SGPocket: A New Graph Convolutional Neural Network for Ligand-protein Binding Site Prediction.

BACKGROUND: Drug research is a long process, taking more than 10 years and requiring considerable financial resources. Therefore, researchers and industrials aim to reduce time and cost. Thus, they use computational simulations like molecular docking to explore huge databases of compounds and extract the most promising ones for further tests. Structure-based molecular docking is a complex process mixing surface exploration and energy computation to find the minimal free energy of binding corresponding to the best interaction location. OBJECTIVE: Our work is developed in the ligand-protein context, where ligands are small compounds like drugs. In most cases, no information is known about where on the protein surface the ligand will bind. Thus, the whole protein surface must be explored, which takes a huge amount of time. METHODS: We have developed SGPocket (meaning Spherical Graph Pocket), a binding site prediction method. Our method allows us to reduce the explored protein surface using deep learning without any information about a ligand. SGPocket uses the spherical graph convolutional operator working on a spherical relative positioning of amino acids in the protein. Then, a final step of clustering extracts the binding sites. RESULTS: Tested and compared (with well-known binding site prediction methods) on a hand-made dataset, our method performed well and can reduce the docking computing time. CONCLUSION: Thus, SGPocket allows the reduction of the exploration surface in the molecular docking process by restricting the simulation only to the site(s) predicted to be interesting.

Machine-learning methods for ligand-protein molecular docking.

Artificial intelligence (AI) is often presented as a new Industrial Revolution. Many domains use AI, including molecular simulation for drug discovery. In this review, we provide an overview of ligand-protein molecular docking and how machine learning (ML), especially deep learning (DL), a subset of ML, is transforming the field by tackling the associated challenges.

Convolutional neural networks and temporal CNNs for COVID-19 forecasting in France.

This paper focus on multiple CNN-based (Convolutional Neural Network) models for COVID-19 forecast developed by our research team during the first French lockdown. In an effort to understand and predict both the epidemic evolution and the impacts of this disease, we conceived models for multiple indicators: daily or cumulative confirmed cases, hospitalizations, hospitalizations with artificial ventilation, recoveries, and deaths. In spite of the limited data available when the lockdown was declared, we achieved good short-term performances at the national level with a classical CNN for hospitalizations, leading to its integration into a hospitalizations surveillance tool after the lockdown ended. Also, A Temporal Convolutional Network with quantile regression successfully predicted multiple COVID-19 indicators at the national level by using data available at different scales (worldwide, national, regional). The accuracy of the regional predictions was improved by using a hierarchical pre-training scheme, and an efficient parallel implementation allows for quick training of multiple regional models. The resulting set of models represent a powerful tool for short-term COVID-19 forecasting at different geographical scales, complementing the toolboxes used by health organizations in France.

AMIDE v2: High-Throughput Screening Based on AutoDock-GPU and Improved Workflow Leading to Better Performance and Reliability.

Molecular docking is widely used in computed drug discovery and biological target identification, but getting fast results can be tedious and often requires supercomputing solutions. AMIDE stands for AutoMated Inverse Docking Engine. It was initially developed in 2014 to perform inverse docking on High Performance Computing. AMIDE version 2 brings substantial speed-up improvement by using AutoDock-GPU and by pulling a total revision of programming workflow, leading to better performances, easier use, bug corrections, parallelization improvements and PC/HPC compatibility. In addition to inverse docking, AMIDE is now an optimized tool capable of high throughput inverse screening. For instance, AMIDE version 2 allows acceleration of the docking up to 12.4 times for 100 runs of AutoDock compared to version 1, without significant changes in docking poses. The reverse docking of a ligand on 87 proteins takes only 23 min on 1 GPU (Graphics Processing Unit), while version 1 required 300 cores to reach the same execution time. Moreover, we have shown an exponential acceleration of the computation time as a function of the number of GPUs used, allowing a significant reduction of the duration of the inverse docking process on large datasets.