AI and genome engineering: new directions in biology

Listen to this blog story:
Listen to "AI and genome engineering: new directions in biology" on Spreaker.

Leo recalls finishing high school when scientists announced the publication of the first draft human genome. This new scientific revolution intrigued him, but he preferred subjects with a stronger maths focus. He chose to pursue higher education in computer science and maths, moving between three different countries.

During his education in Estonia, he joined Jaak Vilo’s research group to help design algorithms for aligning DNA sequences to the human genome, which he genuinely loved because he could apply his strengths in computational methods.

This ignited his passion for bioinformatics, which he chose as the subject for his undergraduate thesis project. He completed his bachelor's in Computer Science, Engineering and Mathematics at Massachusetts Institute of Technology (MIT), US. After graduation, Leo moved to the UK to complete a quantitative genomics PhD at the Sanger Institute.

“It was great doing my PhD here. Huge amounts of sequencing data were being generated and made available online. For the first time, those sequences were linked to human phenotypes, such as gene expression measurements. I could see this path forward - that you could measure lots of DNA sequences, then see what the downstream biological effects are, from messenger RNA (mRNA) levels, protein levels and eventually all human disease traits. I found that very appealing and spent my PhD building statistical models for these data types, but also considering what experiments should be run to get the best data.”

Dr Leopold Parts,
Group Leader, Generative and Synthetic Genomics, Wellcome Sanger Institute

After he finished his PhD, scientists had a deeper understanding of how DNA influences gene expression levels, but the next important step was studying the impact on protein levels. This required new data, which inspired Leo to venture into experimental research, so he took a joint postdoctoral fellowship at the University of Toronto, Canada and EMBL Heidelberg, Germany. He continued this work at Stanford University, US, focusing on researching the genetics of yeast to understand how genotypes impact protein levels in cells. He enjoyed the freedom of generating and analysing his own data to answer important questions.

By around 2014, CRISPR gene editing had become a popular tool renowned for its flexibility and power. In 2015, Leo joined the Sanger Institute to set up a computational genomics research group and later moved into cellular genetics. His team focused on optimising CRISPR techniques for understanding human cells. Recently, Leo has been appointed as a Group Leader in the Generative Genomics programme. His team are currently exploring alternative ways to engineer genomes to understand the downstream effects in biological processes and systems. His work continues to be very interdisciplinary, combining computer science, mathematics, and biology to develop new approaches for studying and changing genetic data.

Gene editing allows scientists to make precise changes to DNA, such as removing, replacing, or modifying the genetic sequence. It has significantly advanced over the years, from the original labour-intensive mouse ‘knock-out’ techniques to the fast, scalable power of CRISPR.

The Sanger Institute has played a key role in influencing genome editing. In the 1980s, Allan Bradley – who later became Director of the Sanger Centre – worked in Martin Evans’ lab on mouse embryonic stem cells¹. Allan developed a technique to target specific genes in mice. When these genes were mutated to the extent they no longer worked, they were called ‘knock-out’ mice. This work led to Martin winning the 2007 Nobel Prize in Physiology or Medicine for the research underpinning knockout technology². These techniques equipped scientists to breed and grow mammals with particular characteristics, which was highly effective but painstakingly slow.

Use the arrow on the right (or left) to go forwards (or backwards) through the process.

The best-known gene editing tool is CRISPR/Cas9, the discovery of which secured the Nobel Prize in Chemistry 2020³. It acts as ‘molecular scissors’ to make precise cuts in a specific sequence of DNA. This technique transformed gene editing, making it faster, simpler, and more scalable. Scientists create ‘guide RNAs’ to target the DNA they want to edit, and an enzyme called Cas9 cuts the DNA at that spot.

Use the arrow on the right (or left) to go forwards (or backwards) through the process.

Further improvements to CRISPR-Cas9 arrived with the developments of base editing and prime editing⁴, which provide more precise gene editing by targeting single bases and reducing the unwanted off-target effects of CRISPR-Cas9. The rise of innovative artificial intelligence (AI) tools may continue to enhance the safety and effectiveness of gene editing tools.

“I believe we may be midway through a biological revolution. From understanding DNA, to mapping the human genome, to understanding variation across individuals, and now making biology an engineering discipline to design and create useful applications. That will slowly but surely impact all areas of everyday life.

“We can think of biology as being analogous to electricity. Maxwell’s equations arrived in the mid-19th century, but it took decades before you had wiring in every household with people benefitting from electricity. In a similar way, biologists will progress from figuring out how things work to being able to predictably engineer biology.”

Dr Leopold Parts

Artificial intelligence (AI) could help transform gene editing tools, offering greater precision, efficiency, and speeding up discovery. For example, the biotechnology company Profluent has developed OpenCRISPR-1⁵, an open-source AI-created gene editor. This CRISPR-like protein does not exist in nature and early work suggests it may show greater specificity in gene editing, although research is ongoing.

Leo’s lab is using AI to predict the outcomes of CRISPR edits. CRISPR creates breaks in DNA in specific places, and the resulting changes – whether single base substitutions, deletions, or base insertions – are often predictable. AI models improve this predictability, leading to more efficient and precise edits. For example, the team recently used machine learning to assess the success rate of prime editors.

AI tools also excel at interpreting the vast, complex datasets generated by genome sequencing. While the human genome contains 3 billion base pairs, these datasets are relatively small for AI, and its repetitive nature presents a further challenge. Researchers can introduce variation by engineering edits, which AI can use to refine its models and predict how genetic changes will affect biological functions – reducing the need for time-consuming human analysis.

Researchers can also use AI as a ‘virtual experimenter’. For example, if a scientist's predictions are not going as expected, the model can refine the predictions by including new data and running more simulations. This is more efficient than the time-consuming process of waiting for physical experiments to complete.

However, a challenge for researchers using AI is the complexity of biological systems. Even simple biological processes, like how DNA makes RNA, are very intricate, involving several elements such as promoters, enhancers, and insulators. Whilst AI handles this complexity well, it is crucial to distil the findings into clear insights. AI is similar to a ‘black box’, transforming inputs into outputs in ways that are not always easy to ascertain. So there is a need for further research into the inner workings of AI models so that humans can understand them, which is known as Explainable AI.

These advances in artificial intelligence (AI) have led to the emergence of the new field of generative genomics, where AI and gene editing combine to create new biological systems.

At the Sanger Institute, the Generative and Synthetic Genomics research programme is leading the field. Combining their expertise with Sanger’s high-quality, open datasets, the team are enabling researchers around the world to collaborate and benefit from their research. Two initiatives at Sanger that rely on gene editing and AI are the Cancer Dependency Map and the Tree of Life programme.

The bold and ambitious goal of Generative and Synthetic Genomics is to “solve biology” – predicting the function of any DNA sequence, such as whether it encodes a protein, how it interacts with other molecules, or whether it gets expressed in a cell. The team’s ultimate aim is to be able to avoid many gene editing experiments and replace them with AI modelling and prediction. For example, if a new mutation arises in a gene that scientists have not encountered before, the AI model could predict whether it will affect a person’s wellbeing.

Achieving this vision will rely on AI models that are trained by systematically altering DNA sequences and measuring the effects. Scientists can control DNA synthesis and have many techniques to enrich the DNA by introducing variation. By creating diverse datasets – whether through random DNA synthesis, variations from other species, or controlled genetic edits – researchers can build models to predict how DNA will function in different contexts.

Leo envisions big changes on the horizon. Ultimately, combining artificial intelligence (AI) with genome engineering may lead to biology becoming a more engineering-focused discipline, in which it is increasingly possible to design and create biological systems for specific tasks.

In the biotechnology sector, AI has the potential to supercharge innovation. For example, companies may investigate altering virus capsids, the protein shell that encloses its genetic material and attaches to cell receptors. If researchers can modify a virus to travel to specific tissues in the body, they could deliver medicines to specific organs such as the lungs or liver. This approach could be combined with targeted mRNA vaccines, such as the Moderna COVID-19 Vaccine. AI models can predict the most stable version of mRNA, which along with the newly designed capsid, could lead to rapid drug delivery.

“It feels like AI is liberating this ‘design for function’. You no longer need to experiment blindly. Instead, one can do a systematic scan, build an AI predictor, and then use that to generate new things that work as expected. I think it will be an opportunity amplifier – speeding up drug development and facilitating new discoveries. This could radically shorten the time it takes to move from an observation to a useful result. You can already use AI for engineering. We just need enough data to predict how a biological system works and then use it. And that's the thing with using AI, it can help you reap the benefits of research without fully understanding the underlying mechanisms. It’s now unwise to avoid AI because it has such power, but it's not clear where it will have the biggest impact right away.”

Dr Leopold Parts

As AI technologies become more sophisticated at enhancing gene editing tools, we step ever closer towards engineering biology. By combining the predictive power of AI with the precision of gene editing, researchers are exploring some exciting applications, from speeding up drug discovery and personalised medicine to designing entirely new biological systems for biotechnology.

Footnote:

For a simple guide to key concepts in AI, see our article: Using artificial intelligence for genomic research on the YourGenome website.