Welcome to the Era of AI-Driven Drug Discovery
ROSALINDᴬᴵ is the first AI-powered drug discovery engine for pharmaceutical R&D. We democratise access to cutting-edge AI that manages sparse, noisy and small-data challenges enabling drug hunters to drive innovations and cure diseases.
ROSALINDᴬᴵ automates the application of AI by providing a large repository of pretrained and customisable AI models particularly suited to data challenges and address novel and first-in-class targets.
Our partners use ROSALINDᴬᴵ to develop, scale and apply innovative AI solutions to discover novel chemical scaffolds, optimise activities and chemical properties and screen ultra-large chemical libraries.
New Way To
Discover Novel Therapeutics
We design novel small-molecule therapeutics from uncharted areas in chemical space. Our computational tools have learned from billions of known chemicals to design and evaluate new chemical structures never seen before.
Our deep learning methods can address challenging and first-in-class targets with limited available data.
Structure-based Drug Design
Our methods screen and optimise novel therapeutics for specific targets. We extract biological and contextual data from the crystal structure and the binding domain to explore a large search space relevant to the target of interest.
We can design and screen huge chemical libraries to select the most promising compounds with unprecedented speed and accuracy.
Accurate and Data Efficient Property Prediction
Our methods are the most accurate for prediction of ADMET and physicochemical properties critical for progressing drug discovery programs.
We use active and transfer learning together with our proprietary data-efficient methods to ensures that we learn from every experiment to minimise lab testing while producing the best results.
The Future of the Design-make-test-analyse Cycle
Our small molecules are designed with the best possible profile from the getgo. We consider 10s of ADMET and physicochemical properties in the design of novel therapeutics, using reinforcement and active learning to optimise the insilico DMTA cycle. Using our computational closed-loop system, we ensure our generated compounds have the best chance of success.
Discover the value of ROSALINDᴬᴵ
Novel chemistry for challenging targets
Working on a challenging oncology target with a protein-protein interaction pocket, ROSALINDᴬᴵ's de novo design coupled with accurate predictors of activities discovered a novel scaffold in weeks.
ROSALINDᴬᴵ's computational design-test process can start with very limited data. We seeded Rosalindᴬᴵ with the protein crystal structure and just one fragment. Our automated computational design and optimisation process explores a virtual library of millions of compounds and screens these libraries to select the most promising for lab validation.
ROSALINDᴬᴵ completed the computational screening, identifying novel active scaffold, in under six weeks and by lab testing only three compounds.
With ROSALINDᴬᴵ all we need is the 3D model of the protein and one starting data point.
Improving DEL screening
By training our models on entire DEL selections, we can significantly increase the hit rate and the diversity of the screening outcome. Our methods can be trained on extremely unbalanced DEL libraries and still learn generalizable patterns to classify compounds as active or inactive without any information about the target.
Using our computational approach, we can use ML to screen commercial libraries of 10s of millions of compounds in hours and create new focused libraries containing only the likely active compounds.
This process drastically increases the success rate and reduces the time and the cost required to identify hit compounds.
Accurate activity prediction
We've tested ROSALINDᴬᴵ on a large collection of ADMET and physicochemical properties. The result is a comprehensive set of models achieving the best possible accuracy and remarkable generalisation to unseen data and scaffolds.
We know lab testing is time and resource consuming and we know ML can significantly reduce time and cost. Our models are optimised with active learning to guide lab testing towards data points that are most informative so you only have to test what is absolutely necessary.
With our approach, we optimise for data quality rather than quantity and we aim for smarter data, not bigger data.