive. However, it requires complex equipment and reagents, and it is usually performed by highly skilled personnel at centralized laboratory facilities which receive submitted samples, with turnaround time for results from several hours to days. There are also some test kits for use at the point-of care or at-home which provide results in less than an hour, but reagents require careful storage and handling. To overcome some of the drawbacks of molecular tests, hundreds of rapid protein-based diagnostic devices have emerged12. Among them, enzyme-linked immunosorbent assays or lateral flow assays can detect SARSCoV-2 specific human antibodies or viral antigens13. They are also available for at-home use and provide results in 15min, but their performance is substantially poorer, with lower sensitivity than PCR for the detection of mild cases and significant false negatives. Negative antigen tests are usually followed by molecular (e.g., PCR) or repetitive checks along several days for confirmation14. More sophisticated approaches integrate molecular technologies that range from clustered regularly interspaced short palindromic repeats (CRISPR)-based strategies15,16 to improve nucleic acid detection17, to the use of functionalized nanomaterials18,19. However, they are not used for mass screening. Recent innovative approaches for COVID-19 diagnostic also include the combination of health data provided by wearable sensors and selfassessment of symptoms20. Optical technologies for viral detection. Optical spectroscopic methods have been developed to detect viruses in vegetal structures21–23 and in human samples, mostly in blood. They involve polarimetric and fluorescence spectroscopy, and different implementations of Raman spectroscopy for identification of Dengue virus24, hepatitis B and C viruses25, and microfluidic devices to recognize avian influenza A and other respiratory infections26. Further approaches propose the use of nanomaterials targeted to specific viral antibodies to enhance the potentialities of optical spectroscopy to detect the human immunodeficiency virus27. Simpler experimental set-ups pose great interest to develop easy-to-implement testing units suitable for use at the point-of-care, and diffuse reflectance spectroscopy has been applied to discern mosquitos fed with human blood containing Zika virus28 from controls, although signals from potential alterations in tissues and structures due to the infection remain to be discerned from those arising from the virus itself. Since the current pandemic emerged, many innovative optical and photonic techniques are being specifically developed for detection of SARS-CoV-2. Complex microscopy imaging setups have been described to explore Raman scattering measurements, fluorescence imaging and surface plasmon resonance, Fourier-transform infrared spectroscopy and colorimetry29–32. Remarkably significant results for fast COVID-19 diagnosis have been obtained using attenuated total reflection Fourier transform infrared spectroscopy of RNA extracts from 3 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ nasopharyngeal samples combined with machine learning data analysis33, and of pharyngeal saliva samples processed with genetic algorithm-linear discriminant analysis34. Other optical-related detection schemes rely on nanomaterials-enhanced sensing within lab-on-chip devices29 and combined with hyperspectral microscopy imaging35, matrix-assisted laser desorption/ionization time-of-flight mass spectrometry36 or employ fiber optics probes to record diffuse reflectance spectra28. A previous work37 by the same authors showed that hyperspectral imaging of diffuse reflectance in the visible and near-infrared ranges can be used to detect a synthetic viral model commonly employed for the study of SARS-CoV-2. Advances in the knowledge of the mechanisms of airborne spread of the disease have led to research on rapid methods for detecting the SARS-CoV-2 virus and indirect related markers in exhaled air and salivary aerosols. Aimed at developing potential non-contact, fast breath analyzers (‘breathalyzers’), they rely on many different technologies, from well-established gas chromatography, mass spectrometry and photonics biosensors to new concepts of ‘electronic nose’ sensing and terahertz spectroscopy38. Unmet needs in the COVID‑19 pandemic. While there are available diagnostic tests to determine if an individual has an active COVID-19 infection, worldwide spread of the virus and its variants—outpacing most public health measures—shows that there remains an urgent need of easily deployable screening tests to perform mass, repetitive (time seriated) checks to detect asymptomatic infectious individuals. Of particular importance is the identification of those subjects at initial stages—before the onset of symptoms—and the so-called ‘supercarriers’, asymptomatic individuals with a very high viral load, potential ‘super-spreaders’ of the disease39. From national to local and community levels, many types of COVID-19 screening programs are being implemented, even using tests authorized under emergency approvals (or not approved), with varying levels of success. Criteria for this type of extensive testing include high sensitivity and rapid turnaround time, and the (sometimes difficult) availability of authorized molecular tests for confirmation of positive and ‘concerning negative’ cases40. In addition, virus mutations generate different strains which may modify its ability to spread, the severity of related diseases and the performance of public health measures. Unfortunately, by December 2021 there are not any tests authorized (by the United States Food and Drug Administration) to detect specific SARS-CoV-2 variants14, not even for those categorized by the World Health Organization as ‘variants of concern’ (Alpha, Beta, Gamma, Delta, and Omicron) of the disease41. The current COVID-19 pandemic has underlined the significance of searching for easy-to-implement testing tools potentially useful for real-life applications, particularly at the point-of-care and in constrained resource settings. An approach based on light scattering. In the present study, we present a proof of concept of the use of optical diffuse reflectance hyperspectral imaging in the visible and near infrared ranges combined with specific data analysis as a new technique for fast primary screening of SARS-CoV-2 in 5-µL fluid samples deposited on a surface. This approach builds upon (i) the expanding area of hyperspectral imaging42 and their different experimental set-ups and processing approaches43, particularly on close-range44,45 reflectance applications46, (ii) advances on light scattering techniques for the analysis of size and structure of elements (e.g., bacteria, viruses) below the wavelength of the employed light47, and (iii) the aforementioned previous work by the same authors37 that demonstrated how the same methodology can be exploited for detection and quantification of a SARSCoV-2 model in two biofluids (phosphate buffered saline solution and artificial saliva), both as liquid droplets and dry residues. Optical diffuse reflectance results from the complex phenomena of interaction resulting from light incident on thin fluid samples. The mainly involve reflection, refraction, elastic and inelastic (Raman) scattering, absorption, and re-emission (fluorescence), strongly modulated by specific features of the samples, i.e., by the presence of elements with varying degrees of sizes, shapes and potential physical, biological and biochemical crossed interactions in in the solution48. In the present study we surveyed samples of water-based, optically transparent biofluids with viruses (namely, synthetic viral models in saline solution and in artificial saliva, and samples of SARS-CoV-2-positive human nasopharyngeal exudate and fresh saliva, and their corresponding negative controls). Samples were all deposited as relatively small (5 uL) liquid droplets on a supporting plate, and it was hypothesized that most useful information would arise from sub-surface scattering, a physical process of much interest for computer-graphics rendering, usually modeled using the radiative transfer framework49. Note that the diameter of viral particles (and its engineered models) is about 120–140nm, below the lowest wavelength of the employed illumination (< 400nm) and, therefore, optical imaging of individual particles is not feasible and its visualization would require electron microscopy. We have explored hyperspectral image analysis in the visible and near-infrared (VNIR) band of the electromagnetic spectrum because it requires relatively simple optical imaging technology, potentially useful for deployment of easy-to-implement point-of-care devices. However, in the VNIR band, differences among reflectance spectra from positive samples and their negative controls are difficult to discern following the standard spectroscopy approach, that is, looking for distinctive features (i.e., peaks or absorption bands). Instead, we have analyzed averaged differences of pixel spectra relative to the background, integrating them to droplet and patient levels. Using this procedure, conceptually similar to the per-layer information extraction from the multiple, noisy signals from neurons in brain-computer interfaces50,51, it was possible to effectively enhance the embedded information that allowed detecting the presence of the virus and its quantification37. Results The aim of the work presented here is to determine whether the combination of hyperspectral imaging and spectral data analysis can be used for mass screening of SARS-CoV-2. To answer this question, the optical diffuse reflectance spectra of the samples under study in the visible and near-infrared ranges were converted to 4 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ pseudo-absorbance (PA) spectra and processed using spectral feature descriptors (SFDs), partial least-square discriminant analysis (PLS-DA), and machine learning (a feed-forward neural network, FFNN) to extract the information, following the methodology summarized in Fig.1. Three different experiments were performed: Experiment 1 appraised the discernment of two types of synthetic SARS-CoV-2 models (with and without the characteristic spike protein), Experiment 2 evaluated the classification of SARS-CoV-2-positive and negative human nasopharyngeal exudate samples (the same used for conventional PCR tests), and Experiment 3 comprised an observational, descriptive study of fresh samples of SARS-CoV-2-positive and negative human saliva (see details in “Materials and methods” section and Supplementary Information). Spectral information from SARS‑CoV‑2 spike pseudotyped lentiviral particles (Experiment 1). The first step in this study was to answer whether the PA spectra obtained from fluids containing viral particles possess sufficient information for optical detection. We compare pixel spectra obtained from droplets containing SARS-CoV-2 spike pseudotyped lentiviral particles (S-LP) with preparations of lentiviral particles pseudotyped with the G glycoprotein of the vesicular stomatitis virus (VSV-G, i.e., without the spike protein, here named as G-LP) prepared in both, phosphate buffered solution (PBS) and artificial saliva (AS), and their respective negative controls. A total of 193 droplets (corresponding to 109 G-LP, 32 S-LP and 52 controls) were analyzed (see Table1). These viral particles are comparable to the SARS-CoV-2 in overall shape and diameter, have a double lipid envelope and were engineered to also resemble the molecular structure of the surface of the SARS-CoV-2 virions, characterized by the protruding spike proteins. They allowed for a preliminary assessment of our approach while reducing the biosafety requirements needed to handle SARS-CoV-2 samples52. Differences were observed in the overall PA spectra obtained from samples carrying the virus and their negative controls, as shown in Fig.2a,b. These differences appeared more evident at higher viral concentrations. We performed a partial least squarediscriminant analysis to reduce the intrinsic complexity of large subsets of pixel spectra using two latent variables and explore the differences among positive and negative pixels, as illustrated in Fig.2d. The performance score obtained from a stratified tenfold cross-validation (Fig.2c) showed that there was sufficient information to differentiate the individual pixel spectra obtained from preparations with viral particles from those used as a negative control. In fact, the performance score showed a very strong linear correlation (r2 = 0.95) with the viral concentration in PBS preparations, and a strong correlation (r2 = 0.84) for preparations in AS (Fig.2c). Note that the viral concentrations used here were in the range of those found in expelled respiratory fluids from SARS-CoV-2 positive cases53. These findings suggested that this method may be applied to detect the intended pathogen in human specimens. Body fluids such as saliva contain a wide variety of biomolecules and might even carry other viral species. Consequently, we sought to find whether two similar viral particles of the same type, shape and size could be distinguished based only on the presence of the characteristic spikes. We calculated a spectral feature descriptor (namely, SFD) as a measurement of the relative difference between the diffuse reflectance of the sample under analysis and that of its supporting plate over a certain wavelength range, in this case between 483 and 610nm. We Figure1. Schematic of the hyperspectral imaging assay. (a) Three different types of samples were analyzed. Top, samples containing spike pseudotyped lentiviral particles (i.e., synthetic coronaviruses). Middle, human saliva of SARS-CoV-2 suspects. Bottom, inactivated nasopharyngeal swabs for SARS-CoV-2 PCR tests. (b) Several fluid droplets were placed on a supporting plate. (c) The samples were illuminated using two halogen lamps. Sub-surface scattering is illustrated by the red arrows. (d) A sliding sensor recorded hyperspectral images the VNIR range. (e) The reflectance spectrum of each pixel within the hyperspectral matrix was then digitally pre-processed to obtain the pseudo-absorbance spectra. (f) Spectral features descriptors were obtained from pixel spectra for analysis. (g) A feed-forward neural network (FFNN) was trained37 to detect viral content from spectral features and output a pixel-based binary classification. (h) A partial least square-discriminant analysis (PLS-DA) was performed37 using pixel reflectance spectra. 5 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ then compared this spectral descriptor from S-LP samples with G-LP preparations obtained in a previous study37 (see Table1). Normality of the distribution was discarded using the Kolmogorov–Smirnoff test. The Wilcoxon rank sum test performed on the values of the said descriptor demonstrated statistically significant differences (p-value = 0) between both viral preparations and the controls for each concentration, as shown in Fig.2e,f. It must be noted that the differences in the spectral descriptor are more evident at higher viral concentrations, with no overlapping between the inter-quartile ranges from both viral preparations (particularly in PBS). These differences point at the molecular composition of the virions, related to the proteins integrated in their membrane. Findings in this section suggest that our proposed method could be used to discern the presence of the SARS-CoV-2 spike protein in viral solutions in the synthetic biofluids under study (saline solution and artificial saliva). Further analysis could help to determine whether the spectral feature descriptor represents a structural or molecular difference between viruses or a different interaction of the virus with the medium in the fluid. Detection of SARS‑CoV‑2 in nasopharyngeal exudates (Experiment 2). We also sought to answer whether the proposed optical technique could be used for the analysis of inactivated nasopharyngeal exudate specimens employed for PCR testing. Typically, the transport media in the commercial swabs contain a lysis buffer such as guanidinium thiocyanate which inactivates the viral particles while preserving the genetic material54. Hence, the molecular structure of the viral particles is not maintained and therefore, the optical information embedded in these preparations may differ. As previously determined, the optical technique described here can obtain virus-specific information related to the proteins integrated in the viral capsid. To test whether optical detection under safer conditions is feasible, we analyzed inactivated nasopharyngeal exudates from 72 subjects—including 31 positive (24 men, 6 women, and 1 pediatric female case) and 41 negative (16 men, 23 women, and 1 pediatric male and 1 pediatric female cases)—using the same two pixel-based classification methods, i.e., PLS-DA and FFNN. Following the same classification strategy described in a previous publication37, PLS-DA Model 2 and the FFNN model were built from individual pixel data. Their outputs provided per-pixel classifications, subsequently integrated at droplet and patient levels to enhance the mutual information contained in the viral debris. An additional PLS-DA Model 3 was built using patient-averaged spectra, and its output provided a direct per-patient classification (on the same patient sets). Note that training sets employed for PLS-DA models combine the training and validation sets used for the FFNN, while test sets are the same for both classification procedures. The total number of patients was randomly split in a fourfold cross validation of the FFNN. Table2 shows the sample distribution for Trial 1 (see Supplementary Information for Trials 2, 3 and 4). The spectral feature descriptor (SFD) was also computed in a band with visible differences among averaged PA spectra (between 870 and 910nm) to analyze the viral load of samples. Positive cases were confirmed by quantitative reverse transcription PCR (qRT-PCR) in three levels of viral load, high (H: 106 copies mL−1), medium (M: 104 copies mL−1) and low (L: 102 copies mL−1). Table2 shows the sample distribution among the different experimental groups. Figure3a shows their mean pseudo-absorbance pixel spectra and Fig.3b the value of the spectral feature descriptor (SFD) in the band between 870 and 910nm. Figure3c–h shows the classification results obtained by the PLS-DA Model 2 (Trial 1), and Fig.3i–n the results given by the FFNN (Trial 1, see Supplementary Information for Trials 2, 3 and 4). Classification results are shown as receiver operating characteristic (ROC) curves with the corresponding values of the area under the curve (AUROC). Table 1. Experiment 1: Sample distribution of synthetic viral models (SARS-CoV-2 spike pseudotyped lentiviral particles (S-LP) and lentiviral particles pseudotyped with the G protein of the vesicular stomatitis virus (G-LP)) and their respective negative controls (S-LP Ctrl and G-LP Ctrl) in phosphate buffered solution (PBS) and in artificial saliva (AS) at droplet and pixel levels. Concentrations were C1 = 800 TU µL−1, C2 = 1500 TU µL−1, C3 = 3000 TU µL−1 and C4 = 4000 TU µL−1. Data from G-LP samples were obtained from a previous study37. PBS AS G-LP G-LP Ctrl S-LP S-LP Ctrl G-LP G-LP Ctrl S-LP S-LP Ctrl Droplet C4 24 2 4 4 15 3 4 4 C3 9 2 4 4 14 3 4 4 C2 9 2 4 4 14 3 4 4 C1 9 2 4 4 15 3 4 4 Total 51 8 16 16 58 12 16 16 Pixel C4 25,504 1605 4636 4725 13,381 2642 5065 5076 C3 10,532 1600 4703 4243 12,755 2651 4535 4373 C2 8536 1061 3701 3091 11,315 1847 4084 4534 C1 8205 1134 3203 3519 16,594 2826 4742 4989 Total 52,777 5400 16,243 15,578 54,045 9966 18,426 18,972 6 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ Figure2. Experiment 1: Spectral discrimination of SARS-CoV-2 spike pseudotyped lentiviral particles. (a) Mean pseudo-absorbance (PA) pixel spectra of spike pseudotyped lentiviral particles in phosphate buffered solution (S-LPPBS) for high concentration (HC = 4.0 × 103 TU µL−1) and low concentration (LC = 0.8 × 103 TU µL−1) and their respective negative controls (CtrlPBS). (b) Mean pseudo-absorbance (PA) pixel spectra of spike pseudotyped lentiviral particles in artificial saliva (S-LPAS) for high concentration (HC = 4.0 × 103 TU µL−1) and low concentration (LC = 0.8 × 103 TU µL−1) and their respective negative controls (CtrlAS). (c) Overall score value of the tenfold cross-validation essay performed in partial least square-discriminant analysis (PLS-DA Model 1) in both preparations (per pixel). Dashed lines represent their linear correlations. (d) Scatter plot of the two latent variables (V1 and V2) used in the PLS-DA Model 1 for different viral concentrations. Brown and green dots represent positive and control samples respectively. (e,f) Spectral feature descriptor (SFD), computed in the band 483–610nm for S-LP and G-LP samples in phosphate buffered solution (PBS), artificial saliva (AS), and the corresponding culture media as control (Ctrl). ***p-value = 0 using the Wilcoxon rank sum test. Outliers were removed for clarity. 7 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ As we anticipated, the differences in the PA spectra obtained in Experiment 2 (nasopharyngeal exudate samples in the inactivation medium) were less informative than those obtained from the previous preparations of Experiment 1 (synthetic viral models in saline solution and in artificial saliva). The per-pixel PLS-DA Model 2 was built using the samples from 45 patients (20 qRT-PCR-positive, 25 qRT-PCR-negative). The test group included 27 subjects (11 positive and 16 negative) as described in Table2. A total of 12 latent variables were used, and the variance captured was 95.15%. The outputs of the PLS-DA were used to classify the individual pixel spectra. After discarding normality, the two-tailed Wilcoxon rank-sum test performed on the classification output showed statistically significant differences between positive and negative samples (p-value = 0), as shown in Fig.3c. The ROC curve was constructed to determine the pixel classification cut-off value that optimized both, sensitivity and specificity (Fig.3d,e). The obtained AUROC showed a good accuracy (AUROCpixel = 0.88). However, a noticeable improvement was observed by integration at droplet level (AUROCdroplet = 0.95), as shown in Fig.3f,g. Finally, at patient level, we found a very good agreement between the qRT-PCR assays and our proposed technique (sensitivity = 100%, specificity = 87.5%), as illustrated in Fig.3h. Next, the same pixel PA spectra were classified using a FFNN. Following the procedure described in a previous work37, the input of the network was a set of 28 spectral shape features computed in four spectral fringes, and the output provided a pixel-level binary classification. In this case, the neural network was trained using samples from 33 patients (15 qRT-PCR-positive, 18 qRT-PCR-negative), validated using samples from 12 patients (5 qRTPCR-positive, 7 qRT-PCR-negative), and tested in the same subset of 27 patients used in the per-pixel PLS-DA Model 2, as summarized in Table2. Note that both test and validation splits were randomly assigned from the training group used in the PLS-DA Models 2 and 3. Similarly, after rejecting normality, the Wilcoxon rank-sum showed statistically significant differences in the output of positive and negative pixels (p-value = 0), as illustrated in Fig.3i. The performance of the classifier, as in the case of the PLS-DA, improved importantly from pixel level (AUROCpixel = 0.88) to patient level (AUROCpatient = 0.93), as shown in Fig.3j–n. The performance of both classification methods was comparable in terms of sensitivity and specificity at all levels under consideration, as shown in Table3. Additionally, we performed a fourfold cross-validation essay with random splits (over patients, see Supplementary Information), obtaining an overall sensitivity and specificity of 94.7 ± 4.6% (mean ± std) and 92.7 ± 2.8% (mean ± std) (see Supplementary Information). The overall AUROC was 0.95 ± 0.01 (mean ± std). The additional PLS-DA Model 3 was built upon the averaged pixel spectra per patient for comparison (see Supplementary Information). Its overall sensitivity and specificity were 90.9% and 87.5% respectively, in very good agreement with the results provided by the per-pixel models given by PLS-DA Model 2 and by the FFNN (see Table3, Supplementary Information). Findings in this section show that proposed optical analysis could classify positive and negative cases of SARS-CoV-2 using the same inactivated nasopharyngeal exudates than conventional PCR tests. Spectral information from fresh saliva from SARS‑CoV‑2 patients (Experiment 3). Human saliva has been proved to be a body fluid suitable for detecting ongoing infections of SARS-CoV-255–57. Therefore, we analyzed the optical PA spectra obtained from saliva specimens to determine if we could extract useful information using the proposed technique. As indicated, data collection for this study was carried out during the first wave of the COVID-19 pandemic. Under such extraordinary conditions, it was very difficult to obtain and manage SARS-CoV-2 positive samples. Based on the results described in previous sections, we limited our analysis of fresh saliva samples to an initial, descriptive study with a reduced number of cases to evaluate the feasibility and potential interest of our technique for this application. Under biological containment conditions equivalent to biosafety level 3, we studied the pixel spectra obtained from a total of 192 droplets from 6 positive (3 men and 3 women) and 26 negative (8 men and 18 women) cases determined by PCR test. These sets are clearly imbalanced with respect to the presence of SARS-Cov-2 and gender. Nevertheless, differences were observed in the PA spectra of positive and negative cases, as illustrated in Fig.4a. Table 2. Experiment 2: Sample distribution of nasopharyngeal exudates at different levels (total numbers of patients, droplets and pixels) among the experimental Training (Tra), Validation (Val), and Test groups (Trial 1). Positive (Pos) and negative (Neg) cases were determined by qRT-PCR. Positive cases include three viral load levels (H = 106 copies mL−1, M = 104 copies mL−1 and L = 102 copies mL−1). Note that classification algorithms provide results as (per-pixel, per-droplet and per-patient) ‘positive’ (with any level of viral load) or ‘negative’ assignations. PLS-DA partial least square-discriminant analysis, FFNN feed-forward neural network. Pixel Droplet Patient Pos Neg Pos Neg Pos NegH M L H M L H M L Total PLS-DA Tra 144,426 35,715 30,545 808,184 78 24 18 150 13 4 3 20 25 Test 50,264 32,902 26,579 146,440 30 18 18 96 5 3 3 11 16 FFNN Tra 111,676 27,924 21,498 744,855 60 18 12 108 10 3 2 15 18 Val 32,750 7791 9047 63,329 18 6 6 42 3 1 1 5 7 Test 50,264 32,902 26,579 146,440 30 18 18 96 5 3 3 11 16 8 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ To explore those differences, a PLS-DA Model 1 (using all samples) was constructed upon two latent variables and the clusters are shown in Fig.4b. The spectral feature descriptor was also computed in the range between 483 and 610nm (Fig.4c) and between 407 and 470nm (Fig.4d). Normality was discarded using the KolmogorovSmirnoff test. The Wilcoxon rank sum test revealed statistically significant differences in the values of the SFD (p-value = 0). Despite the relatively low number of positive cases presented in this section, these findings suggest that the method presented here might be used to differentiate the presence of the SARS-CoV-2 in human fresh saliva samples. We also found statistically significant differences in the values of the spectral feature descriptor for positive and negative cases, both for male and female samples (Fig.4c,d, p-value = 0). Figure3. Experiment 2: Analysis of SARS-CoV-2 nasopharyngeal exudates. (a) Mean pseudo-absorbance (PA) pixel spectra from positive (viral loads H = 106 copies mL−1 (high), M = 104 copies mL−1 (medium) and L = 102 copies mL−1 (low)) and negative cases of SARS-CoV-2. In addition, the panel shows the mean pixel spectra of the supporting plate (background). (b) Value of the spectral feature descriptor (SFD) in the band between 870 and 910nm for all cases. Positive (Pos) box includes all H, M and L samples. (c–h) Classification results obtained using partial least square-discriminant analysis (PLS-DA Model 2, Trial 1). (i–n) Classification results obtained using a feed-forward neural network (FFNN, Trial 1). (c,i) Median value of the output used for pixel classification. Positive and negative samples were determined by qRT-PCR. ***p-value = 0 using Wilcoxon rank-sum test. The blue dashed line illustrates the classification threshold. (d,j) Receiver operating characteristic (ROC) curves from the pixel classification. (e,k) Example of the resulting pixel classification in a droplet. Red and green pixels were classified as positive and negative respectively. Note red and green backgrounds indicate a positive sample and its negative control. (f,l) ROC curves obtained from droplet classifications. (g,m) Example of the classification of several droplets from a positive (red background) and a negative (green background) patient. (h,n) ROC curves obtained from patient diagnosis. Outliers were removed for clarity. 9 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ Discussion The development of fast and reliable screening techniques represents a critical enabler in a pandemic scenario, as quick, easy-to-implement and cost-effective test methods are essential for the detection of positive cases to curb propagation and to provide viral prevalence data58. In the specific context of the COVID-19, given the implications of false negatives for the spread of the disease59, highly sensitive tools are urgently needed for mass screenings to identify virus carriers58,60, even at the cost of reduced specificity relative to other respiratory viral species. In resource-constrained settings such screening approaches may also be useful to protect high-risk patients in those scenarios12. Here we have demonstrated the strong potential of hyperspectral image analysis in the visible and nearinfrared range to contribute to fighting the transmission of the SARS-CoV-2, as it allows for high-throughput testing by scanning many samples simultaneously using standard, relatively simple, optical equipment. This technique does not require the addition of any reagent to the samples or the amplification of nucleic acids, and may therefore be easily deployed in transport hubs, mass events, and during vivid outbreaks. In this study the detection and classification results of exudate samples were obtained applying two different and independent analytical procedures: partial least square-discriminant analysis and feed-forward neural networks. PLS-DA is a statistical multivariate data processing technique commonly employed for descriptive analysis and predictive classification of highly dimensional data, including genomic data sets61, metabolomics62 Table 3. Experiment 2: Values of sensitivity (SE), specificity (SP) and area under the receiving operating characteristic (AUROC) curve obtained at per-patient classification from both types of analysis, partial least square-discriminant analysis (PLS-DA Models 2 and 3) and the feed-forward neural network (FFNN), of inactivated nasopharyngeal exudates using the same patient sets (Trial 1). PLS-DA (Model 2) (perpatient, from per-pixel classification) FFNN (per-patient, from per-pixel classification) PLS-DA (Model 3) (perpatient, from patientaveraged spectra) SE (%) SP (%) AUROC SE (%) SP (%) AUROC SE (%) SP (%) AUROC Pixel 83.3 78.8 0.88 83.5 79.2 0.88 – – – Droplet 97.0 89.6 0.95 97.0 88.5 0.95 – – – Patient 100.0 87.5 0.98 100.0 87.5 0.93 90.9 87.5 0.97 Figure4. Experiment 3: Spectral information from SARS-CoV-2 in fresh saliva specimens. (a) Mean (± std) of the pseudo-absorbance pixel spectra from fresh saliva samples. Positive (pos) and negative (neg) cases were determined by PCR test. (b) Scatterplot of the two latent variables obtained in partial least square-discriminant analysis (PLS-DA) Model 1 from the pixel spectra. Positive and negative PCR tests are brown and green dots respectively. (c) Value of the spectral feature descriptor (SFD) obtained between 483 and 610nm (grey band in (a)) for positive and negative cases. (d) Value of the spectral descriptor SFD obtained between 407 and 470nm for positive and negative cases. (c,d) Both panels include differentiation between male and female pixel sets. ***p-value = 0 using the Wilcoxon rank sum test. Outliers were removed for clarity. 16 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ Normality was tested using the Kolmogorov–Smirnoff test at 95% significance. Wilcoxon rank sum tests were used for comparison of non-normal distributions. To characterize statistical distributions, we used box plots displaying the median values and the inter-quartile ranges. The whiskers were used to show 2.7 times the standard deviation, and the outliers beyond that range were removed for clarity. Here, p-values below 0.0001 were considered as zero. Data availability All data needed to evaluate the findings are present in the paper and the Supplementary Materials. Additional data related to this work are available from the authors under reasonable request. Received: 22 September 2021; Accepted: 28 January 2022 References 1. Lau, H. et al. The positive impact of lockdown in Wuhan on containing the COVID-19 outbreak in China. J. Travel Med. 27, 037 (2020). 2. Wells, C. R. et al. Impact of international travel and border control measures on the global spread of the novel 2019 coronavirus outbreak. Proc. Natl. Acad. Sci. 117, 7504–7509 (2020). 3. Lopes-Júnior, L. C. et al. Effectiveness of mass testing for control of COVID-19: A systematic review protocol. BMJ Open 10, e040413 (2020). 4. Hellewell, J. et al. Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts. Lancet Glob. Health 8, e488–e496 (2020). 5. Rossman, H. et al. COVID-19 dynamics after a national immunization program in Israel. Nat. Med. 27, 1–7 (2021). 6. Udugama, B. et al. Diagnosing COVID-19: The disease and tools for detection. ACS Nano 14, 3822–3835 (2020). 7. Hay, J. A. et al. Estimating epidemiologic dynamics from cross-sectional viral load distributions. Science 5, 78 (2021). 8. Lewis, G., Sheringham, J., Bernal, J. L. & Crayford, T. Mastering Public Health: A Postgraduate Guide to Examinations and Revalidation (CRC Press, 2014). 9. Shental, N. et al. Efficient high-throughput SARS-CoV-2 testing to detect asymptomatic carriers. Sci. Adv. 6, 5961 (2020). 10. U.S. Food and Drug Administration. COVID-19 Test Uses: FAQs on Testing for SARS-CoV-2. Q: What is the Difference Between Surveillance, Screening, and Diagnostic Testing for COVID-19 Testing? (2021). https:// www. fda. gov/ medic aldevic es/ coron aviruscovid19andmedic aldevic es/ covid19testusesfaqstesti ngsarscov-2. Accessed 2 Feb 2022. 11. European Centre for Disease Prevention and Control. Diagnostic Tests Versus Screening Tests (2021). https:// wiki. ecdc. europa. eu/ fem/ Pages/ Diagn ostic% 20tes ts% 20ver sus% 20scr eening% 20tes ts. aspx. Accessed 2 Feb 2022. 12. Vandenberg, O., Martiny, D., Rochas, O., van Belkum, A. & Kozlakidis, Z. Considerations for diagnostic COVID-19 tests. Nat. Rev. Microbiol. 19, 1–13 (2020). 13. Cheng, M. P. et al. Serodiagnostics for severe acute respiratory syndrome-related coronavirus 2: A narrative review. Ann. Intern. Med. 173, 450–460 (2020). 14. U.S. Food and Drug Administration. Coronavirus Disease 2019 Testing Basics (2021). https:// www. fda. gov/ co nsu mers/ consu merupdat es/ coron avirusdisea se2019testi ngbasics. Accessed 2 Feb 2022. 15. Broughton, J. P. et al. CRISPR–Cas12-based detection of SARS-CoV-2. Nat. Biotechnol. 38, 870–874 (2020). 16. Ackerman, C. M. et al. Massively multiplexed nucleic acid detection with Cas13. Nature 582, 277–282 (2020). 17. Gootenberg, J. S. et al. Nucleic acid detection with CRISPR-Cas13a/C2c2. Science 356, 438–442 (2017). 18. Moitra, P., Alafeef, M., Dighe, K., Frieman, M. B. & Pan, D. Selective naked-eye detection of SARS-CoV-2 mediated by N gene targeted antisense oligonucleotide capped plasmonic nanoparticles. ACS Nano 14, 7617–7627 (2020). 19. Ahmadivand, A. et al. Functionalized terahertz plasmonic metasensors: Femtomolar-level detection of SARS-CoV-2 spike proteins. Biosens. Bioelectron. 177, 112971 (2021). 20. Quer, G. et al. Wearable sensor data and self-reported symptoms for COVID-19 detection. Nat. Med. 27, 73–77 (2021). 21. Lee, H. et al. Detection of cucumber green mottle mosaic virus-infected watermelon seeds using a near-infrared (NIR) hyperspectral imaging system: Application to seeds of the “Sambok Honey” cultivar. Biosyst. Eng. 148, 138–147 (2016). 22. Wang, D. et al. Early detection of tomato spotted wilt virus by hyperspectral imaging and outlier removal auxiliary classifier generative adversarial nets (OR-AC-GAN). Sci. Rep. 9, 1–14 (2019). 23. Nguyen, C. et al. Early detection of plant viral disease using hyperspectral imaging and deep learning. Sensors 21, 742 (2021). 24. Firdous, S. & Anwar, S. Optical Diagnostic of Dengue Virus Infected Human Blood using Raman, Polarimetric and Fluorescence Spectroscopy (InTech Open, 2017). 25. Anwar, S. & Firdous, S. Optical diagnostic of hepatitis B (HBV) and C (HCV) from human blood serum using Raman spectroscopy. Laser Phys. Lett. 12, 076001 (2015). 26. Yeh, Y.-T. et al. A rapid and label-free platform for virus capture and identification from clinical samples. Proc. Natl. Acad. Sci. 117, 895–901 (2020). 27. Manoto, S. L. et al. Exploring optical spectroscopic techniques and nanomaterials for virus detection. Saudi J. Biol. Sci. 28, 78–89 (2021). 28. Fernandes, J. N. et al. Rapid, noninvasive detection of Zika virus in Aedes aegypti mosquitoes by near-infrared spectroscopy. Sci. Adv. 4, 0496 (2018). 29. Lukose, J., Chidangil, S. & George, S. D. Optical technologies for the detection of viruses like COVID-19: Progress and prospects. Biosens. Bioelectron. 178, 113004 (2021). 30. Maddali, H., Miles, C. E., Kohn, J. & O’Carroll, D. M. Optical biosensors for virus detection: Prospects for SARS-CoV-2/COVID19. ChemBioChem 22, 1176 (2021). 31. Thi, V. L. D. et al. A colorimetric RT-LAMP assay and LAMP-sequencing for detecting SARS-CoV-2 RNA in clinical samples. Sci. Transl. Med. 12, 7075 (2020). 32. Stanborough, T. et al. Optical detection of CoV-SARS-2 viral proteins to sub-picomolar concentrations. ACS Omega 6, 6404–6413 (2021). 33. Kitane, D. L. et al. A simple and fast spectroscopy-based technique for Covid-19 diagnosis. Sci. Rep. 11, 1–11 (2021). 34. Barauna, V. G. et al. Ultrarapid on-site detection of SARS-CoV-2 infection using simple ATR-FTIR spectroscopy and an analysis algorithm: High sensitivity and specificity. Anal. Chem. 93, 2950–2958 (2021). 35. Alafeef, M., Moitra, P., Dighe, K. & Pan, D. Hyperspectral mapping for the detection of SARS-CoV-2 using nanomolecular probes with yoctomole sensitivity. ACS Nano 15, 13742–13758 (2021). 36. Yan, L. et al. Rapid detection of COVID-19 using MALDI-TOF-based serum peptidome profiling. Anal. Chem. 93, 4782–4787 (2021). 17 Vol.:(0123456789) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ 37. Gomez-Gonzalez, E. et al. Hyperspectral image processing for the identification and quantification of lentiviral particles in fluid samples. Sci. Rep. 11, 1–12 (2021). 38. Waltz, E. Five COVID breathalyzers. IEEE Spectr. 58, 10 (2021). 39. Yang, Q. et al. Just 2% of SARS-CoV-2-positive individuals carry 90% of the virus circulating in communities. Proc. Natl. Acad. Sci. 118, e2104547118 (2021). 40. U.S. Food and Drug Administration. FDA FACT SHEET. Screening for COVID 19: Deciding Which Test to Use When Establishing Testing Programs (2021). https:// www. fda. gov/ media/ 146666/ downl oad. Accessed 2 Feb 2022. 41. World Health Organization. Tracking SARS-CoV-2 Variants (2021). htt ps:// www. who. int/ en/ activ ities/ track ingSARSCoV-2varia nts/. Accessed 2 Feb 2022. 42. Khan, M. J., Khan, H. S., Yousaf, A., Khurshid, K. & Abbas, A. Modern trends in hyperspectral image analysis: A review. IEEE Access 6, 14118–14129 (2018). 43. Lu, Y., Saeys, W., Kim, M., Peng, Y. & Lu, R. Hyperspectral imaging technology for quality and safety evaluation of horticultural products: A review and celebration of the past 20-year progress. Postharvest. Biol. Technol. 170, 111318 (2020). 44. Daukantas, P. Hyperspectral imaging meets biomedicine. Opt. Photonics News 31, 32–39 (2020). 45. Mishra, P., Lohumi, S., Khan, H. A. & Nordon, A. Close-range hyperspectral imaging of whole plants for digital phenotyping: Recent applications and illumination correction approaches. Comput. Electron. Agric. 178, 105780 (2020). 46. Cucci, C., Delaney, J. K. & Picollo, M. Reflectance hyperspectral imaging for investigation of works of art: Old master paintings and illuminated manuscripts. Acc. Chem. Res. 49, 2070–2079 (2016). 47. Wyatt, P. J. Differential light scattering and the measurement of molecules and nanoparticles: A review. Anal Chim Acta X 7–8, 100070 (2021). 48. Boutier, A. Laser Metrology in Fluid Mechanics: Granulometry, Temperature and Concentration Measurements (Wiley, 2013). 49. Xiao, B., Zhao, S., Gkioulekas, I., Bi, W. & Bala, K. Effect of geometric sharpness on translucent material perception. J. Vis. 20, 10–10 (2020). 50. Nicolas-Alonso, L. F. & Gomez-Gil, J. Brain computer interfaces, a review. Sensors 12, 1211–1279 (2012). 51. Schwemmer, M. A. et al. Meeting brain–computer interface user performance expectations using a deep neural network decoding framework. Nat. Med. 24, 1669–1676 (2018). 52. Crawford, K. H. et al. Protocol and reagents for pseudotyping lentiviral particles with SARS-CoV-2 spike protein for neutralization assays. Viruses 12, 513 (2020). 53. Wang, Y., Xu, G. & Huang, Y.-W. Modeling the load of SARS-CoV-2 virus in human expelled particles during coughing and speaking. PLoS ONE 15, e0241539 (2020). 54. Pastorino, B. et al. Evaluation of chemical protocols for inactivating SARS-CoV-2 infectious samples. Viruses 12, 624 (2020). 55. Azzi, L. et al. Saliva is a reliable tool to detect SARS-CoV-2. J. Infect. 81, e45–e50 (2020). 56. Williams, E., Bond, K., Zhang, B., Putland, M. & Williamson, D. A. Saliva as a non-invasive specimen for detection of SARS-CoV-2. J. Clin. Microbiol. 58, e00776 (2020). 57. Wyllie, A. L. et al. Saliva or nasopharyngeal swab specimens for detection of SARS-CoV-2. N. Engl. J. Med. 383, 1283–1286 (2020). 58. Mercer, T. R. & Salit, M. Testing at scale during the COVID-19 pandemic. Nat. Rev. Genet. 22, 415–426 (2021). 59. Woloshin, S., Patel, N. & Kesselheim, A. S. False negative tests for SARS-CoV-2 infection—Challenges and implications. N. Engl. J. Med. 383, e38 (2020). 60. Mina, M. J., Parker, R. & Larremore, D. B. Rethinking covid-19 test sensitivity—A strategy for containment. N. Engl. J. Med. 383, e120 (2020). 61. Boulesteix, A.-L. & Strimmer, K. Partial least squares: A versatile tool for the analysis of high-dimensional genomic data. Brief Bioinform. 8, 32–44 (2007). 62. Szymańska, E., Saccenti, E., Smilde, A. K. & Westerhuis, J. A. Double-check: Validation of diagnostic statistics for PLS-DA models in metabolomics studies. Metabolomics 8, 3–16 (2012). 63. Feng, L. et al. Hyperspectral imaging for seed quality and safety inspection: A review. Plant Methods 15, 1–25 (2019). 64. Peerbhay, K. Y., Mutanga, O. & Ismail, R. Commercial tree species discrimination using airborne AISA Eagle hyperspectral imagery and partial least squares discriminant analysis (PLS-DA) in KwaZulu–Natal, South Africa. ISPRS J. Photogramm. Remote Sens. 79, 19–28 (2013). 65. Folch-Fortuny, A., Prats-Montalbán, J. M., Cubero, S., Blasco, J. & Ferrer, A. VIS/NIR hyperspectral imaging and N-way PLS-DA models for detection of decay lesions in citrus fruits. Chemomet. Intell. Lab. Syst. 156, 241–248 (2016). 66. Huang, H., Liu, L. & Ngadi, M. O. Recent developments in hyperspectral imaging for assessment of food quality and safety. Sensors 14, 7248–7276 (2014). 67. Bonifazi, G., Capobianco, G. & Serranti, S. Hyperspectral imaging and hierarchical PLS-DA applied to asbestos recognition in construction and demolition waste. Appl. Sci. 9, 4587 (2019). 68. Chevallier, S., Bertrand, D., Kohler, A. & Courcoux, P. Application of PLS-DA in multivariate image analysis. J. Chemom. 20, 221–229 (2006). 69. Golhani, K., Balasundram, S. K., Vadamalai, G. & Pradhan, B. A review of neural networks in plant disease detection using hyperspectral data. Inf. Process Agric. 5, 354–371 (2018). 70. Tang, J. W. et al. Where have all the viruses gone? Disappearance of seasonal respiratory viruses during the COVID-19 pandemic. J. Med. Virol. 93, 4099 (2021). 71. Park, K. Y., Seo, S., Han, J. & Park, J. Y. Respiratory virus surveillance in Canada during the COVID-19 pandemic: An epidemiological analysis of the effectiveness of pandemic-related public health measures in reducing seasonal respiratory viruses test positivity. PLoS ONE 16, e0253451 (2021). 72. Brittain-Long, R. et al. Multiplex real-time PCR for detection of respiratory tract infections. J. Clin. Virol. 41, 53–56 (2008). 73. Díez-Fuertes, F. et al. A founder effect led early SARS-COV-2 transmission in Spain. J. Virol. 95, e01583 (2021). 74. Hodcroft, E. B. et al. Spread of a SARS-CoV-2 variant through Europe in the summer of 2020. Nature 22, 1–9 (2021). 75. Takahashi, T. et al. Sex differences in immune responses that underlie COVID-19 disease outcomes. Nature 588, 315–320 (2020). 76. Hussain, I. & Bowden, A. K. Smartphone-based optical spectroscopic platforms for biomedical applications: A review. Biomed. Opt. Express 12, 1974–1998 (2021). 77. Scherr, M., Battmer, K., Blömer, U., Ganser, A. & Grez, M. Quantitative determination of lentiviral vector particle numbers by real-time PCR. Biotechniques 31, 520–526 (2001). 78. Tran, N. K. et al. Novel application of automated machine learning with MALDI-TOF-MS for rapid high-throughput screening of COVID-19: A proof of concept. Sci. Rep. 11, 1–10 (2021). 79. Kutner, R. H., Zhang, X.-Y. & Reiser, J. Production, concentration and titration of pseudotyped HIV-1-based lentiviral vectors. Nat. Protoc. 4, 495–505 (2009). 80. U.S. Food and Drug Administration. COVID-19 Test Uses: FAQs on Testing for SARS-CoV-2. Q: Can Laboratories Report Ct Values for Authorized Molecular Diagnostic COVID-19 Tests? (2021). https:// www. fda. gov/ medic aldevic es/ coron aviruscovid19andmedic aldevic es/ covid19testusesfaqstesti ngsarscov-2. Accessed 2 Feb 2022. 18 Vol:.(1234567890) Scientific Reports | (2022) 12:2356 | https://doi.org/10.1038/s41598-022-06393-3 www.nature.com/scientificreports/ Acknowledgements The authors would like to gratefully acknowledge the assistance of the members of the Explosive Ordnance Disposal—Chemical, Biological, Radiological & Nuclear (EOD-CBRN) Group of the Spanish National Police, whose identities cannot be disclosed, and who are represented here by JMNG. The authors also thank Manuel Guerrero-Claro for his technical assistance with the implementation of the neural networks, and the companies CAMBRICO BIOTECH (Sevilla, Spain), CER “Dr. Gregorio Medina Blanco” (Espartinas, Sevilla, Spain) and SAMU (Sevilla, Spain) for their collaboration in the collection of human samples. This research was funded by Grants Number COV20-00080 and COV20-00173 of the 2020 Emergency Call for Research Projects about the SARS-CoV-2 virus and the COVID-19 disease of the Institute of Health ‘Carlos III’, Spanish Ministry of Science and Innovation, and by Grant Number EQC2019-006240-P funded by MICIN/AEI/10.13039/501100011033 and by “ERDF A way of making Europe”. ABR was supported by Grant Number RTI2018-094465-J-I00 funded by MICIN/AEI/10.13039/501100011033 and by “ERDF A way of making Europe”. This work has been supported by the European Commission through the Joint Research Center (JRC) HUMAINT project. Author contributions E.G.G. conceived the study, generated the hypotheses, and designed the imaging and analysis methodology and the experimental setup. E.G.G., J.M.R., J.M.N.G., B.F.M. and A.B.R. conceptualized the research. J.M.R. and J.M.N.G. designed the operational methodology. E.G.G., J.M.N.G., J.M.R., R.P.G. performed the experiments. E.G. provided contributions to the methodology, machine learning net and data driven approach. E.G.G., J.M.N.G., J.M.R., I.F.L., F.J.M.G., P.G.G., R.P.G., D.R.L., M.A.P.E., J.A.C., O.M., J.C.G.M., S.S.T., A.B.R. analyzed the data. B.F.M., C.R.V., M.M.L., F.J.G.C., L.O.C. prepared and analyzed virus solutions. M.H.G., M.R. and J.M.N.G. collected and analyzed human samples. M.R. and R.G.D. performed molecular analysis. I.F.L., F.J.M.G., P.G.G., D.R.L., R.P.G. contributed to data curation and programming. J.P.R., A.P.M., C.G.G., M.J.M.B., M.I.R.L., R.S.P. contributed to the experimental design. A.B.R., E.G.G., B.F.M. drafted the manuscript. All authors revised the manuscript critically for important intellectual content and approved the final version. Competing interests EGG, FJMG, RPG, DRL, PGG, JMR, IFL, BFM and JMNG have filed a patent related to the proposed methodology. They intend to make this technology affordable and suitable for research and potential use. MHC works as staff physician for CER ‘Dr. Gregorio Medina Blanco (Espartinas, Seville, Spain). MR is the Chief Scientific Officer of CAMBRICO BIOTECH (Seville, Spain). All other authors declare no competing interests. Additional information Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1038/ s4159802206393-3. Correspondence and requests for materials should be addressed to E.G.-G. Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. © The Author(s) 2022