Based on sources from the available literature, the acute toxicity of Novichoks was perceived to be several times higher than that of conventional NAs (Ellison 2007; Mirzayanov 2008). Novichok A-230 was claimed to be 5–8 times more toxic than compound VX (a relative comparison of the LD50 values under the same conditions). Additionally, A-232 was said to be 10 times as harmful as Soman. Nerve agents A-242 and A-262, toxic derivatives of A-230 and A-232, were classified as ultra-highly toxic despite not specifying any value. However, these primary sources lack information on the acute toxicity of A-234 (Mirzayanov 2008). According to the data published in the seminar paper by Karev, the above reports on Novichok toxicological data were not valid (Karev 2009). In the case of A-232, it showed a value one order of magnitude lower than VX and A-234 two times lower than VX. The acute toxicity of the A series nerve agents, lower than conventional NAs, was somewhat confirmed by other estimated data (Franca et al. 2019). The LD50 value for Novichok A-230 was lower by an order of magnitude, while the compounds A-232 and A-234 were approximately three times less toxic than VX. It is worth mentioning that the similarity in toxicity between A-232 and A-234 was assumed here based on structural similarity. The results for these two Novichoks were highly different compared to the data published by Karev (2009).

Carlsen (2019) also attempted to verify these data for Novichoks (n = 6), using quantitative structure–activity relationship (QSAR) models for estimation and calculated the median lethal dose for oral administration rats that were converted to the human dose. Carlsen presented utterly different data on the relative toxicity of A-series nerve agents, contrary to Mirzayanov's claims. The estimated LD50 value for compound VX was used to indicate the reliability of the data. In the works of Karev (2009) and Franca et al. (2019), it reached the value of 10 mg/person (70 kg). Converted to mg/kg, this value is 0.14 mg/kg, which is very consistent with the calculated value of LD50 for humans: 0.1 mg/kg (Carlsen 2019). Estimated data proved that Novichoks (A-230, A-232, A-234, A-242, and A-262) are 5–75 times less dangerous than VX, and in the case of Iranian 'Novichok', almost 1000 times less toxic. While the values for A-232 and A-234 are similar to the data obtained by Franca et al. (2019), the median lethal dose for A-230 is very different. Assuming that a "standard" person weighs 70 kg, the LD50 values for Novichoks were recalculated based on the sources of the above literature and summarised in Table 4.

Using the analogous FDA method included in the TEST programme (ver. 4.2.1), we obtained precisely the same results for Novichoks as in the work of Carlsen (2019). Furthermore, to estimate the LD50 parameter, we also used the Consensus method included in TEST (ver. 5.1.2) and the second software, QSAR Toolbox. As the results between the Consensus and FDA methods differ, the question is which is more reliable? The first is supported by the fact that it applies all QSAR methods included in the TEST to assess toxicity and is additionally recommended by the US EPA (2022). Furthermore, the consensus method was reported to be the most reliable estimate provided by the TEST software (Melnikov et al. 2016). On the other hand, the FDA method is backed by the generation of new models based on the closest analogues of the test chemical. The latest TEST software version (ver. 5.1.2) does not include the implemented FDA method. Furthermore, the LD50 values obtained using the QSAR Toolbox overwhelmingly correlate with the consensus method; only two compounds from the A series nerve agents were the values more comparable to the FDA method. Taking into account the information above, we tend to evaluate the results obtained using the consensus method, primarily supported by verification using the QSAR Toolbox, as more trustworthy. The LD50 values for Novichoks estimated in our work using TEST (Consensus method) and QSAR Toolbox mainly differs from the data discussed earlier and are included in Table 4. The only compounds whose LD50 values were similar are A-230, A-232, and 'Iranian' Novichok. According to our estimates, the most dangerous Novichok was A-232, in contrast to the values calculated by Carlsen (2019), where the nerve agent A-242 would be the most toxic, and Franca et al. (2019) suggested A-230 as the most toxic Novichok. Unfortunately, sources from the literature only provide LD50 for 2–6 organophosphorus compounds from the Novichok group. Therefore, our work is unique because it includes up to 17 such A-series nerve agents.

Attention should also be paid to extrapolating doses between species, converting the rat to a human oral median lethal dose. The allometric scaling between species for dose conversion from animal to human studies is one of the most controversial areas of pharmacology and toxicology. The allometric approach considers differences in body surface area related to animal weight while extrapolating doses between species (Nair and Jacob 2016). This article's conversion to human toxicity followed the guidelines for animal-human dose conversion based on body surface area, i.e., rat doses were converted to human equivalent doses by dividing the rat dose by 6.2 (Nair and Jacob 2016; Carlsen 2019). Science changes in phases, experiencing anomalies that lead to a crisis and revolution, resulting in a new, immature scientific paradigm that, over time, becomes the new normal (Hartung 2021). Toxicology has encountered a series of such anomalies that have led to a crisis. One of them is the generally accepted guide for dose conversions between species, which is not necessarily the right one. As evidenced by the various studies, for example, many inflammatory mediators assume very different roles in different species; for example, TLR4 signaling differs in humans and mice (Schmidt et al. 2010). The above studies prove that animals are not particularly good predictors of humans in areas where we have comparative data across species. In toxicodynamics, a well-known example is that humans are 1000 times more responsive to inhibition of Na/K-ATPase by the cardiac glycoside ouabain than mice (Kent et al. 1987). Moreover, the difference in susceptibility to bacterial endotoxins can be up to a million times greater in range (Hasiwa et al. 2013). Thus, the above examples indicate that humans are not 70 kg mice in toxicology (Leist et al. 2008). A study based on a broad system approach confirmed the low predictability of animal responses to inflammation (Seok et al. 2013). The low-level predictability of animal studies in research areas, which allows direct comparison of data between species, raises serious doubts about the usefulness of animal data as crucial tools for predicting human safety. Perhaps this is the reason for the differences in prediction, or perhaps it is another proof of the validity of Hartung's concept? (Hartung 2009, 2021). Regardless, these studies were essential as an initial screening before undertaking acute toxicity studies on animals, concerning reactive substances such as Novichoks.