AI at Roche and Genentech: End-to-End R&D, From Target Discovery to Scaled Manufacturing
Jan 9, 2026
At Roche and Genentech, AI and machine learning are embedded across research, development, and manufacturing. This is an ecosystem, not a stack of tools. Each step feeds the next so we learn faster, design better, and move with more confidence.
By integrating AI end-to-end, teams are identifying new targets, iterating molecular designs, running smarter trials, and preparing for manufacturing earlier. The result: molecules that were previously out of reach, decisions backed by stronger evidence, and a pipeline that moves with fewer bottlenecks.
Discovery: Targets and Molecular Design at Speed
Target discovery is slow because biology is complex. Instead of spending months combing through massive datasets, researchers now use autonomous agents built on multiple AI biology models to pinpoint promising targets and select indications with more precision.
AI is now core to discovery: all antibody programs and about 90% of eligible small molecule programs integrate AI to streamline design and improve predictability. Recent results include an oncology molecule redesigned to reduce unwanted immunogenicity and move faster into development; a degrader created 25% faster with a structure that wouldn't have been reached without AI; and an immunology backup small molecule delivered in seven months versus more than two years for the original pre-AI lead. For teams looking to build these capabilities, see AI Research Courses.
The Lab in a Loop
Data from the lab and clinic feed AI models that propose experiments and generate molecular designs. Scientists run the experiments, feed results back, and the models improve. It's a tight loop: data, computation, experimentation, discovery.
"The beauty of the 'lab in a loop' is that it augments scientists, rather than replacing them. AI allows us to iterate quickly and at scale, freeing our teams to ask bigger questions - and answer them better and faster than ever." - Aviv Regev, Head of Genentech Research and Early Development (gRED)
To scale this approach, the Computational Sciences Center of Excellence (CSCoE) launched in 2025. This joint Roche-Genentech group accelerates AI projects across research and early development, standardizes data and tools, and spreads best practices.
Clinical Development: The Clinic in a Loop
Clinical development is full of friction: up to 80% of trials face delays and 20-30% of patients typically drop out. AI-powered simulations now help teams test trial scenarios, balance trade-offs, and optimize enrollment criteria and endpoints to reduce risk and dropout.
Teams analyze patient data in real time to select sites and strategies faster, cutting planning from weeks to minutes. AI image analysis reduces expert review time by over 90% while delivering more consistent assessments-crucial in ulcerative colitis and ophthalmology, where precise readouts directly inform decisions.
In oncology, validated ML models trained on prior trials predict outcomes such as tumor response and survival. Those predictions guide future trial designs and decisions at readout, enabling faster learning and clearer go/no-go calls.
"Artificial intelligence has become a true force-multiplier across our development engine, including in clinical trials. AI-driven tools help us do everything from generating endpoints more quickly to monitoring treatment response more precisely in trials. They give us real-time insights and cut months - and potentially years - off the traditional timeline, ensuring we can bring life-saving medicines to the people who need them faster and more reliably." - Venkat Sethuraman, Senior Vice President and Global Head of Data Science and Analytics
Manufacturing: From Manufacturability to Smart Factories
Production constraints are addressed early. AI models assess manufacturability during discovery so teams advance molecules that can be produced at scale, avoiding late-stage surprises.
As programs move to late-stage and commercial manufacturing, AI improves efficiency, quality, and scalability. Digital twins, predictive maintenance, quality monitoring, autonomous robots, and adaptive training help reduce costs and increase output; these are practical examples of AI for Operations.
"Even before a medicine is approved and commercialized, we use AI to simulate, design, and optimize manufacturing processes that are resilient, compliant, and ready to scale to deliver higher value for our patients." - Daniele Iacovelli, Head of Data and Digital for Roche's Global Manufacturing Network
Aligned with sustainability goals, computer vision and ML enhance defect detection and consistency while reducing waste. Agentic AI farms and automation will span development, manufacturing, and supply to improve reliability for patients and simplify work for employees.
"In manufacturing, our ambition with AI goes beyond efficiency. It is about building a future-ready supply network that patients can always rely on: resilient by design, intelligent by default, and able to deliver the medicines people need, exactly when they need them." - Daniele Iacovelli
Collaboration Multiplies Impact
Impact grows when biology, chemistry, clinical insight, and advanced computation come together. That's why strategic collaborations are central to scaling AI across discovery and development.
With NVIDIA, teams improved protein structure prediction, reduced model training time, and accelerated image analysis. The Equifold model-trained in part with the NVIDIA BioNeMo platform-predicts dynamic conformational ensembles of antibodies, turning GPU-hours of surface property computations into seconds and enabling early-stage exploration at scale. The collaboration also extends to manufacturing with plant digital twins that model material flow, processes, and staffing to optimize design and operations.
With Recursion, researchers are mapping human biology to find targets in neurodegenerative disease. A recent microglial map-built from ~46 million images linking 17,000 genes to microglial cell features-helps uncover pathways that traditional methods miss.
With Medra, Genentech is connecting data, AI, and physical lab work. The platform integrates with internal ML infrastructure and LIMS to continuously take predictions, run experiments with general-purpose robotics, and optimize protocols-tightening the "lab in a loop."
With Amazon Web Services, the gRED Research Agent helps automate time-consuming knowledge work by scanning large datasets and cross-referencing journals. It could save more than 43,000 hours of manual biomarker validation across therapeutic areas each year.
These collaborations show how unified compute, foundation models, and real-world workflows can expand the scale, speed, and scope of scientific discovery.
What This Means for Scientists and R&D Leaders
- Build closed loops. Connect data generation, modeling, experimentation, and validation so each cycle improves the next.
- Invest in data quality early. Curated, interoperable datasets raise model performance and shorten downstream timelines.
- Validate in the real world. Pre-train, fine-tune, and stress test models against prospective data before relying on them for decisions.
- Design for manufacturability from day one. Include developability and production constraints in discovery to de-risk late stages.
- Use simulation to cut trial friction. Model enrollment, endpoints, and site strategy to reduce delays and dropout.
- Collaborate across domains. Pair domain experts with computational scientists to ask better questions and iterate faster.
For a broader view of methods and case studies, this review of AI in drug discovery is a useful reference: Nature Reviews Drug Discovery (2024).
If you're upskilling your team on practical AI skills by role, see this curated index: AI courses by job.
Looking Ahead
Genentech helped start modern biotechnology. The next wave is being built by combining human ingenuity with advanced AI across discovery, development, and manufacturing.
For patients, this means more treatments that reach the clinic faster with higher confidence. For scientists, it means fewer manual bottlenecks, tighter feedback loops, and the space to focus on the hard problems that move medicine forward.
Your membership also unlocks:
