Diamond Age Data Science and the Boston Computational Biology and Bioinformatics Meetup recently packed the house for a discussion on one of the buzziest topics in science today: AI in computational biology and drug discovery. Taking a grounded perspective, the panel got real about what AI can do today, where it is still tripping up, and how scientists and companies can keep up.
Eleanor Howe, Founder and CEO of Diamond Age Data Science, moderated the conversation. Joining her were four experts with very different vantage points. Alex Zinoviev of Lilly shared how she leads the company’s mRNA platform and uses generative models to push sequence design into new territory. John Hutchinson of Diamond Age brought two decades of bioinformatics chops and the perspective of a superuser who leverages AI to accelerate his work. Sonia Timberlake, an independent consultant who guides companies and investors on AI strategy, offered the startup and VC angle. And Enoch Huang, VP of Machine Learning and Computational Sciences at Pfizer, added the seasoned perspective of someone who knows when to double down on computational approaches and when to tap the brakes.
The first question set the tone. What exactly do we mean when we talk about AI? The panel agreed it is not a stand-in for “anything done on a computer.” Here, AI means models trained on data – from predictive systems that recognize patterns to generative tools. This includes tools that help produce code, predict protein structures, and generate new molecular sequences based on previous data.
So where is AI actually delivering? Sequence design is a big one. With mRNA or synthetic proteins, there are no tidy biological rules to follow, and generative models can help chart the options and flag promising directions. AI can also be a savior for messy lab data. Scattered spreadsheets, cryptic metadata, and endless Slack threads can finally be pulled into something scientists can query and trust. Dashboards are another win. What once took weeks of coding can now be built quickly, letting researchers explore results in real time. And coding itself has gotten a turbo boost. Large language models have become go-to productivity boosters, turning ideas into prototypes faster than ever. But it’s important to note that AI-generated code isn’t flawless. Users still need to know how to code, review outputs carefully, and fix the inevitable mistakes.
AI has additional limits as well. The real bottleneck in drug discovery is not designing molecules, it is making and testing them. Virtual designs are worthless if they cannot be synthesized and validated. Then there is automation bias, that very human urge to trust a model just because it sounds convincing. Without rigorous validation, that shortcut can be costly. And unlike in software or language, there are no solid benchmarks to prove that AI in biology is delivering reliable results. Until that changes, testing will remain the weak link.
Looking ahead, the panelists pointed to breakthroughs worth watching. Larger training datasets could increase the accuracy of perturbation models, making them more broadly useful. Testing frameworks that inspire confidence from scientists, leaders, and regulators will be critical. And in clinical development, smarter tools for patient recruitment and site selection could ease some of the toughest bottlenecks to getting new therapies to patients.
All of this is also reshaping the role of scientists. Less time will be spent grinding out code. More time will be spent asking sharper questions, interpreting results, and marrying computational predictions with the messy realities of biology. Embedding computational scientists directly into discovery teams may turn out to be the fastest way to close that gap.
If your team is ready to cut through the hype and put AI to work in ways that actually matter, Diamond Age can help. From strategy to implementation, we partner with biotech and pharma leaders to turn complex data into insights that drive discovery forward.
Let’s talk.