Cerella navigated complex in vitro-in vivo correlations within a sparse data set to make accurate predictions for an in vivo pharmacokinetic (PK) profile and indicate compounds that met the PK requirements for this project.
The Challenge
Drug discovery is a multi-parametric challenge, trying to balance drug efficacy with safety, bioavailability and other PK properties. In this example, our collaborator had identified potential anti-infective compounds with good activity, but without the necessary PK profile to match.
Our Targets
Activity: Must meet or exceed five activity endpoint targets, measured in cell-based assays
ADME: Must have good microsomal stability, permeability, solubility and plasma protein binding
PK*: Must have a high Cmax, long half-life, low IV clearance and good oral bioavailability
*PK targets for mouse model, in which compounds were being measured
The Solution
Cerella was trained on a data set of 2,260 compounds across over 65 activity, ADME and PK endpoints.
The activity data had come from this specific anti-infectives project, but their ADME and PK data spanned multiple other projects. This was useful, as it meant our models had the opportunity to learn relationships between ADME and PK parameters across a diverse range of chemistry. Overall, the data was very sparse, with only 10% of all possible measurements recorded.
Cerella’s predictions and uncertainty estimates were combined with a multi-parameter scoring profile to weigh up the importance of different properties against the compound’s objectives. Based on this, only of 3 out of 3,000 newly designed compounds were selected for further progression, and comparison with the initial lead anti-infective compound. Data for these compounds were measured and compared to Cerella’s predictions.
All the new compounds displayed excellent in vivo PK profiles, even when some in vitro PK parameters had not been met. Cerella was able to navigate the complex relationships between in vitro and in vivo properties, and enable our collaborators to minimise the number of expensive in vivo measurements they needed to carry out, quickly identifying high quality candidates.