DeepEnzyme

The project DeepEnzyme was a 2020 winner of the PREMATURATION call for projects, led by the Innovation Lab of IP Paris.

Our AI targets in-silico drug design, developing algorithms and software to assess molecule efficiency and discover new active compounds.

The project relies on a cooperation between two laboratories from Ecole Polytechnique, combining their expertise in deep learning, structural bio-informatics, and biochemistry.

This cross-functional approach aims to enhance the accuracy of screening processes, reducing experimental costs and shortening development time. By identifying weak molecular interactions early, it increases the likelihood of hit identification, ultimately improving the overall efficiency of drug discovery and advancing the development of successful treatments.

The team is developing a hybrid in-silico approach for large-scale drug prediction.

Their AI consists of:

• A multimodal, physics-inspired deep learning approach predicting quantitative protein-ligand interactions (PLI) by incorporating a novel AI-based solvation model. This approach leverages recent developments in Large Language Models (LLMs) and Graph Machine Learning (GML) by analyzing protein structures as both amino acid sequences and geometric graphs.
• Fitting of state-of-the-art diffusion and geometric graph neural network models for molecules and proteins, addressing modern challenges in drug discovery.
• The unique strength of our AI model lies in its training on a dataset generated from physics-based simulations, which is significantly larger than available experimental data. This makes our AI more universal, robust, and accurate.

• A unique database built from public data, enriched with experimental results from École Polytechnique. We leverage in-house expertise on proteins dependent on vitamin B9, key targets for anti-microbial and anti-cancer therapies.
• A novel state-of-the-art deep learning approach incorporating protein, molecule and solvent information.