The 2019 coronavirus (COVID-19) pandemic revealed an urgent need for the acceleration of vaccine development worldwide (1). Within months of the start of the COVID-19 pandemic, several pharmaceutical companies announced their objective to develop vaccines against COVID-19 by using different technologies (2–4). Each of the technologies presents advantages and pitfalls regarding the development of efficient COVID-19 vaccines (5–7). However, the situation requires that all the alternatives should be explored to shorten the course of the COVID-19 pandemic (8, 9). An increased speed of development induces risks, which must be taken seriously (10, 11). Hence, we analyzed the risks of the different technologies to produce the vaccines proposed by the pharmaceutical companies providing vaccines against COVID-19. This analysis was performed using scientific publications from early 2020 and utilizing the artificial intelligence (AI)-based search engine Risklick (12, 13). By calculating the risk score using a Probability and Impact Matrix and a semi-automated AI interface, we estimated the major risks faced by the pharmaceutical companies providing the vaccines.

A year later, our risk analysis on COVID-19 vaccine development was compared to the contemporary situation. Overall, we observed that the AI-based analysis anticipated the shortage of production and distribution faced by the pharmaceutical companies providing the RNA vaccines. The AI-based analysis highlighted the other issues faced by the pharmaceutical companies providing the vaccines, such as the delayed arrival of vaccines based on technologies other than RNA. Each issue faced by the vaccine providers is, as expected in the analysis, connected to the technology of their vaccine. Hence, our analysis exposes how AI-based technologies could become essential to the future of treatment development and the anticipation of associated risks.

The COVID-19 pandemic revealed the necessity for accelerated vaccine development worldwide (3, 7). In this critical context, we performed a risk analysis on the different vaccine technologies proposed by the manufacturers against COVID-19 in the early months of 2020 by using the AI-based search engine Risklick (12, 13). We graded seven different risks across seven different vaccine technologies by using a Probability and Impact Matrix (as illustrated in Supplementary Figure 1, for the risks linked to safety for each vaccine). Then, the risk score of every risk category for each vaccine technology was calculated. Risk calculation was limited by the low number of available publications at the time. An ideal quantitative risk assessment would require more data to reduce the limitations and strengthen the conclusions of the analysis. This resulted in a table of risks for each individual vaccine technology (Table 1). In this table, we already noticed that each vaccine technology possessed strengths and weaknesses different from one another, indicating the potential critical issues for the development of future vaccines. To simplify the visualization of the risks, a resume of risks per technology was performed using radar charts (Figure 1). To allow comparison, all technologies of the vaccines were compared to virus-like particle (VLP). This choice was made in the perspective of VLP being the least probable vaccine technology to effectively reach the market against COVID-19 in the near future. There are currently only four VLP-based COVID-19 vaccines in clinical trials and even in the optimistic cases, production could not reach more than 100 million doses before 2022 (16). This was confirmed in analysis with VLP presenting a particularly high risk for the time to availability and manufacturing scalability.

In June 2020, we already observed that the pharmaceutical companies providing vaccines would face different issues. Taken separately, DNA-based vaccines have the highest risk of providing only short-term immunity (Table 1, Figure 1). Shipping is a particular risk for the RNA-based vaccines and, to a lesser degree, viral vectors, which require extremely low temperatures. Inactivated viruses and live attenuated viruses raised concerns regarding their safety and the cost of goods sold. Protein subunits appear less risky across most categories, but the duration of immunity and time of availability still appeared as a considerable risk. Finally, the viral vectors presented serious risks connected to production, particularly for the categories of th manufacturing scalability and manufacturing process.

After 1 year, we compared our results to the contemporary situation (May 22, 2021). Presently, 18 different vaccines are approved in one or more countries (17). When selecting the vaccines with authorization in more than two countries, this number falls to 10 (three inactivated viruses, five virus vectors, and two RNA-based vectors).

First, we noticed that an extremely low number of the vaccines we classified as “high risky” managed to reach market authorization so far (these being inactivated virus, live attenuated virus, and VLP; Table 1). Moreover, based on the WHO database (18), the most widely represented technology in the clinical phase of the vaccine development is the protein subunit, which was also the least risky technology according to our analysis (Figure 1). However, none of the vaccines using this technology have reached market authorization to date. In fact, one of the major risks of the protein subunit compared to the competitors was its relatively important time to availability (Figure 1), which was much lower in the RNA-based vaccines and viral vectors. Hence, the major risk associated with the protein subunit vaccine might be the reason why vaccines based on such technology are slower to reach approval compared to vaccines based on the other technologies, despite the high number of clinical trials running using this technology at the time this article was written.

With respect to the other technologies, RNA-based vaccines were rapidly commercialized probably due to low cost and high scalability of production, which translate into an extremely low risk for the time to availability (Supplementary Figure 1). However, their high risk associated with shipping was verified when supply chains started suffering from storage temperature issues following market authorization (19). With respect to the viral vector technologies, the major risk was associated with manufacturing. This was demonstrated by the shortfall of the Oxford/AstraZeneca doses in Europe for 2021 (20) and Sputnik in Latin America (21). Hence, the major risk calculated in 2020 appeared to play a critical role in the deployment of the viral vector vaccines today as it did for the RNA vaccines. With respect to vaccine distribution, the early stage of vaccine distribution in 2021 highlighted difficulties with respect to the cold chain distribution process (22). This situation provoked the evident risks of supply chain failures and caused an important pressure on vaccine supply chain infrastructures (23, 24). Such risks will need to be addressed in the future, particularly from the perspective of mRNA vaccine distribution (25). Altogether, the risk analysis performed using the Risklick AI in early 2020 offered an accurate prediction of risks that would be faced by the different pharmaceutical companies providing vaccines in the year 2021.

The COVID-19 outbreak resulted in one of the biggest waves of publications in the history of modern science (26, 27). This influx of COVID-19-related data made literature retrieval and monitoring one of the greatest challenges of the pandemic (28). The COVID-19 pandemic highlighted the need for vaccine developers and health authorities to provide general guidance for the faster development and preparation of vaccines while offering the highest safety and efficacy standards (1, 29). Shortly after the start of the COVID-19 pandemic, several pharmaceutical companies providing vaccines announced their objective to develop vaccines against COVID-19, using different technologies, all within a matter of months (2, 3).

The choice of the type of vaccine is crucial: carrier or vector, adjuvant, excipients, dosage form, and route of administration; all directly impact the potential efficacy of the vaccine against COVID-19, but also the logistics of manufacturing, storing, and distributing the vaccine might affect the availability of the product (5, 6, 9). Hence, this simple variation in the composition of the vaccines will directly impact the policy of mass vaccination and the access to vaccine doses worldwide, affecting the duration of the COVID-19 pandemic (8).

In this context, we performed a risk analysis on the different technologies used by the pharmaceutical companies providing vaccines against COVID-19 in the early months of 2020 by using the AI-based search engine Risklick (12, 13).

Our results demonstrate that the AI-based analysis of risks allows the identification of the vaccine category with the highest chances of success. Moreover, the Risklick AI allowed us to foresee the different issues pharmaceutical companies providing vaccines would face, notably during the production and distribution phases of their product.

Our results highlight how AI-based technologies will become essential in the development of various therapies in the future. Moreover, our results demonstrate that, by using in-depth risk analysis, the global performance of vaccine developers could be increased by anticipating major issues, allowing corrective, and preventive actions. Such an approach, if generalized, would represent an important gain of time and efficiency for every scientist and manufacturer involved in providing care and drugs. The adoption of detailed risk analysis by all the scientists and manufacturers involved in the COVID-19 pandemic could represent a game-changer that could positively impact the future of therapy development worldwide.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

PA designed the study. QH and NB wrote the manuscript. QH, PA, and LM performed the experimental work. DA, SF, NB, and DT developed the AI models. All authors had full access to the data, helped draft the report or critically revised the draft, contributed to data interpretation, reviewed, and approved the final version of the report.

This study was funded by the Innosuisse project funding number 41013.1 IP-ICT.

QH, NB, LM, and PA are work for Risklick. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.