Living organisms and AI may seem worlds apart, yet both share a fundamental reliance on data to determine their function and enable functionality. In organisms, that information is stored in DNA sequences, or genes. For AI, the information is largely external and typically delivered in massive troves as training data (though it is also found in the code that governs models' algorithms).  

The fit between biology's wealth of data and AI's input requirements is evident. And it promises to be a boon for a relatively new field of science, called synthetic biology, that is concerned with reading (sequencing), editing (synthesizing), and writing (printing) DNA to create new entities. Synbio, as it is known, is already changing industries, such as manufacturing, pharmacology, and agriculture. And its influence will continue to spread as genetic innovations and inventions give rise to new applications, including in areas such as renewable energy, healthcare, foods, and crops.

AI's ability to sort data and find meaningful (and often novel) connections means it has quickly become a central element of synbio's technology platform, which, with engineering and biology, is part of the three-component loop that powers this emerging field of science (see figure 1).

Synbio, has traditionally been expensive and slow. But that is changing as technological advancements at the cross-roads of genomics and digitization push the science in a variety of ways. For example, the digital domain's relatively easy and low-cost data manipulation has created unprecedented economies of scale, accelerated development of new processes, and cut the cost and length of experiments. This is evident in the cost of DNA sequencing. In 2001, the first human genome was published after more than a decade of work and at a cost of about $100 million. That same process can now be done in a day for about $600, according to the National Human Genome Research Institute. 

AI promises to provide synbio with the next technological leap forward. Advancements in machine learning and the improving quality and quantity of data have the potential to generate new and useful genes, speed analysis of possible applications, and conduct (low-cost, low-risk, and rapid) virtual testing.

An example of those possibilities emerged in April 2024, when researchers published a paper describing a large language model (LLM), dubbed CRISPR-GPT, capable of automating and enhancing gene editing experiments (for more on large language models see "Language Modelling: The Fundamentals," Jan. 23, 2024). Such a model could enhance our ability to edit DNA sequences known as CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats), which can be used to cure diseases (and potentially reverse aging). The potential is clear, but the advances come with ethical issues. They include the possibility of illegal altering of human genomes to enhance traits and, because edits can be passed to future generations, the possibility of compounded mistakes and questions of consent.

New developments in AI, and notably the advent and improvement of generative AI, have opened the door to additional creativity in synbio. For example, LLMs, which use data and algorithms to generate meaningful text responses to queries, have been adapted to the lexicon of biology by replacing words with nucleotide bases, such as adenine, cytosine, thymine, and guanine. This enables the LLMs to optimize experiments to generate new DNA sequences (and thus new virtual organisms) precisely, quickly, and cheaply, in response to human prompts. The resulting molecules and organisms promise to be useful in accelerating drug discovery, food engineering, and climate remediation. For instance, experiments and studies suggest that microalgae could be engineered to (more effectively) remove toxins from the air, treat heavy metal pollution, and for desalinization.

Generative AI could also be used to predict the outcomes of gene editing experiments, reducing time spent investigating eventual dead ends, broadening the scope of testing, and delivering resultant savings in cost and time. Those applications are similar to software engineers' use of generative AI to test code.

Our usual practice is to identify key trends watch over the course of a decade. But the likely speed of development and the breadth of potential inherent to the combination of AI and synbio is such that looking beyond a five-year horizon seems both overly adventurous and a step too far into speculation. 

With that caveat, the key trends that we will be watching are:

• The growth of industrial biomanufacturing: This implies success in the scaling up of the production of biological materials and processes. The Internet-of-Things' (IoT) sensors and other devices should contribute to real-time collection and monitoring of biological data, which AI models will process accurately and at scale. Also, digital twins (virtual replicas of physical process or system) will enable the testing of biomanufacturing processes in a virtual environment and experiments on organisms behavior in different environments and outside the laboratory. This should significantly reduce production costs and improve the efficiency of biomanufacturing processes. The application of industrialization and scalability to DNA printing should also support next-generation DNA printers to improve precision and add automation, which will facilitate the scaling up of processes to create organisms and other bio materials. 

• Support for the circular economy: AI, IoT, and data analytics could be catalysts for a thriving circular economy, particularly by enabling more sustainable use of renewable resources to minimize and recycle waste, and by supporting energy generation from waste. Generative AI could facilitate innovation in biomaterials design and industrial production at affordable prices, supporting local production and the growth of biological goods. Taking a step up, we think AI could become the engine for a bioeconomy (based on products, services and processes derived from plants or microorganisms) and thus help to address issues related to food security, climate change, and sustainability.  

• Big data in biology: Multi-omics is the aggregation of different types of scientific data, including information from proteomics, genomics, metabolomics, and transcriptonomics. Multi-omic data can be used as data to train AI models in similar ways that we currently use text, video, images, and code as inputs into multi-modal generative AI models. The combination of multi-omics with AI, will give scientists a more holistic view, for example, of the human body. And that could be still further augmented with data such as X-ray images, patient health records, and personal and family history.