AI and genomics pave the way to new data-driven scientific discoveries

Ronnie was surrounded by science from a young age. Image credit: Ronnie Crawford.

I grew up in a tiny village west of Cambridge and, from a young age, was surrounded by science. Many of my family members were scientists, and my mum loved doing kitchen experiments with me. When I was eight, I brought a book into school on immune-cell interactions for show-and-tell. Biology soon became my passion!

After leaving school, I chose a combined undergraduate and master’s degree in biomedicine at the University of York. I initially disliked computer science. We were required to use a programming language for statistics and graphics called R, and I was terrible at it. But later, I started building video games using the Java programming language as a hobby, which made me realise the issue was how coding had been taught to me.

After graduating, I joined the Sanger Institute to work in a laboratory, wrapping up experiments as the mouse facility was closing. I thought this would provide useful experience for a biomedical career, but I found the work did not suit me, and it put me off a lab-based path. However, I became increasingly interested in how others at the Sanger Institute were using computer science. I wanted a career that supported biomedicine but took me in a fresh direction; I had also considered science communication. After some deliberation, I decided on a second master's in computational science. This time around, I enjoyed computer science, especially AI, even before ChatGPT became popular.

Whilst studying part-time for my second master’s, I joined Sanger’s Tech Talent Scheme, which offers work placements in IT and informatics teams. During the two-year scheme, I sampled different roles through three rotations, working with Dr Vivek Iyer (Human Genetics Informatics), Dr Jake Almagro-Garcia and Dr Richard Pearson (Genomic Surveillance Unit), and John Boyle (Science Solutions). The teams helped me start small and build my computational skills over time. I am very grateful to everyone who shared their knowledge and helped me thrive.

I work in the Informatics and Digital Solutions (IDS) team, led by the Institute’s Chief Information Officer, James McCafferty. My role has two main strands: supporting researchers by troubleshooting AI and machine learning challenges and developing computational models within Ben Lehner’s Generative and Synthetic Genomics group. This dual focus aligns with the Institute’s AI for science strategy, which combines infrastructure and discovery.

In my operational work, I meet researchers to help solve their research problems. Scientists tend to fall into two groups: those comfortable with machine learning, who typically know precisely what they need to solve a problem, and those less experienced with AI but know what outcome they want. I especially enjoy working with the second group because these requests demand more creative problem-solving. I love tracking down the appropriate tools and techniques to address a scientific problem. I have also gained a reputation as “The AI Guy”, which brings a flood of queries – some beyond my scientific remit. I have to remind people that questions about chatbots and Large Language Models (LLMs) like ChatGPT go to the Operational Excellence team.

For my research, I spend most of my time sitting at a desk coding. I use data generated by Ben’s group, which focuses on building predictive AI models to understand biology. The team has made me feel like an integral member, and experiencing ‘academic life’ first-hand has helped me consider my longer-term goals. Ben also leads a research team at the Centre for Genomic Regulation in Barcelona, and I work closely with them too. I love method development; despite sounding rather dry and robotic, it is highly creative and lets me explore many ideas and tool combinations. I often meet people who trained as biologists, like me, and are considering whether to dabble in coding. I try to be encouraging, because although it looks intimidating, it is less scary than it seems.

I am building an AI model that predicts a protein’s stability just by looking at its amino acid sequence. Proteins are long strings of amino acids – each represented by a letter. Just as ChatGPT predicts the next word in a sentence, I use a base model from Meta that can predict the next likely segment of a protein sequence. While the AlphaFold protein structure database revolutionised protein structure prediction, there is still a gap when it comes to predicting how stable proteins are – how long they last before breaking down. The base model infers structural features; I then feed its mathematical representation into my local model, which estimates where proteins sit in space and judges their stability. Eventually, this work could be used to help design better drugs, understand how protein changes lead to disease or design more stable proteins for use in lab experiments. Ben’s team has a cute tradition of using Asian food-themed names for their models, so I named my model DUMPLING (Delta of Mutations in Protein Language embeddings for INdel effect predictinG).

I enjoyed this project so much that it has inspired me to focus on research in the next stage of my career. I am currently looking for PhD programmes in Cambridge that combine method development with meaningful biological questions.

In 2023, I learnt that the Institute’s staff machine-learning community had dissolved during the pandemic. Colleagues wanted it back, so I approached my manager, John Boyle, about reviving it. John felt the community would be a great way to connect our team more closely with the research programmes. I launched a series of engagement events and talks with the BioDev Network – and you can read about one of the AI hackathons. With John’s support, the community has grown to a couple of hundred members.

Later that year, I planned a short series of lightning talks, but John nudged me to build a full-day event. I am glad he did because it pushed me out of my comfort zone. As soon as it ended, many people asked when the next one would be. The idea snowballed into a multi-day event, and before I knew it, we were planning an Explainable AI conference with international experts.

Organising a large-scale conference is challenging, and public speaking ties my stomach in knots. But I wanted to face that feeling, so I took the opportunity and gave the opening address, and the conference was a success.

Working across the Institute’s research programmes, I have seen different routes to scientific discovery. While all researchers aim to advance biological understanding, some lead with data, others with theory and others with translation. Understanding these approaches makes collaboration easier and has sparked my fascination with learning the philosophy of science. There is a strong ethos of learning together here – it is not about one group telling another what to do.

That collaborative spirit extends beyond the research. At a previous Sanger Startup School, a scheme that supported scientific innovation, I enjoyed working with colleagues to explore and pitch genomics and biodata projects in an entrepreneurial setting. Our team proposed an AI-bias validation platform to help researchers assess bias in their AI models before deployment. The scheme taught me how to sell ideas and think commercially – skills I will carry forward, whether or not we decide to revisit and launch the platform.

It feels like I pick up a new hobby every month – I love trying new things. I am a dabbler rather than a deep-dive expert in any one thing. I enjoy making music: I learnt the trumpet growing up, and I am slowly learning guitar; I also like making electronic music. I spend time on puzzles and gaming, and I recently built my own computer. Boredom is my nemesis, though I know I need to make more space for rest.

Stop worrying about making the ‘right’ choices. Do different things, and remember, you can always change direction later. That eight-year-old with the immune-cell book could never imagine coding AI models for a living!