Key Takeaways from AACR, PEGS, SLAS Europe and Bio-IT World
In spring 2026, four major life sciences conferences converged on the same practical question: how does AI become genuinely useful in day-to-day drug discovery R&D?
AACR, PEGS Boston, SLAS Europe, and Bio-IT World each examined the field from a different angle, but three patterns kept resurfacing.
Three patterns across four conferences
- AI value depends on implementation, not adoption. Across all four events, the most relevant sessions focused on where and how AI delivers decisions R&D leaders can act on – not on model architecture, but on data quality, validation rigor, and integration with experimental workflows.
- Data infrastructure now rivals models for attention. Knowledge graphs, ontologies, and harmonized datasets appeared consistently as practical prerequisites for scaling AI in life sciences. AI-ready data – biomedical data that is harmonized, quality-controlled, connected to metadata and ontologies, and usable in reproducible workflows – was treated as an implementation requirement, not a buzzword.
- Biological relevance is the validation standard. Better model systems, multimodal imaging, translational evidence, and rigorous validation shaped the conversation about what separates useful AI from technically impressive demonstrations.
AACR 2026 – Oncology decisions became data problems
Target discovery was not the conference theme. Novel targets made up a small fraction of abstracts. The proteome is largely mapped. Disease associations are broadly known. The selection problem has moved one layer downstream: which known targets enable a drug program given the constructs available today?
That shift was visible across sessions. The questions on display were concrete and bounded – not biological questions in disguise, but data and modeling questions sitting inside biology conversations:
Which indications match a specific ADC payload? Which antigen pairs co-express reliably enough for AND-gate logic in bispecifics? Which tumor types internalize a given ADC construct? Which targets translate across species? Where does a therapeutic window actually exist in the patient population?
These are not the kind of questions that get answered with a single dataset or a general-purpose AI tool. They require integrated evidence across expression, disease context, model systems, modality constraints, safety signals, and translational relevance.
Jan Majta, PhD noticed: AACR felt like “a long list of computational problems wearing biology hats.”
These questions are already on desks at platform companies, with timelines attached. The field is not waiting for someone to frame the problem. It is looking for teams that can answer it.
PEGS Boston 2026 – Biologics AI confronted developability and safety
PEGS Boston 2026 drew approximately 2,300 participants across 21 tracks, with three tracks dedicated to AI/ML for biologics design and optimization. These sessions drew particularly high interest – but the most valuable discussions went beyond molecule generation to focus on developability, manufacturability, and safety.
The emphasis on “Safety First” was hard to miss, especially around immunogenicity risk assessment and pipelines to reduce anti-drug antibody formation. Andrew Martin’s presentation, “Application of AI to Developability Screening, a Skeptic’s View,” framed the mature question well: not whether AI can suggest candidates, but whether AI-supported workflows help identify which candidates are developable, safe, and worth advancing.
For biologics teams: AI creates value when it connects sequence, structure, expression, developability, immunogenicity, and disease biology data into prioritization workflows – not just generation pipelines.
SLAS Europe 2026 – Screening, imaging, and biologically relevant models
At SLAS Europe in Vienna, AI was not confined to a single track. Magdalena Otrocka, PhD, observed that it cut across imaging, safety assessment, phenotypic screening, and automation – and opened the State of the Industry Panel as the first topic of discussion.
The dedicated AI and Data Science track, running in parallel across all scientific sessions, reflected how central the topic has become. One session, “From Prediction to Precision,” focused specifically on AI-driven image analysis, toxicity and safety prediction, and the use of multiple data modalities to complement information extracted from biomolecular images.
The core tension: in high-throughput screening, generating data is no longer the bottleneck. Extracting interpretable biological signal from large, multimodal datasets is.
Particularly important was the pairing of AI with more biologically relevant model systems. Florian Fuchs from HeartBeat Bio presented cardiac organoids that demonstrated how combining different data types – imaging, functional readouts, molecular profiling – can produce a more complete picture of drug response than any single modality alone. It was a concrete example of where model relevance and multimodal AI analysis intersect.
The broader pattern: 3D culture, patient-derived iPSC systems, and organoids are becoming available at screening scale. AI-driven analysis helps – but only when the underlying biological model is relevant enough to support downstream development decisions.
Bio-IT World 2026 – Useful AI starts with data models and ontologies
Bio-IT World’s 25th edition centered on operationalizing AI, with strong attention to data infrastructure and knowledge representation. Two tracks stood out: “AI for Drug Discovery & Development” and “AI for Oncology, Precision Medicine & Health” – both heavily focused not just on making AI models work, but on demonstrating measurable impact and the role of data preparation in making that impact possible.
For Ardigen – attending for the 10th consecutive year – the event confirmed a shift in what audiences care about:
“The most telling moment at Bio-IT was a question from the audience: ‘What is the model behind this?’ – except they weren’t asking about the AI model. They wanted to know about the data model and ontology inside the knowledge graph. That’s where the field’s attention has moved.” – Dawid Rymarczyk, PhD, Ardigen
Life sciences audiences are looking beneath the AI interface. They want to understand how biological entities, evidence, and relationships are represented – because a model is only as useful as the data structures around it.
The implementation challenge is more specific than it sounds. Even clinical trial data from two different sources in CDISC format – supposedly a shared standard – can differ enough to require substantial harmonization before it supports reliable downstream analysis. If standardized data is not truly interoperable, proprietary and experimental datasets require even more careful preparation.
Dawid also pointed to a broader acceleration pattern: tasks that took years now take weeks; tasks that took weeks now take hours. Literature mining, annotation, and standard analysis coding are increasingly handled by specialized orchestrated agents, freeing scientists to focus on hypothesis development and therapy design. But this acceleration only holds when the underlying data infrastructure – harmonization, metadata, ontologies, quality control, provenance, and governance – is in place.
How Ardigen can help
Ardigen helps biotech and pharmaceutical teams turn complex biomedical data into decision-ready insights – through AI-ready data foundations, knowledge-driven workflows, and scientific applications for drug discovery and precision medicine.
If your team is working on AI-ready data infrastructure, computational biology workflows, knowledge graph design, multimodal data integration, or translational decision support, we can help build the foundations that make AI operationally useful across your discovery programs.
Let’s talk about building AI-ready workflows for your discovery pipeline
Frequently Asked Questions
What happened at the major life sciences conferences in 2026?
AACR, PEGS Boston, SLAS Europe, and Bio-IT World each approached AI in drug discovery from a different angle – oncology, biologics, screening, and data infrastructure – but converged on the same message: AI value depends on implementation, not adoption. Data readiness, biological relevance, and workflow integration emerged as the real bottlenecks across all four events.
Was AI a big topic at PEGS Boston 2026?
Yes – three tracks were dedicated to AI/ML for biologics design, with the strongest interest around developability, safety, and immunogenicity risk rather than molecule generation alone. Across sessions, the consensus was that AI delivers the most value when it connects sequence, structure, expression, and disease biology into end-to-end prioritization workflows.
What were the highlights of Bio-IT World 2026?
Bio-IT World’s 25th edition centered on operationalizing AI, with particular attention to knowledge graphs, ontologies, and data models. Across sessions, the focus was less on AI algorithms and more on how biological entities, evidence, and relationships are represented and structured – a sign that data representation has become as important as model capability in life sciences AI.
