CPI collaborated with a major consumer product company to apply advanced computational modelling, machine learning, and pharmacokinetic simulation approaches to support the safety and development assessment of two APIs. The project aligned with the companies' commitment to replace, reduce, and refine animal studies and, wherever possible, reduce the proportion of new product development involving animals.

A multi-model assessment of off-target effects, transporter liabilities, metabolic stability, and predicted metabolites — delivered as a single decision-ready report.

Purpose-built Assay Central models across three targets — including a cross-species reconstruction — applied to a client molecule, its competitor, and thousands of patent derived analogues.

Machine learning and generative design were used to create selective adenosine A3 receptor agonists, balancing target activity with drug-like properties, synthesizability, and patent novelty.

A layered assessment combined BindingDB similarity searches with purpose-built target and kinase models to identify potential liabilities and guide follow-up testing.

Literature mining, machine learning, and large-scale screening were combined to prioritize known and potential inhibitors of a target cytokine.

Validated regression and classification models were used to screen more than 5 million commercially available molecules for dual activity across four kinases.

Microscale thermophoresis was used to test the compound's binding to recombinant human PPT1, alongside control inhibitors.

Machine learning models trained on the client's polymer data guided the design and optimization of new polymers, with later measurements largely confirming the predictions.

A progressive series of machine-learning assessments evaluated consumer product ingredients across estrogen, androgen, and aromatase receptor targets; the resulting work was published in separate papers.

Client data, machine-learning models, and an approved-prodrug-fragment database supported prodrug analog design and predictions relating in-vitro stability and conversion to in-vivo pharmacokinetics.

Generative modeling was adapted to design PROTAC structures and propose strategies to reduce hERG IC50 while improving solubility across multiple PROTACs and a small molecule.

MegaPredict BindingDB, a GraphSage model trained on 475 human kinases, and models for several immunology targets were used to assess a client-provided PROTAC structure.

Literature curation, transporter and cell-based models, corneal PAMPA data, and an ensemble approach were combined to predict and rank compounds for potential ocular bioavailability.