The New Experimental Stack: Building the Lab That AI-Led Discovery Actually Needs

November 18, 2026

Part 2 of 3: AI & the Future of Protein Drug Discovery in APAC

In Part 1, I described the AI Acceleration Paradox: AI has compressed the front end of protein drug discovery while shifting the bottleneck downstream to experimental validation, where many labs are not yet designed to keep pace.

AI-ready laboratory integrating protein interaction analysis, automated workflows, and data architecture for drug discovery.

This article examines what an AI-ready experimental stack looks like, which technologies are changing validation economics, and how they can operate as a coherent learning system rather than a collection of instruments.

My perspective draws on more than two decades across research and commercial roles in APAC, from protein-target research at Würzburg, NUS, and A*STAR to Affymetrix, Thermo Fisher, and DKSH Technology. That experience has shown me how labs are structured, where constraints emerge, and where change can create the most value.

From Data Ggeneration to dDecision Generation

The essential mindset shift is simple: the experimental layer should not merely generate data. It should generate decisions.

As AI increases the number of candidates entering validation, more instruments and more data do not automatically improve prioritization. Data quality, turnaround time, interpretability, and integration with computational models matter as much as throughput.

The new experimental stack brings together three capabilities: high-resolution interaction analysis, automated workflow orchestration, and AI-ready data architecture.

AI-ready laboratory integrating protein interaction analysis, automated workflows, and data architecture for drug discovery.

Figure 3: The three-pillar AI-ready discovery backbone, connecting the AI design engine to prioritized, IND-ready leads through a closed feedback loop.

Pillar 1: High-Resolution Interaction Analysis

The core questions in protein characterization have not changed: Does a candidate bind? How tightly, specifically, and durably? Under which conditions? Earlier tools made answering these questions painstaking and throughput-limited.

AI does not remove the need for experimental answers. It increases the number of candidates, accelerates the required response, and raises the resolution needed for results to inform the next design cycle.

Next-generation biophysical platforms can therefore create competitive advantage.

Grating-Coupled Interferometry (GCI), including the Creoptix WAVEsystem, measures fast kinetics across the sensor surface. Its resolution of rapid dissociation is valuable in fragment-based discovery and for candidates whose binding may not be sufficiently durable.

NanoTemper Spectral Shift measures binding in free solution through changes in emission spectrum. Avoiding immobilization can benefit difficult targets, including membrane proteins, intrinsically disordered proteins, and molecules whose biology is altered by surface attachment.

Refeyn Mass Photometry measures individual molecular mass in solution using minimal sample, supporting assessment of stoichiometry, aggregation, and large protein complexes.

Together, these platforms emphasize label-free or solution-based characterization and produce outputs that can move more directly into computational workflows. Data that requires extensive manual interpretation before reuse slows the learning cycle.

AI-ready laboratory integrating protein interaction analysis, automated workflows, and data architecture for drug discovery.

Figure 4: Matching technology to bottleneck. There is no universal answer. The right platform depends on your target class and the specific decision point you need to resolve.

Pillar 2: Automated Workflow Orchestration

Labs making the strongest progress with AI-led design are not necessarily those with the most expensive instrument. They are those organized around a closed loop: Design, Build, Test, Learn.

The main delays often sit between steps: manual sample preparation, data exports requiring reformatting, and decisions waiting in inboxes. Individually small handoffs can accumulate into months.

Laboratory automation and laboratory operating systems (Lab OS) reduce these handoffs. The focus is shifting from automating isolated tasks, such as pipetting or centrifugation, to orchestrating complete workflows in which scientists define experimental intent and connected systems execute it.

By SLAS 2026, companies including HighRes Biosolutions and Opentrons were demonstrating agentic AI in the lab. Automata’s LINQ platform illustrates the Lab OS model by combining modular robotics with orchestration software that can connect computational direction to experimental execution.

The commercial logic is clear: if AI produces a stronger candidate set in days but experimental confirmation still takes weeks because of serial processing and manual handoffs, much of the value is lost. Closing the loop is therefore a profitability decision, not simply an operational convenience.

Pillar 3: AI-Ready Data Architecture

Data architecture often receives the least attention, yet it can create the greatest friction.

Data volume is not data quality. AI models need structured, contextualized, comparable, and reusable data. In biology, reagent lot, buffer, temperature, assay format, instrument, and operator can all affect results. A binding constant is meaningful only with the conditions that produced it.

If that variation is not captured, data cannot be used effectively to train or validate models. The result is a warehouse full of information that cannot be reliably found or compared.

Policy is moving in the same direction. The US AI-Ready Bio-Data Standards Act of 2026 directs NIST to develop frameworks for machine-readable, AI-usable biological data. Whatever its wider influence, it signals that data will increasingly be judged by both scientific quality and computational reusability.

Experimental design and data architecture must therefore be planned together. Assays should capture structured metadata, protocols should be version-controlled, and results should remain linked to their experimental conditions. This is how a lab turns data into organizational intelligence.

AI-ready laboratory integrating protein interaction analysis, automated workflows, and data architecture for drug discovery.

Figure 5: Most APAC labs operate at Level 1–2. AI-led discovery requires Level 3–4. The gap is bridged by data architecture, not by additional instruments.

Why This Matters for Profitability

AI discovery narratives often focus on computational breakthroughs. These advances matter, but they can make the investment question appear mainly about which AI platform to license.

A better starting question is: What is the experimental bottleneck, and what does that bottleneck cost per decision?

Profitability improves when weak candidates are eliminated earlier. Failure during binding or functional validation costs far less than failure after years of clinical development.

High-resolution interaction analysis, workflow orchestration, and AI-ready data architecture enable earlier, more confident decisions. They are not overhead; they are decision infrastructure.

The Diagnostic Question Every Team Should Ask

Before investing in AI platforms or next-generation instruments, every team should ask: Where does our system stop learning fast enough?

The answer varies. One team may be constrained by cell-based functional validation; another may have excellent kinetics data trapped in spreadsheets; another may lose weeks to manual steps between capable instruments.

That diagnosis should determine the investment sequence. The highest-return intervention is often not the most exciting technology, but the change that removes the slowest or least reliable decision point.

Part 3 will translate this framework into APAC pathways for small and medium-sized developers, government-backed programs, and statutory boards. It will explore how organizations at different stages can build or access an AI-ready experimental backbone without owning every component.

Author’s Bio

Raquel Magalhães is a PhD scientist with over 22 years of experience in life sciences and biotech across APAC. She began her research career studying protein targets at the University of Würzburg, completed a PhD at NUS, held research and commercial roles at A*STAR, Affymetrix, and Thermo Fisher, and now leads Business Development at DKSH Scientific Solutions in Singapore. *Views expressed are her own.

[← Back to Part 1] [Continue to Part 3 → The APAC Path to Success]

Sources:

About the Author

Raquel Magalhaes joined DKSH in 2023 as a Group Business Development Manager based at the Singapore office. She brings a wealth of experience in Biotechnology sales, cultivated through her impactful role at Thermo Fisher Scientific, and her research background with prestigious institutions. In her current role, she is responsible for driving sustainable growth in the Life Sciences sector.

Raquel Magalhaes

Life Science