We collected UV-C sensitivity of ssRNA viruses form published studies and carefully selected D₉₀ values (Table 1). The selection was based on the careful assessment of methods that were used to determine UV-C sensitivity. The selected UV-C sensitivity of an ssRNA virus is determined via the standard method (Bolton and Linden, 2003), with the log₁₀ survivors as a function of UV dose and represented as D₉₀.

The molecular size and nucleotide sequences of genomes used in this study were directly obtained from available NCBI genome database (Table 2 and Table 5). PyNNFV model was developed based on the frequency of each type of pyrimidine dinucleotides (TT, TC, CT, and CC) which varies based on genome sequences. Pyrimidines are almost 10 times more susceptible to photoreaction (Smithyman and Hanawalt, 1969), while strand breaks, inter-strand cross links and DNA-protein cross links form with less frequency (1:1000 of the number of dimers and hydrates) (Setlow and Carrier, 1966). Three simple rules were formulated for sequence-dependent dimerization (Becker and Wang, 1989); “(i) When two or more pyrimidines are neighboring to one another, photoreactions are observed at both pyrimidines, (ii) Non-adjacent pyrimidines exhibit little or no photoreactivity, and (iii) Purines form UV photoproducts when they are flanked at 5′ side by two or more adjacent pyrimidine residues.” Therefore, we considered 100% probability of formation of photoreaction products when PyNN are flanked by pyrimidines on both sides and 50% probability when PyNN are flanked by purine on either side. The individual PyNNs were counted by the exclusive method (each pyrimidine considered in one PyNN combination only). Research studies showed the proportion of photoreaction products in the order of TT > TC > CT > CC (Douki, 2013), thus same sequence was followed in counting individual PyNNs. Table 3 shows the method used for PyNNFV calculation in this study. A mathematical function was written to calculate PyNNFV from the potential PyNNs to exist in the genome of RNA (Eq. 1).

The PyNNFVs from complete genome sequences of 16 ssRNA viruses and corresponding reported D₉₀ values were used to plot a model graph. Then, the correlation between PyNNFVs and D₉₀ values was analyzed by fitting the appropriate regression model (linear regression).

To determine the relationship between genome size and UV-C sensitivity, the D₉₀ values were plotted against the genome size of various ssRNA viruses (Figure 1). The data were best fitted to log linear regression model with r² = 0.63. The results revealed that there was a decisive relationship between genome size and UV sensitivity across the range of 3569–29751 bp.

Further to evaluate the influence of base composition and sequence along with genome size on UV-C sensitivity, the D₉₀ values were plotted versus PyNNFV (Figure 2). Linear regression model was best fitted with r² = 0.90. Therefore, based on the value of r squared a moderate positive relationship was found between PyNNFV and UV-C sensitivity of the virus. The following linear regression equation shows the correlation between D₉₀ values and PyNNFV.

Also, to predict the distribution of UV-C sensitivities and estimates of the true population mean using this model, 95% prediction and confidence intervals were shown in Figure 2. To confirm the adequacy of the fitted model, studentized residuals versus run order were tested and the residuals were observed to be scattered randomly, suggesting that the variance was constant. It can be indicated from Figure 3 that predicted values were in close agreement with the experimental values and were found to be not significantly different at p > 0.05 using a paired t-test. Despite some variations, results obtained predicted model and actual experimental values showed that the established models reliably predicted the D₉₀ value. Therefore, the predictive performance of the established model can be considered acceptable. The applicability of the models was also quantitatively evaluated by comparing the bias and accuracy factors (Table 4 and Eqs 3 and 4).

The average mean deviation (E%) were used to determine the fitting accuracy of data (Eq. 5). Where, n_e is the number of experimental data, V_E is the experimental value and V_P is the predicted value.

In most cases, as shown in Table 4, the accuracy factor (AF) values for the genomic model were close to 1.00, except for Measles virus (0.83), Semliki forest virus (0.86). The bias factor (BF) values for the predicted models were also close to 1.00, ranging from 1.02 to 1.21 for the parameter studied. These results clearly indicate that there was a good agreement between predicted and observed D₉₀ values. Ross et al. (2000) stated that predictive models ideally would have an AF = BF = 1.00, indicating a perfect model fit where the predicted and actual response values are equal. However, typically, the AF of a fitted model will increase by 0.10–0.15 units for each predictive variable in the model (Ross et al., 2000). Genomic model, as in this study, that forecasts a response may be expected to have AF and BF values ranging from 0.83 to 1.21 or an equivalent percentage error range of 0.82–25.67%.

Owing to good model fitting, the PyNNFV genomic model was used to predict UV sensitivity of coronaviruses including SARS-CoV-2 and different HuNoV genogroups. PyNNFV values of target viruses were calculated from genomic sequences obtained from the NCBI database. The UV sensitivities were predicted by substituting PyNNFV value in Eq. 2. Table 5 shows PyNNFV values and corresponding predicted D₉₀ values of target viruses. Predicted D₉₀ of SARS-CoV-2 virus (21.5 J/m²) (Table 5) is closer to the estimated D₉₀ of SARS-CoV-1 (18 J/m²) from the experimental study (Table 1). Kariwa et al. (2006) irradiated 2 mL of SARS-CoV-1 in 3-cm petri dishes without stirring UV-C light at 134 μW/cm² for 15 min, and observed reduction in infectivity from 3.8 × 10⁷ to 180 TCID₅₀/mL with equivalent to D₉₀ value of 226 J/m². In contrast, Darnell et al. (2004) showed 4 log reduction of SARS-CoV-1 at UV-C exposure of 4016 μW/cm² for 6 min which is equivalent to D₉₀ value of 3610 J/m². The authors conducted the experiment in a 24 well plate containing 2 mL virus aliquots without mixing. These two studies neither calculate the average irradiance nor provide conditions for uniform UV-C dose distribution throughout the test fluid and thereby reported higher values. The model predicted D₉₀ value of MERS-CoV (28.1 J/m²) that is found to be higher than SARS-COV-2, whereas murine hepatitis coronavirus (MHV) strains showed similar UV-C sensitivity (D₉₀ values = 20.3 to 21 J/m²). For α- and γ-coronaviruses, the predicted D₉₀ values (17.8 to 18.3 J/m²) were lower than the β-coronaviruses (Table 5). Saknimit et al. (1988) demonstrated the efficiency of UV-C irradiation on the inactivation of MHV and CCV coronaviruses using 15 W UV-C lamp at a distance of 1 m and reported efficient UV-C inactivation after 15 min treatment. From this data, the estimated D₉₀ values for MHV and CCV (γ-coronavirus) were 17 and 15 J/m², respectively, and observed to be slightly lower (∼20%) than the model predicted values (Table 5). Overall the results show that coronaviruses are highly sensitive to UV-C light than other ssRNA viruses reported in Table 1. From the UV sensitivity data obtained using the genomic model, it was observed that UV doses ranging from 90 to 141 J/m² are required for 5 log reduction of human pathogenic coronaviruses (SARS-CoV-1, MERS-CoV, 2019-nCoV). Here we demonstrate an example of UV exposure using a low-pressure mercury lamp. If the UV-C lamp source provides an average irradiance of 0.4 mW/cm² or 4 W/m² (under uniform dose distribution conditions), a mere 35 s treatment is adequate to inactivate β-coronaviruses (99.999% or 5 log reduction). Since the developed model relies on total PyNNFV (not on specific gene sequences), slight viral mutations should not cause significant variations in UV sensitivity. For instance, if the PyNNFV value of SARS-CoV-2 changes up to ±10%, the model predicted UV sensitivity (D₉₀ value) ranges from 20.4 to 22.6 J/m² with the change of just ±2.6%.

The predicted D₉₀ values of HuNoVs are 69.1, 89, and 77.6 J/m² for genogroups, GI, GII, and GIV, respectively (Table 5). The results revealed that the UV-C sensitivity of GII was lower with higher predicted D₉₀ value in comparison to GI and GIV. To the best of our knowledge, limited experimental data is currently available on UV-C sensitivity of HuNoVs. Some research studies used RT-qPCR method to estimate MNV survivors and validated with virus infectivity assay (Wang and Tian, 2013; Rönnqvist et al., 2014; Walker et al., 2019). The reported validation results showed that the values obtained with RT-qPCR method are overestimated compared to standard virus infectivity assays (Wang and Tian, 2013; Rönnqvist et al., 2014; Walker et al., 2019). For instance, Rönnqvist et al. (2014) reported 4-log reduction of MNV at a UV dose of 60 mJ/cm² with the infectivity assay, whereas just 2-log decline of MNV and HuNoV RNA levels was found at a UV dose of 150 mJ/cm² by the RT-qPCR method. The experimental D₉₀ values of conservative surrogates (MNV, echovirus and caliciviruses) obtained via viability assay are reported to be in the range of 60–100 J/m² (Table 1).

Validation of the UV-C inactivation kinetics of specific pathogens such as SARS-CoV-2 is not possible (without the use of appropriate surrogates) because of the need for sophisticated biosafety level (BSL)-3 containment, and to protect the researchers, and the public from health risk in environmental settings. For HuNoV, research on reproducible cultivable systems that obtain high titers are still on-going. Hence, criteria for the selection and application of surrogates are required to ensure that the surrogates mimic the behavior of the SARS-CoV-2 or HuNoVs under specific treatment conditions, while ensuring safety of personnel and also decreasing labor, cost and time. Also, surrogates are useful in process validation studies at scale up that can reduce the uncertainties linked with UV-C dose measurement.

As seen from Table 5, the model predicted D₉₀ value (∼21.5 J/m²) of SARS-CoV-2 was comparable to MHV strains (non-pathogenic to humans) of the β-coronavirus group (∼21 J/m²), higher than α-coronaviruses (TGEV, CCV, and FIPV) and γ-coronavirus (AIBV) (∼18 J/m²). Also, since both SARS-CoV-2 and MHV are β-coronaviruses, MHV-strain A59 may show similar behavior under various culture conditions making it a potential surrogate for SARS-CoV-2 for UV-C inactivation kinetics and validation studies.

For HuNoVs, the predicted D₉₀ values of all genogroups (69–89 J/m²) were higher than D₉₀ values of the reported caliciviruses (60–67 J/m²) in our study, echoviruses (75 J/m²), except being lower than MNV-1 (100 J/m²) (Tables 1, 5). Use of surrogates that exhibit similar or slightly higher D₉₀ values to target pathogens can avoid the risk associated with improper inactivation, hence our results indicate that MNV-1 is the better choice (though conservative) to validate UV-C inactivation of all HuNoVs under laboratory experimental setup conditions.

In conclusion, a predictive genomic-modeling method was developed for estimating the UV sensitivity of SARS-CoV-2 and HuNoVs. Results of the model validation showed that the developed model had acceptable predictive performance, as assessed by mathematical and graphical model performance indices. We predicted the D₉₀ values by conducting extensive genomic modeling. Although the parameters reported here may suffice to estimate the UV sensitivity, experimental research directed to address various knowledge gaps identified in this study is required to maximize the accuracy of predicted models. Additional parameters will be computed to the predictive model as needed, including terms for the presence of chromophores or UV absorbers and for possible UV scattering.