Biologics R&D is a lot like a camera lens. The aperture controls how much light reaches the sensor—like your eye's pupil, which opens and closes to control the light. A wider aperture gives you a shallow depth of field, used for portrait photos where the background looks blurry. A narrower aperture creates a deep depth of field, great for landscapes where everything stays sharp.
In the Discovery phase of R&D, you want a wide aperture to capture as much of the scene as possible, but then you’ll filter it carefully. In the Lead Optimization phase, you’ve essentially narrowed down the focus to one subject, making it sharp. When you move to the Chemical Processing and Controls (CMC) and Scale-up phases, you want to lock in the focus and record the settings you used, so anyone can recreate the photo.
The shift from looking at lots of possibilities to focusing on specific ones in a drug discovery pipeline is why traditional pipelines don’t work well. It’s also why for immunotherapy discovery, you need a “relationship-first” architecture that can track how things change.
The Right Tool for the Wrong Stage
Over the last few years in my work supporting immunotherapy discovery, I have spent time listening to scientists describe how they work and what tools they lean on for at each stage. What I’ve noticed is that scientists depend on different categories of software tools depending on where they are in the R&D workflow.
That’s because each stage demands different modes of thinking:
Discovery calls for divergent tools like broad search, hit triage, the ability to pull in large numbers of sequences and filter aggressively - "What's interesting about this group of sequences?"
Lead optimization calls for convergent tools for side-by-side comparison, delta tables, developability scoring, and decision gates – “Which candidate do we advance?"
Qualification/CMC calls for traceability/verifiability tools for batch records, analytical method references, or regulatory annotation chains - "Can we prove what we’re claiming?"
Scale-up calls for process handoff tools for CRO request triggers, expression yields, or purification parameters - "Does this translate?"
Together, the different modes reveal a problem that has plagued our industry. Most platforms are built to excel at one of these modes and then get stretched to cover the rest. For example, a software tool designed for triage doesn't naturally produce an audit trail, whereas a tool designed for regulatory traceability wasn't built for open-ended search. So, the judgment a scientist forms at one stage either gets manually re-entered at the next, or it gets lost entirely, possibly reconstructed later from memory, from emails, from someone's recollection of a decision made at a quarterly lab meeting.
The Handoff Is Where the Reasoning Gets Lost
And that is a real cost of mismatched tooling. This is not just inefficiency; it is a specific and recurring failure at every handoff, where the reasoning behind a decision doesn’t travel with the decision itself.
The question organizations need to answer is not "what does the platform do?" The question is "where does scientific judgment live, and what happens to that judgement at each handoff?"
For complex biologics like multispecific antibodies (MsAbs) or antibody drug conjugates (ADCs), the modality itself is a coordination puzzle.
R&D for a monoclonal antibody (mAb) has a well-defined path: sequence -> structure -> function. A MsAb has two or more defined paths, plus a format question, plus a payload decision if it is an ADC or radiotherapeutic, plus a linker chemistry decision that interacts with everything upstream. Each branch is a potential fragmentation point and a place where the insight generated in one arm of the discovery program never reaches the scientist deciding in the other arm.
For decades, the answer has typically been to build a better database. Luma's answer is different.
Dotmatics Luma captures more than just the data; it also captures the context and the scientific conclusion. For example, the Surface Plasmon Resonance (SPR) overlay that showed binding interference in arm A should be a referenceable object that reaches the scientist in arm B, before a redundant experiment is commissioned.
We've Been Trying to Solve the Wrong Problem
Luma inverts the query-to-question paradigm. The point here is not about scientists doing more with less or by working harder, but rather it is about scientists working on fewer questions because the platform already answered the adjacent ones. With Luma, instead of asking "show me all KD values for this construct," you get "here's what you concluded last time you were in this part of design space." But that only works if the data model treats modality complexity and context as first-class citizens.
Most LIMS and ELN platforms were designed around the mAb format. MsAbs and ADC modalities break the mAb object model, because the format is the experiment. You can't separate the biologic format from the decision logic that produced it. Luma closes this architectural gap, by storing the process and the design rationale alongside the molecule.
When a scientist asks whether a platform can reduce redundant experiments, many vendors answer by promising better search functionality. But that's simply a retrieval improvement. Luma changes what gets recorded in the first place, and that's a different reality entirely and where the real workflow timeline compression comes from.
Scientific Judgment Shouldn't Evaporate at the Handoff
We also see that most platforms have solved the storage problem, which is about where the data live. The harder problem for complex modalities is that the scientific judgment behind the data was never captured as a first-class object, because it lived in someone else’s head. When that scientist moves on to another program, the reasoning goes with them. What's left is the result without the rationale and that means the next scientist facing the same design space starts from scratch.
This is what is meant by treating context as a first-class object. It is not merely theoretical either; in fact, it is how BioGlyph Luma works today and a great example of what this looks like in practice. Imagine if a MsAb glyph, an SPR overlay, the alignment, the developability plot all live in the system as something the next scientist can find and reason from, versus as a PDF attachment to a report nobody reads.
The organizations that win the next decade of biologics development won't be the ones with the most data. They will be the ones where scientific judgment doesn't evaporate at the handoff, because they built an architecture AI can reason across from the very first step.
See how this plays out in a real antibody discovery challenge—click here to register for our upcoming webinar, AI Alone Will Not Accelerate Science, on August 12.