By Dr Neil Taylor
BioKorea 2026 showcased many of the trends currently shaping drug discovery, from AI-driven innovation to emerging therapeutic modalities.
There were many great presentations throughout BioKorea 2026, and two in particular brilliantly cut to the point. Different speakers, different contexts, from different organisations, but they belonged together.
The first came from an Eli Lilly presentation on accelerating drug development through AI-driven open innovation. The speaker represented their framework by displaying a set of concentric circles: at the core, and dominating the slide, “data” (scientific artefacts) carefully structured and curated. On top of that, an orchestration layer of agents. Finally, the outer layer, domain-specific skills and applications.
The visual logic was the point. Everything in the outer rings depends entirely on what’s at the centre. Strip away the applications, the agents, the interface and what remains is data. Scientific data. The quality and scope of that data determines everything about what the system built on top of it can and cannot be trusted to do.
In a conference week full of AI enthusiasm, this was a quietly grounding statement. It didn’t diminish the potential of AI in drug discovery. It located that potential correctly. The organisations advancing most meaningfully with AI aren’t those with the most sophisticated architectures. They’re the ones who have done the painstaking, unglamorous work of building scientific data assets that are genuinely fit for purpose – curated with domain expertise, structured with scientific intent, and validated against experimental reality.
For structural biology, protein structure data and structure-based drug discovery specifically, this framing is clarifying. Protein structure data is not created equal. A crystal structure deposited with well-documented experimental conditions, appropriate resolution, and expert curation carries fundamentally different informational value than one processed at scale without contextual judgement. AI systems trained on the former can be trusted in ways that systems trained on the latter cannot. The data layer is not a prerequisite to be cleared on the way to the interesting work, it is the interesting work.
The second slide came from an Amgen presentation and displayed something visually striking: thirteen distinct therapeutic modalities arrayed across a single screen, encompassing small molecules, bispecific antibodies, PROTACs, CAR-Ts and TCRs, siRNAs, peptibodies, oncolytic immunotherapy viruses, and antibody-drug conjugates. The full breadth of what modern drug discovery can strive for and create.
The principle underneath the visual was Biology First, Modality Second. The modality best suited for a target is the one that the biology demands – not the one a platform company has optimised, not the one that’s attracting the most capital, and not the one that worked for the last programme.
This is a principle that sounds obvious but is routinely violated. The history of drug development is full of programmes where a preferred modality was pursued past the point where the biology was signalling something different. The discipline to start with deep target understanding – to let the structure, the mechanism, and the patient context speak first – and then select the tool accordingly is harder than it sounds when platform preferences, commercial timelines, and investor narratives are all pointing in a particular direction.
Held together with the data strategy framing from the earlier session, a coherent picture emerges: the science that will produce the next generation of medicines starts with biological understanding built on high-quality, expertly curated data, uses AI as a tool in service of that understanding rather than a replacement for it, and selects modality based on what the target actually requires.
The reason these two slides felt significant at BioKorea, specifically, is context. The conference was full of evidence that drug development is accelerating – new modalities, faster clinical timelines, and AI-assisted discovery at every stage of the pipeline. That acceleration is genuinely exciting.
But speed without scientific rigour is just faster failure. As the industry moves more quickly and reaches for more ambitious therapeutic approaches, the foundational principles embedded in those two slides matter more, not less. The data has to be right and the biology has to lead.
Everything built on top of that can move as fast as the science allows, but the foundation doesn’t get to be an afterthought.
BioKorea was an important reminder that successful AI in drug discovery still begins with trusted scientific data, biological understanding and rigorous experimental evidence.
For more insights into AI in drug discovery, computational chemistry and research-grade scientific software, follow follow Dr Neil Taylor on LinkedIn. Or, if you’d like to arrange a demonstration of DesertSci’s Proasis, please get in touch with our team.
