Cerella

Cerella is a deployable AI platform developed by Optibrium that accelerates drug discovery by extracting hidden insights from complex, sparse, and uncertain experimental data. Designed for medicinal chemists, computational scientists, and R&D teams in pharmaceutical, agrochemical, and related life sciences industries, Cerella enables a streamlined journey from drug concept to candidate by applying deep learning to predict compound properties, prioritise experiments, and guide decision-making with confidence.

Drug discovery data presents significant challenges: vast chemical spaces, expensive downstream experiments, missing or inaccurate measurements, and complex structure-activity relationships all make it difficult to identify the best compounds and allocate resources effectively. Cerella addresses these challenges directly, providing robust, uncertainty-aware predictions even when working with limited data sets, and helping teams avoid unnecessary experiments while maximising the return on investment of their research efforts.

Core Capabilities

• Uses early-stage experimental data to predict late-stage outcomes, reducing the need for expensive downstream measurements that cannot be addressed by conventional methods.
• Uncovers hidden opportunities by accurately filling in missing, uncertain, or inaccurate data across a project's experimental matrix.
• Creates value from all available data, including results from parallel or previous projects, to inform future discovery strategies.
• Identifies the most valuable measurements to perform, maximising the return on investment of experimental efforts.
• Targets high-quality candidate compounds with confidence through robust uncertainty estimates attached to every prediction.
• Handles large-scale datasets, demonstrated by processing 700,000 compounds across 1,000 experimental endpoints for Takeda Pharmaceuticals.

Demonstrated Real-World Success

• MSD: Deep learning transformed the accuracy of bioactivity and property predictions, with results used to guide experimental prioritisation.
• AstraZeneca: Cerella translated in vitro data to in vivo savings, with findings published in Molecular Pharmaceutics.
• Constellation Pharmaceuticals: Saved the equivalent of two months of work for 12 chemists by identifying inactive compounds; published in the Journal of Chemical Information and Modeling.
• FMC: Guided compound purchasing and screening decisions to improve efficiency in agrochemical discovery.
• International Flavors & Fragrances (IFF): Accurately predicted complex in vivo sensory responses; published in the Journal of Computer-Aided Molecular Design.
• Genentech: Achieved a 76% reduction in the number of assays required across kinase programmes.
• Open Source Malaria: Identified the only compound experimentally proven as active in the OSM competition; published in the Journal of Medicinal Chemistry.
• Takeda Pharmaceuticals: Successfully handled 700,000 compounds and 1,000 experimental endpoints; published in Applied AI Letters.
• Undisclosed collaborator: Obtained accurate in vivo pharmacokinetic profiles from limited in vitro ADME data.

Scientific Foundation

• Cerella is built on a novel deep learning neural network method specifically designed for the imputation of bioactivity and property data, as described in peer-reviewed literature dating from 2019.
• The platform supports both QSAR-style predictive modelling and imputation-based approaches, enabling teams to select the most appropriate method depending on the nature and completeness of their data.
• A proof-of-concept study conducted in collaboration with Intellegens, the University of Cambridge, and AstraZeneca demonstrated Cerella's ability to predict rat in vivo pharmacokinetic parameters and full concentration–time PK profiles from chemical structure alone.

Cerella is a deployable solution offered by Optibrium, making it suitable for integration into existing R&D workflows across pharmaceutical, agrochemical, and flavour and fragrance organisations. Teams interested in adopting AI-guided discovery can contact Optibrium directly to explore how Cerella can be applied to their specific projects and data environments.