Why Biopharma Scientists Spend Up to 80% of Their Time on Data Cleanup

Monoclonal antibodies and other biologics still anchor therapeutic pipelines across the industry, but the software behind them hasn't kept pace. On a recent episode of The Digital Pulse, host Trisha Pillay spoke with Lily Helfrich, a product manager at Benchling, about what it takes to build a data foundation that serves antibodies today and new modalities tomorrow. Helfrich says antibodies make up more than half of large-molecule pipelines, raising the stakes for getting the data right.

Why Legacy Biopharma Data Systems Slow Antibody Discovery

Legacy tools, Helfrich says, tend to treat an antibody as a flat object: a sequence, or a GenBank file. In reality, it is built in layers. She likens it to a Lego model, one large structure made up of bricks that each have their own properties. Flatten it, and scientists lose sight of how the parts connect. That is where lineage breaks down. A team might keep sequences in one registry, expression and purification results in an electronic lab notebook, and plate-based work in a third system. Piecing them back together takes time. Helfrich points to McKinsey research suggesting scientists and computational teams are “spending up to eighty per cent of their time finding and cleaning data.” Every hour spent on that is an hour not spent on experiments, and AI projects that need clean, connected inputs suffer too.

Building a Biologics Data Language Layer 

Faced with that mess, many IT teams build their own registry on relational databases and cloud data lakes. Helfrich calls it an expensive trap. A schema designed around a standard IgG can buckle when it meets bispecifics, multispecifics or fusion proteins, and each new format brings more maintenance and more extension work. Roadmaps fall behind the science, and before long people are back in Microsoft Excel.

Benchling’s answer is what Helfrich calls a language layer. Instead of hard-coding a fixed list of antibody types, the platform relies on Veritas, a nomenclature developed by industry figures at Amgen, to give the data structure while leaving customers free to define their own formats. A scientist types in a string and the system draws the antibody, works out how the chains and domains relate, and picks out the CDRs and frameworks within the variable regions. No custom code is needed. The same approach, Helfrich says, can stretch to other multi-component molecules, including antibody-drug conjugates.

View interview Is AI Quietly Creating a GDPR Crisis in Clinical Research?

Is AI Quietly Creating a GDPR Crisis in Clinical Research?

Generative AI is transforming clinical trials, but a failure to distinguish between pseudonymised and anonymised data is creating costly regulatory gaps.

Image

Lessons From an 80 000 Antibody Project

Migration is one of the first questions prospective customers ask, because decades of data may be unstructured, uncharacterised or stored as flat sequences with no record of which chains belong together. Helfrich says AI is well suited to the problem. Native agents can analyse thousands of unclassified legacy antibodies, infer each format, generate a Veritas string, validate the data and register entities automatically.

She described a customer that replaced six disconnected platforms, covering registration, notebooks, inventory, assay tracking and request management, across four research sites. The project took 15 months. The company onboarded about 250 scientists, standardized more than 250 antibody formats and migrated 80,000 legacy antibodies. Within a month of going live, it had registered nearly 4,000 new ones. With the data structured, Helfrich says, IT could shift from cleaning records to bigger goals such as "lab in a loop" workflows and predictive modelling.

Unifying Biology and Chemistry Data 

Discovery is moving past the divide between small and large molecules. Antibody-drug conjugates, peptides and other hybrids combine both, yet most software still splits biology and chemistry into separate systems. For an ADC, antibody data, linker and payload chemistry, and conjugation details such as drug-to-antibody ratio can end up in different databases or in free-text fields. Peptides face a similar problem as registered in small molecule systems; they often lose the sequence-based view that helps scientists understand them.

Benchling's response is a single data model built on HELM notation as a common standard, with domain-aware languages such as Veritas and SMILES layered on top so antibody scientists and chemists each see what they need. The company has already added support for modified RNA and oligos, is building ADC workflows that detect possible conjugation sites, and is extending the same foundation to peptides.

Helfrich says the stakes are highest for AI. Models are only as good as their training data, and today's antibody models predict binding and developability well but have far less data on how a full antibody format affects drug success. Benchling's Model Hub brings tools such as AlphaFold, Boltz and Chai directly to scientists, and the company expects teams of agents, which it calls science foundries, to handle routine tasks while scientists review and QC the results. Her advice for biopharma leaders is to structure data at the source, make registration a natural part of everyday workflows, and invest in an adaptable enterprise language layer instead of temporary point solutions or rigid databases. If you would like to find out more about this, visit benchling.com or follow Lily Helfrich on LinkedIn. Listen to the full episode of The Digital Pulse for more on modernising biopharma R&D.

Takeaways

• Why fragmented systems leave scientists spending up to 80 per cent of their time managing data.
• Why complex antibodies need structured, adaptable data models.
• How AI can accelerate antibody data migration and validation.
• How one customer consolidated six platforms and migrated 80,000 antibodies.
• Why connecting biology and chemistry data matters for emerging modalities.
• How structured data supports more effective AI models.
• Bringing AI tools into everyday scientific workflows.
• Building adaptable data foundations for future drug discovery.

View interview How AI Is Reshaping Data Integrity in Pharma Manufacturing

How AI Is Reshaping Data Integrity in Pharma Manufacturing

Veragence CEO Nour AlSaleh discusses how AI is transforming data integrity in pharma manufacturing, regulatory compliance, the "human-in-the-loop" requirement, and patient safety in the era of AI-driven GxP systems.

Image

Chapters

00:00 Why biopharma’s data foundation needs to keep pace

01:39 Why antibody records need more than flat sequences

04:54 The hidden cost of building custom biologics registries

08:27 Veritas as a flexible language layer for antibodies

11:51 AI-assisted migration of legacy antibody libraries

14:11 How one team consolidated six platforms and 80,000 records

16:16 Why hybrid modalities break siloed biology and chemistry systems

24:02 A shared data model for biological and chemical languages

29:02 Why AI models need complete molecular context

35:10 Model Hub, science foundries, and human review

38:13 Building adaptable software architecture for future pipelines