{"id":9004111222145616,"date":"2026-09-23T15:32:39","date_gmt":"2026-09-23T14:32:39","guid":{"rendered":"https:\/\/www.hgf.com\/?p=9004111222145616"},"modified":"2026-09-23T15:32:39","modified_gmt":"2026-09-23T14:32:39","slug":"protecting-the-ai-drug-discovery-ecosystem-building-an-ip-stack-around-targets-molecules-and-medicines","status":"publish","type":"post","link":"https:\/\/www.hgf.com\/de\/wissens-hub\/protecting-the-ai-drug-discovery-ecosystem-building-an-ip-stack-around-targets-molecules-and-medicines\/","title":{"rendered":"Protecting the AI Drug Discovery Ecosystem: Building an IP Stack Around Targets, Molecules and Medicines"},"content":{"rendered":"

Despite the considerable costs and effort in target identification, candidate generation, selection and advancement to clinical trials, 90% of candidates that enter clinical trials during drug discovery fail (Sun D, 2022). The complexity of biological systems remains a major stumbling block in the drug development pipeline. Large datasets such as multiomic datasets and complex analysis methods (for example genome-wide association studies, transcriptomic profiling, proteomic interaction mapping, and metabolic sequencing) have been used to help identify targets with the potential to interact with a given drug and to also identify biological relationships that influence therapeutic activity (Liu Y. Z., 2026) (Pun, 2026) (Rehman AU, 2024). The combination of multiomics with AI, can identify complex relationships across heterogeneous and noisy datasets and can further correlate variables that may be difficult to detect using conventional statistical approaches. This may enable improved prediction of candidate molecules that modulate disease biology (Genentech, 2024).<\/p>\n

Traditionally, pharmaceutical patent portfolios have centered on protecting therapeutic compounds and their medical uses. However, ongoing developments in relation to the use of multiomics, biological datasets and AI-enabled analysis, means commercially significant value may be generated long before a candidate molecule enters clinical development and may be created by different organisations, for example TechBio, working alongside traditional pharma. Proprietary datasets, disease models, target identification systems, biomarker strategies and machine-learning workflows may become valuable assets in their own right. The challenge is no longer simply protecting a medicine. It is protecting the ecosystem that created it.<\/p>\n

Intellectual Property strategy for TechBio organisations requires a broader and more integrated approach moving beyond patent protected chemistry and instruments to include assets which are typically thought of as challenging or undesirable to patent. This requires the IP team to consider an IP portfolio architecture that considers not only patents in relation to the underlying tools, but also the way data is owned, licensed, and shared. Trade Secrets, contractual and database right strategies can be utilised as tools to capture IP rights and commercial value around multimodal datasets for identification, selection and validation of the targets that subsequently leads to generation of potential candidate molecules\u00a0(Roche, 2026).<\/p>\n

In addition to creating a broader IP portfolio to ensure the IP and commercial strategies are aligned, Freedom to Operate (FTO) risk will have to move beyond considering competitor patent rights to evaluate potential challenges and restrictions arising across a complex omics stack. FTO may increasingly depend on access to data, computational tools and platform integration technologies. A company may own patents covering a therapeutic candidate, yet remain constrained by third-party dataset licences, restrictions on model training, proprietary biomarker rights or contractual obligations arising from collaborative development programmes. Consequently, FTO analyses may need to assess both conventional patent risks and rights associated with the broader discovery ecosystem\u00a0(Zhavoronkov, 2026),\u00a0(Liu B. H., 2025).<\/p>\n

IP portfolio considerations at different layers of discovery and drug development process<\/strong><\/p>\n\n\n\n\n\n\n
Layer <\/strong><\/td>\nConsiderations<\/strong><\/td>\n<\/tr>\n
Biology – Biological insight \u2013 Disease understanding and identification of where target intervention can take place<\/strong><\/p>\n

 <\/p>\n

May include data and AI sublayers for example multiomic datasets, AI-enabled prediction models and workflows<\/em><\/td>\n

\n
    \n
  • Ensure clear understanding of the data used (curated datasets, public data), model weights, training data, ownership, access controls to such data and right to use data.<\/li>\n
  • Understanding the contribution made by human researchers remains important when assessing inventorship and ownership issues arising from AI-assisted processes. Organisations should maintain appropriate records of human decision-making, including target selection, model design, hypothesis generation, interpretation of outputs and candidate selection (see TJ008\/20 (EPO, 2021) and USPTO inventorship guidance (USPTO, 2026)).<\/li>\n
  • Where a classification method serves a technical purpose, the steps of generating the training set and training the classifier may also contribute to the invention’s technical character (EPO, 2026).<\/li>\n
  • Simply applying machine learning to biological data is unlikely to be sufficient. Patentability will frequently depend on demonstrating a technical purpose and technical effect associated with the claimed invention.<\/li>\n
  • Consider if the method should be publicly disclosed (as would be required for patent application filing) or if it should be kept as a Trade Secret.<\/li>\n
  • Consider if the \u2018method type\u2019 claim would be enforceable if patent right obtained.<\/li>\n<\/ul>\n

     <\/td>\n<\/tr>\n

Chemistry \u2013 screening, generation and synthesis<\/strong><\/p>\n

 <\/p>\n

May include sublayers including\u00a0 target selection, <\/em>generation of candidate molecules against target, multiple cycles of in silico optimisation, characterisation of candidate molecules and experimental validation <\/em><\/p>\n

 <\/td>\n

\n
    \n
  • Consideration of the way in which the design of candidate molecules, optimization strategies, and the selection of output is directed by humans for inventorship and ownership issues.<\/li>\n
  • Consider claims types such as Markush claims, claims to lead compounds, synthesis steps, stereoisomers, salts, formulations.<\/li>\n
  • Consider supporting evidence of candidate molecules for patent protection \u2013 clear technical effect \u2013 structure \/ function \u2013 therapeutic link.<\/li>\n
  • In order to meet the requirements of Art. 83 EPC, the proof of a claimed therapeutic effect has to be provided in the application as filed, in particular if, in the absence of experimental data in the application as filed, it would not be credible to the skilled person that the therapeutic effect is achieved. An insufficiency in this respect cannot be remedied by post-published evidence (G2\/21)\u00a0(EPO, 2023)\u00a0(EPO, 2026).<\/li>\n<\/ul>\n<\/td>\n<\/tr>\n
Clinical – Medical Use – clinical validation<\/strong><\/p>\n

 <\/p>\n

May include <\/em>biomarker selection, <\/em>patient stratification, and <\/em>clinical feedback to earlier layers<\/em><\/td>\n

\n