Our computational drug discovery capabilities are enhanced by incorporating state-of-the-art machine learning models and cheminformatics tools to facilitate rapid and efficient discovery and design of novel compounds from hit identification through to lead optimization.

Leverages FASTA sequence (1D representation of the protein) and QSAR model in combination to predict the Bioactivity in low data regimes. Proprietary technology that leverages protein sequence data to increase the accuracy of the model, when the protein structure is not available.

Ro5’s Clinical Trial Analytics module identifies the most important biomarkers and contextualizes biological processes using the knowledge graph. Pharmaceutical companies can thus react fast during the ongoing trials, leveraging the tool’s ability to highlight the important signals and interrogate them with biological context extracted from our knowledge graph.

A unique and proprietary 3D Bioactivity model that is based on point cloud technology. It leverages the 3D structure of the binding site and the 3D structure of the molecule and predicts how each atom of the ligand interacts with every atom of the binding site, through attention mechanisms. Moreover, using attention, we can visualize the specific interactions that contribute positively or negatively to bioactivity.

A unique and proprietary 3D Convolutional Neural Network system that learns spatial information of the molecule-protein complex to predict bioactivity. It consists of a bioactivity predictor and pose classifier. Once a correct pose is chosen, the bioactivity can be predicted. It is superior to other state-of-the-art models due to its ability to disentangle whether the pose is correct or incorrect. This model is one of the core components of our virtual screening capabilities, as well as 3D de novo, which guides the generation of the molecule.

Sophisticated generative process with transcription from 3D spaces to SMILES that allows for the optimization of chemical properties and bioactivity, generating new molecules with similar pharmacophore distributions while maintaining key functional groups. Currently, there is no effective way to traverse between 3D space to a comprehensible and computer-interpretable representation without Ro5’s platform – this is one of the key components that makes our 3D De Novo both reliable and scalable.

A framework to store biomedical data and contextualize it using existing scientific knowledge. During the drug discovery process, or related academic work, researchers need to deal with large datasets, which limits their ability to take into account all of this data and see the complete picture. By storing and contextualizing these datasets using Ro5’s knowledge graph, AI and EDA platforms, we provide a big picture view to the scientist of their work, its relationship with all existing work, and new hypotheses.

Novel and/or optimized compounds must be synthesizable, both for industrial purposes and for verification (screening). The ease of synthesis highly influences the scale of compounds to be tested in vitro. For that matter, it is paramount to verify that compounds can be synthesized, preferentially from commercially available precursors. To that end, our AI-based retrosynthesis system constructs a backward chain of reactions that successively leads to the target (root) molecule, deconstructing a rather complex, and often commercially unavailable target molecule with its synthesizable descendants.

The scalable substructure extraction and clustering tool allows us to operate around molecules based on their ADMET properties and activity, thereby providing deeper insights. Chemists typically do not have the tools to draw such insights from large amounts of data, although this can be effectively accomplished with Ro5’s substructure and scaffold extraction combined with our clustering algorithms.