Distributed Operations in Drug Discovery: Balancing Internal Expertise and External Execution

In this CDD Vault webinar, Dean Dragoli (Vice President, Medicinal Chemistry, Alumis) moderated a discussion with Kenji Kozuka (Senior Director of Drug Discovery, Ardelyx) and Jeff Warrington (Principal Scientist, Digital Chemistry and Design, Novo Nordisk).

The session focused on a practical question many R&D teams face: how to allocate work between internal scientists and external CRO partners while preserving quality, speed, and institutional knowledge. The conversation covered decision criteria for outsourcing, models for CRO engagement, trust and IP, global execution, and the growing role of digital infrastructure and AI in coordinating distributed operations.

The Core Question: What Work Belongs Where

Dean opened with a simple but critical question that every organization must answer repeatedly: what do you work on in house and what do you send outside. The answer depends on:

• Company size and stage
• Available internal capabilities
• Project complexity and novelty
• Timelines and budget

From a biology and translational perspective, Kenji emphasized that study complexity is a primary driver. For highly complex or conceptually fragile work, he prefers to keep execution in house, where scientists driving the questions stay directly connected to the data. Routine or standardized assays, in contrast, are strong candidates for outsourcing, especially when cost and throughput are priorities.

New targets and first-in-class programs tend to demand more internal oversight. Teams often still use specialists at CROs, but with much tighter interaction and review cycles. As projects mature and protocols stabilize, more work can move external without losing scientific control.

Three Modes of CRO Partnership

Jeff described three common modes of working with CRO partners in drug discovery:

• Task or project based
  Short term assignments in which a CRO executes a specific assay, synthesis run, or study. This model is transactional and works well when the task is clearly defined and easy to specify.
• Portfolio outsourcing
  A more distributed model in which a network of CROs covers most of the functional steps in the discovery pipeline. Internal teams focus on strategy, design, and integration while external partners perform a wide range of experimental work.
• Integrated partners
  A single external organization provides an end to end platform that includes chemistry, biology, DMPK, and often in vivo models under one roof. This model suits smaller or younger companies that lack internal lab infrastructure and need an external incubator for execution.

Early stage startups tend to use these models deliberately to compensate for limited internal capabilities. Larger companies now also apply them at scale, even though they have substantial infrastructure, especially when they branch into new modalities or require specialized platforms that would be inefficient to build internally.

Trust, IP Concerns, and Long Term Relationships

Trust and information sharing came up repeatedly. All three speakers noted that scientists often worry about revealing too much structural or assay detail to external partners, particularly when work is proprietary or when a novel assay represents a key source of advantage.

Kenji’s approach is to:

• Share enough information for the CRO to execute high quality work
• Withhold unnecessary proprietary detail when the risk benefit balance does not justify full disclosure
• Ensure robust confidentiality agreements and data handling clauses are in place

Trust builds over years, not weeks. With CROs that have delivered high quality data consistently, Kenji no longer worries about disclosure to the same extent. For new CROs or new sites within a known CRO, he treats the initial engagement almost like a hiring process, reviewing the study director’s background, experience, and communication discipline before committing critical work.

Dean highlighted that trust is mutual. When sponsors show respect for external scientists, recognize their problem solving, and treat them as peers rather than anonymous vendors, productivity often increases. After site visits, he has seen very concrete effects: more compounds delivered, more proactive issue resolution, and greater willingness to start complex in vivo studies before all contracting details are finalized.

Data Flow, Centralization, and Quality Control

Distributed operations only work if data flows back in a structured and secure way. The group discussed two patterns:

• CRO uploads directly into the sponsor’s central data system
• Internal staff receives files and then curates and uploads them

Direct upload by CRO teams increases speed and reduces manual data entry for in house scientists. It requires higher trust and strong permissioning, since external users interact with the same data environment that internal teams use.

Jeff framed centralization as essential when multiple partners and internal teams work on shared assets. Data warehousing needs to support:

• Central access to structures, results, and metadata from many sources
• Compartments or “firewalls” between programs when required
• Clear separation of permissions between internal and external users

Within this context, CDD Vault serves as a hub where structures, assay results, and study metadata from multiple CROs and internal labs converge rather than sitting in scattered spreadsheets and email attachments. Centralization makes cross site comparisons, such as running the same study at two different CROs to check reproducibility, practical instead of ad hoc.

Global Networks, Tariffs, and Time Zones

The panel also addressed practical concerns that have become more visible over the past few years.

Tariffs have added cost and occasional delays to shipments of compounds and materials. In practice, the group views tariffs as an irritant rather than a strategic driver. The incremental cost is usually small compared with the total study cost, and they do not let tariff considerations dictate CRO selection. The more important criteria remain data quality, timelines, and existing relationships.

Working across time zones creates operational friction, but the speakers have adopted straightforward mitigation patterns such as alternating meeting times for fairness and relying on asynchronous communication for non urgent questions. When relationships are strong, small inconveniences in scheduling do not dominate partner selection.

When to Keep Work in House

During the Q&A, an audience member asked whether there are assays that should never be outsourced. The panel converged on one main category: highly proprietary assays that encode a core part of the company’s differentiation.

Complex, internally developed assays that translate a unique disease insight into an experimentally tractable readout are difficult to recreate and often constitute the most defensible part of a discovery engine. In those cases, sending the entire assay platform to a CRO, especially with an expectation that it might become part of their service menu, can erode long term advantage.

In contrast, chemical entities ultimately become public as structures move into patents and literature. Animal models eventually appear in publications as well. Novel assays that capture a key biology signal remain harder to replicate and therefore warrant additional protection.

Stage of development also matters. GLP toxicology and IND enabling packages almost always require external partners since many organizations lack GLP facilities. Even large pharma sites that have GLP capabilities still outsource portions of this work because it would consume too many internal resources. Earlier exploratory pharmacology or efficacy studies are more flexible and can be run in house longer, especially when interpretation is sensitive or nuanced.

Economics, Opportunity Cost, and Management Discussions

The decision to outsource is often framed as a budget or rate question. The speakers argue that the more important concept is opportunity cost.

Running every critical study or synthesis in house leads to concentration of risk and narrow project portfolios. Outsourcing allows teams to run multiple efforts in parallel and hedge scientific risk across programs, assays, and modalities.

When engaging executive leadership, Kenji finds it more productive to emphasize:

• Which projects internal teams will focus on if certain work goes external
• How much internal time will be freed by outsourcing specific studies
• How timeline compression on key milestones offsets higher per study cost

Dean tied this back to the familiar pressure to deliver new development candidates on regular timelines. Leadership often thinks in terms of twelve to eighteen month windows for producing the next molecule that moves forward. External partners with the right capabilities can make those timelines attainable when internal resources alone would not be sufficient.

AI, Digital Chemistry, and the Future of Outsourcing

In the final segment, the panel turned to AI and its impact on distributed operations. Jeff outlined how digital chemistry and AI driven design now influence nearly every stage of the discovery pipeline, from target selection through hit discovery, hit expansion, lead optimization, and candidate selection.

Internally, many teams are converging on a model where:

• External partners focus on make and test
• Internal teams and digital platforms focus on analyze and design

Generative and predictive models inform structure design, optimize for multiple endpoints, and integrate physics based and machine learning models into iterative design cycles. This trend raises questions about job security for traditional medicinal chemistry roles, but Jeff framed the change in a specific way.

AI systems will increasingly automate the tedious parts of navigation, similar to self driving cars that handle steering and route optimization. They will not decide where to go. Humans still frame the questions, define target product profiles, and judge trade offs between risk, novelty, and feasibility. The last stronghold for non AI work will likely be the problem formulation step, where scientific judgment, clinical strategy, and real world context intersect.