Note: this is the 2nd article in a series of articles discussing radvac.org related biosecurity initiatives. In my last article I discussed using yeast to create "vaccine factories in a tube".  This post was also cross-posted on my Substack. 

Speed matters during a pandemic. Maxwell Tabarrok has estimated that if the FDA’s emergency use authorization of Pfizer’s COVID-19 vaccine had been done four months earlier, then about 240,000 lives would have been saved.^[1]

After the sequence of the SARS-CoV-2 virus became widely known on January 10th, 2020, Moderna designed an mRNA vaccine within hours and started manufacturing the vaccine five days later. Unfortunately, another ten months and three weeks passed before Pfizer’s mRNA vaccine was given a limited emergency use authorization by the FDA. It would be another four months until the vaccine was available to all adults in the United States.

The Rapid Deployment Vaccine Collaborative (Radvac) was founded in 2020 to develop novel platforms for vaccine production and dissemination. Radvac pursues open-source vaccine technology that enables rapid and radically decentralized design, synthesis, and distribution of vaccines in emergency situations. Radvac has developed and published twelve iterations of an experimental intranasal peptide-based vaccine. Radvac recently started researching yeast-based vaccines, something I wrote about earlier.

Radvac recognizes that vaccines have limitations, even if the problems surrounding rapid manufacture and dissemination are solved:

• Fear and skepticism of vaccines remains strong among the general public.
• Adaptive immunity takes about one to two weeks to be realized after vaccination, due to hard-to-change aspects of the human immune system.
• Many pathogens have proven to be stubbornly hard to vaccinate against. The canonical example is HIV, a deadly pathogen for which there is no approved vaccine despite over four decades of research and multiple Phase III trials. Hepatitis C has also resisted thirty years of vaccine research. The development of a vaccine for RSV took 60 years and required a breakthrough in understanding the conformations of RSV’s fusion protein.
• Vaccines may greatly reduce mortality but fail to prevent illness and spread. (cf. the COVID-19 and influenza vaccines).

In light of all this, Radvac has been exploring techniques for the rapid identification of antivirals, with a particular focus on the rapid repurposing of existing drugs and generally recognized as safe (GRAS) compounds.

This work recently culminated in a 60 page white paper, recently published on arXiv, where we survey the landscape of open source data and tools for in silico drug discovery. We also build a custom dataset of viral proteins and benchmark 15 binding affinity prediction tools on it, including our own custom fine-tuned model.

During the spread of an emerging pathogen, the ideal scenario is that in silico tools and in vitro screens lead to the rapid identification of one or more compounds with pre-existing safety records and wide availability that can be deployed as antivirals immediately.

Here I focus on biosecurity strategy for antiviral small molecule drugs, but it’s worth mentioning that in the introduction of the white paper we discuss the pros and cons of nine different classes of antiviral agents.^[2] In addition to small molecules, Radvac also has an active interest in peptides and short oligonucleotides, including small interfering RNAs. Radvac is also researching lectins, in particular FRIL lectin, which is an antiviral protein.

Drug repurposing

Drug repurposing has a long history.^[3] However, there are poor economic incentives for companies to invest in repurposing efforts. The FDA currently only grants three years of marketing exclusivity for repurposed drugs, or seven years for drugs that treat rare diseases. While a company might get three years of marketing exclusivity, if the drug is off-patent (likely) then other companies can still produce and sell it at low cost in a generic form. Doctors then naturally prescribe the cheaper generic form off-label for the repurposed application. This is why there is little economic incentive for companies to invest in repurposing. 

While finding a new chemical entity may be harder and more laborious, it is usually more profitable as the end reward is 10—15 years of exclusive rights for both manufacture and marketing. The lack of economic incentives for industry to focus on antiviral repurposing means that academic and non-profit research in this area is especially important.

During the pandemic, many existing anti-inflammatory drugs were quickly and successfully utilized to treat severe COVID-19. Academic researchers also explored repurposing existing drugs as antivirals. A more detailed accounting of humanity’s efforts to repurposed drugs as antivirals during the pandemic will be the subject of a separate Substack article. The effort yielded one definite repurposing success story -- baricitinib, a drug first approved by the FDA for rheumatoid arthritis. Baricitinib treats COVID-19 via two mechanisms - the first being immunosuppression (helpful for preventing the notorious cytokine storm) and the second being a mechanism that blocks SARS-CoV-2 from entering cells. During the pandemic the FDA issued an Emergency Use Authorization for baricitinib in combination with remdesivir (they reasoned it was necessary to combine baricitinib with an antiviral due to its immunosuppression).

Fraudulent trials, low quality research, and proliferation of pseudoscientific misinformation made the repurposing of hydroxychloroquine and ivermectin very popular, even after further research established they are not effective. Problems with cell culture screens created numerous false positives, leading to wasted downstream resources. We worry that these negative experiences during the pandemic have unfairly tarnished the reputation of drug repurposing.

Broad-spectrum antivirals

Broad spectrum antivirals (BSAs) are drugs that exhibit antiviral activity across several virus families. One of the most widely touted classes of BSAs are nucleoside analogs. These are drugs that mimic one of the base pairs of RNA or DNA, causing the virus’s replication machinery to jam. The approved nucleoside analogs ribavirin, remdesivir, and favipiravir all have established broad spectrum activity in cell culture across several viral families.

Nitazoxanide, an FDA-approved antiparasitic medication, inhibits the replication of a broad range of RNA and DNA viruses in cell culture. Nitazoxanide is one of the best validated BSAs since there are several clinical trials supporting varying degrees of efficacy against several viral families, including influenza, rotavirus, norovirus, hepatitis B, hepatitis C, and SARS-CoV-2.

In 2022 researchers reported that neomycin, an active ingredient in antibacterial ointments, triggers the innate immune system to produce a variety of antiviral factors in the nasal mucosa, mimicking the effects of interferon alpha. While its mechanism of action is different than traditional virus-directed antivirals, intranasal neomycin can be considered an emerging BSA.

Unfortunately, BSAs are generally only weakly effective, almost as a rule. For this reason some BSAs have struggled to obtain regulatory approvals. Although the FDA is not legally required to demand that a new therapeutic is better than the existing standard of care, in recent decades the FDA has evolved toward requiring that drug developers demonstrate a definite improvement over existing treatments. Unfortunately, this can make it very hard for BSAs that are weakly effective against many viruses to be approved by the FDA. For example, although the BSA “poster child” remdesivir showed efficacy against Ebola virus in preclinical work, its Phase III trial was terminated early due to clear inferiority compared to existing monoclonal antibody treatments.

Antiviral combinations

Many antivirals are only weakly active, and thus are left on the wayside by the pharmaceutical industry. Likewise, as discussed above, broad spectrum antivirals and repurposed drugs tend to be only weakly effective.

During the pandemic, experimental repurposing screens showed several weak hits (micromolar potency), but no strong hits (nanomolar potency). Many GRAS supplements and herbs were studied for SARS-CoV-2 activity, and research suggests that mint and licorice root may have weak antiviral potency against SARS-CoV-2 (see Chris Buck’s Substack article on mint).

This motivates combining several weak antivirals to get something that can actually be used prophylactically or to treat an infection. Drugs can even be synergistic in combination -- achieving a greater effect than would be expected assuming they acted independently. Also, when antivirals that act on different parts of a virus’s replication pathway are combined, it becomes much harder to evolve resistance to the treatment.

As with repurposing, combinations can be tricky for industry to tackle, due to the complexities of competition and the cross-licensing agreements that might be needed. Combination antiviral therapies have become standard for HIV and hepatitis C virus, but only as a matter of necessity since both are very rapidly evolving viruses. Combination therapies remain little explored for other viruses and for pandemic preparedness in general.

There are several challenges to designing antiviral combinations. First, the side effect and toxicity profiles of each compound need to be well-understood. If side effects or toxicity profiles overlap, the burden might be too great, possibly causing more harm than good. As an example, both hydroxychloroquine and azithromycin prolong the heart’s QT interval by blocking the hERG potassium channel, and this overlap makes that combination dangerous.

Another challenge when designing drug combinations is predicting drug-drug interactions. Drugs might work independently, synergistically, or antagonistically. While antagonism might seem unlikely, it is common enough to warrant serious concern — in cell culture it seems to crop up around 10-20% of the time, and there are additional mechanisms for antagonism in vivo.

To optimize combinations, brute-force screening in cell culture is difficult due to combinatorial explosion – if considering 100 compounds, the number of two-drug combinations is 4,950 and the number of three-drug combinations is 161,700. Pooled screening is a technique for efficiently screening combinations, but to our knowledge within the antiviral domain it has only been applied to HIV.

In general, combination therapies must be designed strategically, using knowledge of biological mechanisms, metabolic pathways, and established side effects. Large language models might be particularly useful here given their extensive in-built knowledge of biological pathways.

Conclusion

Antiviral drug discovery is hard, and we shouldn’t expect to discover a silver bullet drug during the next outbreak. However, it is very likely there will be some existing or investigational drugs or GRAS products that have weak antiviral activity. If taken in isolation, these compounds would do almost nothing. However, it’s possible a carefully constructed combination could be quite powerful.

In light of all this, developing better methods for designing antiviral combinations appears particularly high leverage. Toward that end, there are several promising avenues for work, including building better public domain datasets,^[4] custom AI agents for combination design, and new techniques for high throughput screening of combinations.

Further reading

• “Benchmarking open-source tools for in silico antiviral drug discovery” - Daniel C. Elton and Preston W. Estep, arXiv pre-print, May 2026.
  
   

1. ^{^}

   Tabarrok’s model predicts a range between 130,000 - 350,000 lives lost due to a four month delay in vaccine availability. Four months earlier would have been around August 10th, 2020. This would have been 20 days after Pfizer submitted positive Phase I/II data on the safety and immunogenicity of their vaccine to the journal Nature. Decades of work with vaccines has shown that immunogenicity is highly correlated with efficacy. See my timeline of vaccine EUAs for more details. Tabarrok’s work is consistent with other modeling work on the detrimental effects of vaccine delay during the pandemic.

2. ^{^}

   Antiviral classes: neutralizing antibodies, antiviral small molecules, soluble human receptors, antiviral CRISPR/Cas systems, interferons, antiviral peptides, antiviral proteins, antiviral nucleic acid polymers, antisense oligonucleotides/small interfering RNAs, and aptamers. Soluble human receptors, CRISPR/Cas systems, and nucleic acid polymers are all very much in the early stages of development. As reported in Science in 2025, researchers have developed synthetic carbohydrate receptors with broad-spectrum antiviral activity in vitro, but much more work is needed to prove out the concept in vivo, and receptors are expensive and difficult to manufacture. You may be wondering about the utility of antibodies for biosecurity. While there are many exciting new AI-enhanced tools for antibody design, antibodies are still expensive and difficult to manufacture and deliver at scale. Antibodies also require cold storage and delivery via IV or injection, making rapid dissemination challenging. Nasal antibody delivery could change that, but is very challenging. Generally, viruses can quickly evolve resistance to antibodies, as was observed during the pandemic. While there is ongoing research on broad-spectrum antibodies and antibody cocktails, it is still early stages. See our whitepaper for more about the pros and cons of different antiviral modalities.

3. ^{^}

   The antidepressant Bupropion was repurposed as a smoking cessation agent by GlaxoSmithKline in the late 1990s. Famously, the two most popular erectile dysfunction drugs on the market today both started as cardiovascular-related medications.

^{^}

We discuss several open source datasets containing binding affinity and in vitro screening data in our white paper, such as bindingdb, AntiviralDB, drugvirus.info, SMACC, and others. What is missing is a dataset containing information from the scientific literature on clinical trial results, side effect profiles, animal studies, etc. EHR data could also possibly be leveraged to gather information on side effects, and (through clinical trial emulation) possibly effective antiviral drugs and combinations.

Show all footnotes