Biotechnology and the next pandemic part 2:

This is part two of a series on modern biotechnology and its impact on pandemics. Part one outlines the historical context of various biothreats, and part three will focus on the state of preparedness of Europe against this evolving threat environment.

From a policymaking perspective, it is not enough to recognise that evolving technologies may come with risks – a little knowledge of how these technologies work goes a long way. Here, we delve into some examples of what this advancement looks like in biotech and what these capabilities imply for the risk landscape.

Introduction

Picture this scenario.

“We are in 2030. Two students passionate about biology established their own laboratory in their garage. Initially driven by scientific curiosity, they quickly gained enough confidence to undertake more advanced experiments. Using a few easy-to-access tools, they are able to use AI to alter the genetic information of harmful bacteria and viruses; and then have these made-to-order by a third party. In other words, they established an in-house system for producing biological weapons.” 

Although this scenario is entirely fictional, many defence and biosecurity experts are concerned that fiction could become reality, especially with the recent advances observed in biotechnology. 

A key challenge for policymakers, but also for the public, is to make sense of the capabilities offered by rapidly emerging  biotechnologies in order to grasp how concrete such a scenario could be in the near future without needlessly catastrophising. Getting to that point  involves some grounding on how these biotechnologies work, interface with industries and are used in our daily lives. Armed with this knowledge, policymakers will be empowered to understand the biosecurity landscape and unlock biotechnology benefits for their societies in a safer way. 

Here, we get a closer look at what technological advancement actually looks like in the field of biotechnology, and explore how these advancements may alter the biothreat landscape. We will focus on two major aspects: first we will see how DNA synthesis technologies are evolving and, second, how the use of AI is transforming biotechnology research.

Shopping for DNA fragments: nucleic acid synthesis

DNA synthesis is the process of creating DNA, either naturally in cells or in a lab. In laboratories, scientists build DNA from scratch from basic building blocks called nucleotides . Multiple biotechnology companies offer highly-tailored DNA synthesis services – essentially the on-demand creation of DNA. The accessibility of made-to-order DNA services allowed scientists in 2002 to create for the first time a dangerous virus from scratch when they successfully built the poliovirus using only its genetic code as a guide. This study, in part funded by the US Defense Advanced Research Projects Agency, gave rise to significant debate on the dangers of manufacturing viruses and the implications for biological terrorism. Two decades later, synthetic viruses continue to be created, and the underlying technology and methods have become common practice.

Does this mean that anyone could access made-to-order DNA fragments and as such create its custom bioweapons? To answer this question there are three aspects worth  considering: first the capability offered by the DNA synthesis technologies, second its accessibility, and third the regulatory and biosecurity landscape.

What does this technology look like in the real world?

It is important to understand that DNA synthesis technology is self-limiting and error-prone. The larger the sequence being made, the greater the likelihood of a mistake in this sequence occurring. As such it is impossible to synthesise DNA fragments of an indefinite size. In simple words, one cannot just synthesise the full genome of a pathogen in one step. 

Recent advancements in the field allowed companies to optimise and scale the process enabling them to efficiently produce thousands of DNA fragments from 500 to 1000 nucleotides at once. But we are still far from being able to create large, virus-sized genomes made up of an estimated 1,700 to 2 million nucleotides. 

So, how are these DNA fragments turned into something functional and alive? Genome assembly methods enable us to stitch together these synthesised fragments into larger structures, capable of combining enough information to, for example, create an entire genome which can be introduced into a cell to create a virus. These assembly processes have been improved further and can now be done with minimal yet highly effective equipment.  We have reached a point where companies can combine, automate and scale both the synthesis and assembly steps – enabling the production of even larger nucleic acid fragments from 5000 up to 20000 nucleotides long. Such achievements would theoretically allow clients of such companies to order the synthesis of viruses like polio, HIV or Ebola. A key barrier that prevents the widespread use of these technologies is the sheer risk of handling these unpredictable agents. Conventional scientific practice makes use of specialised containment facilities adhering to strict biosafety – not having these protections acts as a form of deterrence for most non-expert actors. 

Size matters when it comes to creating DNA, because being able to compile greater amounts of information opens the possibility of building more complex structures such as larger genomes, new synthetic forms of life, and even creating artificial human chromosomes.  While Biotechnological capabilities are advancing at great speed, these technologies have not yet been fully embedded in manufacturing in such a way that new products are reaching markets. But the way these technologies have developed so far gives us reason to believe that over time it will be possible to assemble even larger genetic sequences, paving the way for the creation of even more complex pathogens.

How accessible are these technologies?

There are two main aspects to consider for accessibility: the cost of the tools and the skill of the user. 

While the cost of sequencing (reading) genetic information has been on a steep decline, the costs of producing this same material through synthesis techniques have not seen as rapid a decrease.  Reading a full human genome costs around €1000 while creating a genome of this size through DNA synthesis would theoretically cost around €300,000,000. In other words, writing DNA approximately costs 300,000 times more than reading it. Costs might still be accessible for synthesising small infectious viruses like influenza or coronaviruses (between €2,000 to €4,000) but they might escalate quickly for the design of much more complex pathogens like bacteria. 

But other parts of the DNA production chain – like data storage and biomaterials – are advancing quickly, meaning demand for DNA synthesis is expected to rise—thus driving down costs significantly over the next decade. Alongside these trends, traditionally large and costly lab machinery is being scaled down and made even more accessible in the form of benchtop DNA printers.

In terms of skills needed to use these techniques, there is a general misconception that biotechnology applications are out of reach to a lay person. While for a long time accessible just to researchers, the relevant knowledge base is now within the reach of anyone equipped with an internet connection; access to open-source literature and databases; and the ability to place an order for DNA and lab equipment. It is important to keep in mind that basic genetic and cellular manipulations like virus and bacteria cultures or DNA manipulation are becoming as accessible as following a recipe using commercially available products and kits.

What does this mean for biosecurity?

The decreasing technical barriers and uncertainties that once impeded quick and affordable DNA synthesis are being swept away, making the rapid creation of full viral and bacterial genomes faster and more attainable than ever before. 

As innovation ecosystems look to unlock the societal and economic benefits of the dawn of synthetic DNA, safeguarding these technologies becomes strategically important. 

It may come as a surprise, then, that any private individual can order the creation of DNA from labs without any legal oversight from national governments, and with no mandatory, government-enforced screening of these orders. Currently, companies and laboratories undertaking DNA synthesis technologies self-regulate. This governance approach was set by the creation of the International Gene Synthesis Consortium (IGSC) in 2009, which brought together several firms operating in the field to agree and deploy screening standards. While this is a commendable initiative, this left the regulation of a complex and high-consequence technology in the hands of industry with vested interests at play. Moreover, not even all DNA synthesis providers are part of the IGSC. And, even where providers are abiding by IGSC standards, recent MIT research showed that the protocol may contain major security holes. 

Global consensus is slowly being built around the need to safeguard these technologies. Firstly, the World Health Organization’s 2022 Global Guidance Framework for the responsible use of the life sciences pinpointed DNA synthesis as one of its seven illustrative scenarios to illustrate challenges and gaps in the governance of biorisks. Secondly, the scientific community recently recognised the DNA manufacturing ecosystem as a major point of failure of the current biosecurity landscape. In response, some countries are taking concrete actions to review their biosecurity strategies. The UK identified the democratised access to DNA synthesis as a key area of risk for future generations. The US also recently published a national framework for DNA screening and several bids are being made to the Senate in order to put DNA screening principles into law.

How artificial intelligence is complicating the picture

Creating DNA using the synthesis technologies we’ve outlined above requires having a precise set of instructions. Artificial intelligence is revolutionising how quickly scientists can process information and create these instructions. Here, we delve into two specific ways in which AI is shaping advancements in biotechnology, firstly through the use of biological design tools (BDTs) to produce the instructions for a desired molecule, and secondly the introduction of advanced artificial intelligence systems designed to assist with, or even autonomously conduct, scientific research – AI scientists.

Since the 20th century, scientists have looked to nature for the blueprints to create genetic molecules, analysing and re-creating known structures. BDTs are AI models that have been trained on a vast expanse of biological data gleaned from this earlier exploration. Equipped with this raw information, these models are able to design outputs such as proteins which can be put to use in living organisms in a number of ways. BDTs have spurred a number of innovations including the prediction of protein shapes and design of new vaccines and drugs. These advancements have culminated in significant recognition, including recent Nobel Prizes attributed to Demis Hassabis, John M. Jumper, and David Baker as major contributors to the field.

Applying AI to biological challenges holds great promise, especially for designing new drugs and therapies. However, these same groundbreaking techniques could be reverse engineered to create harmful biological components. Researchers demonstrated the ability to train AI models in order to design new toxins which were engineered to have particularly harmful characteristics, clearly outlining the misuse potential of these technologies. Cognisant of these risks, hundreds of researchers from across the scientific community decided to articulate an ethical approach to BDT-assisted science and signed an agreement that seeks to ensure BDT development will advance without exposing the world to serious harm. The agreement’s purpose is not to suppress or strongly regulate the development or distribution of BDTs but rather to regulate the use of equipment needed to manufacture new genetic materials, the logic being that BDTs are not dangerous in themselves and as long as the predicted sequences are left on a computer. It is the translation of these sequences to the real world through the use of DNA synthesis equipment that ultimately allows for the development of harmful biological entities. 

Adding a grain of artificial intelligence on top of DNA manufacturing capabilities, suddenly unlocks the capability to design and engineer entirely new biomolecules and organisms. The implications are important as the convergence of these technologies may contribute to enhancing the capabilities for the design and production of more complex pathogens than exist in nature, creating new and difficult-to-predict risks.

The most well known AI tools are based on Large Language Models (LLMs), these are programs that can predict and generate meaningful responses to queries by accessing vast amounts of information at speed. LLMs accessibility tonon-experts has led to several scholars exploring the question of whether LLMs could assist malicious users in creating dangerous pathogens. 

This question is still under debate as empirical evidence and research reports have put forward a range of views on whether the LLM capabilities are powerful enough to enable this activity. Nevertheless we expect the capabilities of these models to keep scaling and improving in the future. In line with a recent report from the Centre for Long-Term Resilience, we anticipate major advancements in how AI is used in biology. This growth in capability comes from the improvement of LLM models themselves, coupled with the convergence of LLMs and biological design tools and,  finally, the development of autonomous AI systems which can assist or conduct scientific research.

For several decades, scientists have sought to automate and where possible offload simple and repetitive parts of the research process to computers. Emerging AI capabilities have energised this line of inquiry. 

Earlier this year, research organisation FutureHouse produced the first version of an autonomous AI scientist assistant for biologists  that outperforms scientists on tasks like answering complex scientific questions, writing review articles, and detecting contradictions within the literature. Two institutions were able to use simple human language instructions to command multiple AI programs to interact with laboratory instruments and effectively plan, operate and automate a full end-to-end process for producing chemicals. 

It is possible that, in the next 5-10 years, we could see the emergence of AI models able to reason about biology, make hypotheses, troubleshoot experiments and interact with the physical world by orchestrating lab instruments. Such capabilities, if widespread, will almost certainly lower the barrier of entry and mostly exclude the need for tacit knowledge and years of research experience.

Conclusion

It is easy to see, in general, how biological threats could harm societies. But understanding the  precise ways in which biotechnologies are developing provides a better sense of “where to look” for emerging threats as policymakers. With this knowledge it is possible to articulate how the powerful tools at our disposal could lead to real, material risks. 

The barriers to entry to getting something meaningful out of these technologies are disappearing – we have to act before they are gone entirely.

The scenario described in our introduction might not seem so fictional anymore to the reader. No one wants a future where biothreats could be fully and easily engineered by any actors, without public oversight.

While some biotechnologies are still exclusively used by researchers and scientists, it is worth reflecting on the speed and scale with which these capabilities have so far spread among society. A recent study reported that advanced biotechnologies, once the exclusive domain of scientific experts, now take approximately 4.5 years to reach individuals with modest technical expertise and limited financial means. In the coming decade, authors predicted that this window will shrink to under 3.5 years.

Today’s most advanced technologies may be everyday tools tomorrow.