Unlocking the Next Frontier in AI-Driven Drug Discovery

From target identification and molecular design to patient stratification and clinical trial optimization, artificial intelligence (AI) is showing remarkable promise in accelerating drug discovery and development. In fact, a published analysis of AI-derived molecules found they have an 80-90% success rate in Phase 1.¹

Yet, even the most advanced companies are limited by one primary bottleneck: fragmented data. Both abundant and underutilized, most laboratory, clinical, and real-world data remain siloed and locked away by privacy, regulatory, competitive or technology constraints. This critical junction between the promise of AI and the reality of fragmented, highly-sensitive data is where Federated Computing provides a new path forward - with a common platform at the intersection of edge computing, federated learning, and privacy-enhancing technology.

From Siloes to Scale

Data aggregation requires sensitive data to be transferred, which in addition to posing privacy and regulatory concerns, is resource intensive - complex, time-consuming, and costly.In comparison, Federated Computing (FC) reduces workflow friction by allowing the data to remain in place, securely behind firewalls, while the AI models, algorithms or code are sent to the data. The compute happens where the data resides, at the edge; only aggregated model parameters (‘learning’) or privacy-preserving results are shared back to a central ‘control plane’. This approach allows for:

• Privacy-preserving computation: FC can be complemented with industry best-practice security standards (e.g. encryption at rest, in transit, and in-processing aka confidential computing; as well as remote hardware attestation) along with privacy enhancing technologies (PETs), such as differential privacy, k-anonymization, and homomorphic encryption. Data never leaves its secure environment; only encrypted parameters or model updates are exchanged.
• Cross-organizational, geography, and device AI training: Organizations can jointly train AI models across datasets that could not be legally or operationally combined - while being in compliance with privacy regulations such as HIPAA and GDPR.
• Built-in data traceability and governance: Each data partner maintains control and oversight without risking competitive and intellectual property. At the same time, another advantage of using FC is full data traceability.
• Scalability: Federated networks can become “living” networks, adding new data ‘nodes’ without re-engineering the entire system.
• Data diversity, quality, and quantity: With FC, decentralized, multi-modal data can be harmonized using GenAI engines, without exposing the data to public cloud APIs. AI models, algorithms, or code can then train on greater quantities of more diverse data - reducing bias, improving generalizability, and amplifying insights.

Essentially, FC enables pharmaceutical companies to collaborate securely with peers, emerging biotechs, CROs, academic institutions, healthcare systems, AI developers and any other partner or data source while maintaining privacy, compliance, and IP integrity.

Federated Use Cases

Albeit an emerging technique, FC is not a theoretical innovation. It's a practical solution addressing barriers in AI-driven discovery, research and development, to accelerate innovation cycles. There are many applications for FC; in the simplest of terms, FC adds immense value in any scenario where external, highly fragmented and regulated data would improve predictivity. Some example use cases include:

Predictive Protein, Molecular, ADME, and Toxicity Modeling: FC enables researchers to train or fine-tune protein, molecular, ADME, and toxicity models on distributed and proprietary datasets and chemical libraries. This approach improves model generalizability (e.g. folding prediction, binding affinity estimation); enables richer structure-activity mappings while keeping unique compounds and chemical structures confidential; enhances the robustness of in silico models and improves predictivity in early drug design - all of which can significantly reduce late-stage attrition rates. One example of FC supporting predictive modeling is the FAITE Consortium, comprised of AbbVie, Amgen, Astrazeneca, Johnson & Johnson, and UCB, collectively training models to predict properties of biologics, including viscosity, aggregation, thermal stability, and chemical liability.

Pre-Clinical R&D: With FC, pharma can develop partner ecosystems based on model-data exchanges, where proprietary models and data are ‘exchanged’ to collaboratively develop, train, and fine-tune models without sharing data or IP. A recent example of this is Eli Lilly’s Federated AI program, Lilly TuneLab, which lets biotechs securely run inference and fine-tune Lilly’s proprietary models on their own data - without moving data or IP. Biotechs benefit from gaining access to Lilly’s models, and Lilly benefits from gaining access to more training data as well as building a pipeline of potential acquisition targets.

Clinical Trial Optimization: FC can help optimize clinical trial design, operations, and outcomes. Federated deployment of prescreening algorithms can securely screen distributed patient populations based on inclusion/exclusion criteria without exposing any patient-level data, resulting in faster more accurate patient recruitment, better site selection, reduced burden on site staff and fewer delays. AI models can also be trained on federated real-world and partner data to simulate protocol criteria, test design feasibility across diverse populations, or forecast quality issues - reducing costly, mid-trial amendments and improving representativeness of enrolled populations.

Pharmacovigilance and Safety Monitoring: Federated pharmacovigilance models can be deployed to and monitor distributed datasets in real time, analyzing lab values, adverse event reports, and EHR data; this enables continuous safety surveillance at scale across global sites, without centralizing sensitive patient-level data. The results are earlier detection of safety signals, more proactive risk mitigation, and improved regulatory compliance.

Real-World Evidence (RWE): With FC, researchers can establish evergreen federated real-world data (RWD) networks that simplify collaboration and support ongoing evidence generation—avoiding the inefficiency of rebuilding for each new analysis or study. Through these federated networks, researchers can harmonize multi-modal and multi-geography data, deploy models and conduct advanced analysis to evaluate comparative effectiveness, improve understanding of treatment patterns, and build more comprehensive value stories. The EXAM study is one of the first examples of how a federated network of global sites can be used to predict clinical outcomes, in that case in Covid-19 patients. Another example is AstraZeneca’s Beam - Illuminating Healthcare Through Data Center programthat is using FC to partner with Tel Aviv University and the Meuhedet and Leumit health funds to analyze RWD to understand disease mechanisms, new treatment methods and how to improve patients’ quality of life.

External Innovation: The inherent privacy-preserving characteristics of FC make it a natural fit for use when developing multi-institution research consortia. The objectives for the consortia can be focused on drug development specifically, or be more broadly centered on evaluating disease patterns and predicting outcomes to support treatment targeting and care delivery. For example, the Cancer AI Alliance (CAIA) is a consortium focused on connecting siloed cancer research from different treatment centers. CAIA enables researchers and clinicians to train AI models that learn from participating cancer centers’ millions of clinical data points with the goal of accelerating treatment development and improving care.

A Federated Future

Ultimately, AI in drug discovery and development doesn’t have a data problem. It has a data access problem. FC allows organizations to securely and compliantly access and collaborate around data (no matter where it resides) and AI, responsibly and at scale. It’s how pharma can scale trusted collaboration, accelerate timelines, and unlock the full potential of its investments.

In the race to bring drugs to market faster, the winners won’t be those who have pooled the most data, but those who can learn from the most data. Federated Computing makes that possible.