In the vital search for active compounds, there is great value in any approach that can focus experimental programs on the likeliest success routes, or that can deliver breakthrough insights. Machine learning is one such technology, with the potential to extract more value from existing data in order to guide drug discovery programs. But this data is often sparse and noisy – posing challenges to conventional machine learning and data analysis approaches. The Alchemite™ technology can master such data and, in collaboration with Optibrium, it is helping to reduce costs and optimise the drug discovery process.
Active learning for drug discovery with Cerella
Intellegens partners with Optibrium to embed the Alchemite™ deep learning algorithms within Cerella™ – a software platform for active learning in drug discovery. You can gain more value from your compound data, capturing complex relationships between structures, activities and other properties on a scale that cannot be achieved using conventional tools and supporting chemical space exploration beyond conventional QSAR models. Cerella™, powered by Alchemite™, increases confidence in decision making and lead optimisation and enables you to reduce costs and accelerate discovery cycles by targeting experimental resources more effectively.
Finding a novel antimalarial compound
In work now published in the Journal of Medicinal Chemistry, Intellegens and Optibrium succeeded in a global challenge organised by the Open Source Malaria consortium to design new antimalarial compounds with a novel mechanism of action, despite the sparsity of available experimental bioactivity data. The active compound identified by Alchemite™ deep learning show to have good potency when synthesised and tested. Predictions from Alchemite™ outperformed alternative machine learning approaches assessed by the consortium.
Predicting pharmacokinetics parameters and curves
In work now published in Molecular Pharmaceutics, a project team from AstraZeneca, Optibrium, and Intellegens applied the Alchemite™ method for deep learning imputation to the prediction of PK parameters, based on compound structure and sparse in vitro data. This work can reduce time, cost, and number of animal studies in late-stage drug discovery.
Optimising kinase profiling programmes
At a webinar, hosted by the Pistoia Alliance, scientists from Genentech and Optibrium discussed their use of Alchemite™ to optimise kinase profiling programmes. The method was applied to a data set of approximately 650 kinases and 10,000 compounds, significantly outperforming state-of-the-art quantitative structure-activity relationship (QSAR) approaches, including multi-target deep learning.
Alchemite™ succeeds on global pharma data
Intellegens, Optibrium and Takeda Pharmaceuticals have collaborated on a peer-reviewed study that has been published in Applied AI Letters. Working with Takeda’s proprietary global dataset, the team applied Optibrium’s Augmented Chemistry® platform, demonstrating the ability of deep learning imputation to reduce cost and improve success rates of drug discovery.
Intellegens publications on drug discovery
Find out more about the science behind the Alchemite™ methodology for drug discovery:
Alchemite™ for small molecule drug discovery
Alchemite™ technology enables you to:
- Fill gaps in your compound data, avoiding missed opportunities
- Capture complex relationships between structures, activities, and other properties
- Use accurate uncertainty estimates to prioritise compounds most likely to succeed
- Focus your experimental resources on the most valuable measurements
- Explore chemical space beyond conventional QSAR models
- Standardise and share data and models while ensuring rigorous security for your IP