Antibody engineering has come a long way. Thirty years ago, scientists were just beginning to understand the potential of antibody therapeutics. Today, thanks to advances in molecular biology, biochemistry, and industrial-scale production, the field has moved past asking "what is possible?" and is now asking a much harder question: "what is the best way?"
That shift in mindset was the starting point for a recent presentation that walked through how Dotmatics Luma is helping research teams turn antibodies into next-generation delivery vehicles and how a connected, AI-native data platform is changing the pace of scientific discovery.
A Real-World Challenge: Crossing the Blood-Brain Barrier
To ground the discussion, I'll pose a scenario many neuroscience researchers will recognize. The formation of amyloid plaque is closely linked to tau protein tangles, which drive disease progression in Alzheimer's. If modifying molecules like CYP40 (a molecular chaperone) or HTRA (a serine protease) could slow or reduce tau tangle formation, the next question becomes obvious: how do you actually deliver them?
The blood-brain barrier stands in the way. One strategy is to use an anti-transferrin antibody or antibody fragment to mediate delivery — but that raises a deeper design question: is an antibody even the right format?
It's the kind of question that used to take months of iteration to answer. With the right tools, teams can start answering it today.
The Real Blocker Isn't Ambition, It's Data
Whether you're in academia or industry, working across multiple sites, continents, or departments, the same problem keeps showing up: pulling data together is hard. And as AI becomes part of everyday scientific work, fragmented data doesn't just slow down human researchers — it limits what AI and ML tools can actually do with the information available to them.
What's needed is a way to traverse the "nodes" of research data across multiple dimensions — powering both human scientists and the AI tools working alongside them. That's the problem Luma was built to solve.
Bringing Design, Process, and Results Together
As an AI-native platform, Luma connects design, process, and results into a single ecosystem. This enables faster iteration cycles powered by experimental data across every stage of research. Some teams refer to this concept as "lab in the loop," and it's central to how Luma is built.
In practice, that means:
Better collaboration and planning across teams and sites
More optimized execution of experiments
Faster exploration of insights using AI
A foundation for predictive modeling
It starts with an accurate scientific data model representing the molecules themselves — and the foundation of that model is building blocks: composable units that break immunoglobulins down into structured, reusable components. This gives scientists an intuitive way to look at a molecule (instantly recognizing, for example, a bispecific with an scFv off the Fc and an Fv off the Fc) while also structuring the underlying data for scalable experimentation.
Designing for the Practical with Bioglyph
This is where Bioglyph Luma comes in. Returning to the blood-brain barrier scenario: a single antibody carrying both enzymes might be highly potent, but its size could be a liability for crossing the barrier. Removing one enzyme helps. Concerns about clustering the transferrin receptor might call for removing a Fab. Moving to an scFv reduces molecular size even further.
Bioglyph lets teams walk through that iterative design process visually — in real time, whether collaborating in person or over a call — while also assessing sequence and structure to improve developability and reduce manufacturing liabilities.
It's more than a molecule-drawing tool. It's a collaboration platform. In the scenario let's walked through, a seemingly simple design exercise — just a couple dozen Fabs, a few scFvs, a handful of combinations — quickly produced 144 potential molecules to evaluate. Seeing that number before committing any lab resources makes it possible to have honest conversations about capacity before the work begins, rather than after.
From Design to Execution: Seeing the Full Picture
Once a design panel is set, Luma tracks deep scientific relationships and lineage across the entire process, connecting design data with experimental and process data to build a true operational picture.
From a single dashboard, teams can:
Get high-level project summaries (project name, number of constructs, key metadata)
Drill into expression data across pilot runs to spot variability
Compare properties like aggregation and purity to catch outliers early
Explore quality scores combining titer, purity, and aggregation on a molecule-by-molecule basis
Review assay and binding data across species
And when new questions come up that weren't anticipated in the original dashboard design Luma Agent lets researchers ask questions directly of the data, surfacing connections across every batch, cell line, and condition a protein has been tested under. Because Luma maintains deep relationships between data points from the start, nothing has to be "found" — it's already connected.
What's Next: Moving Design Earlier in the Process
Looking ahead, the vision for Luma is to push design even earlier into the analysis phase. One concept in development would connect tools like Protein Metrics, used to evaluate mass spec data, directly with Bioglyph, sharing not just the molecule's glyph but predicted primary and off-target peaks with scientists at the bench. That means faster, richer analysis, since scientists could compare live instrument data against what the design predicted.
From there, results flow back into Bioglyph, connecting sequence, structure, and function — helping teams evaluate whether a flagged residue is a real liability (say, a buried hydrophobic residue) or a non-issue, and feeding those insights straight back into the design process. That's the lab-in-the-loop model in action: insight from the bench continuously reshaping design, and design continuously informing what happens at the bench.
Solving for Legacy Data with Format Rescue
None of this matters much if decades of historical data stay locked away in old formats, disconnected from new work. That's the problem Format Rescue, one of Bioglyph's newer capabilities, is built to solve. It imports legacy sequences, validates and aligns them, and automatically detects format.
In one real customer example, over 200,000 legacy sequences were processed — roughly 90% were fully annotated automatically, and the remaining sequences were at least partially annotated. That gives research teams a clear, scoped starting point for curation decisions, and — more importantly — a way to finally bring legacy data into the same connected data model as everything moving forward, rather than leaving it to sit side by side in a separate, disconnected schema.
The Bigger Picture: AI-First, Built on Databricks
Luma is built as an AI-first platform on Databricks, with a scientifically aware data model that tracks entities, processes, and lineage throughout discovery — so data can be reused and connected even in ways that weren't anticipated at the outset.
The result is a platform that supports:
Collaboration: Real conversations about data as it's generated, not just after the fact.
Resource planning: Both upfront and as new data comes in.
Automated project updates: Trends and projections that evolve with the data.
Predictive modeling: And eventually, simulation.
Your data and your process are the foundation of your AI journey. Structuring that foundation well today is what makes rapid, confident decision-making possible tomorrow.
Want to see Luma in action? Watch the full on-demand webinar for a deeper look at the discovery workflows, dashboards, and design tools discussed above. Watch the on-demand webinar