This webinar is organised by the ELIXIR 3D-BioInfo Community

            The event is hosted by










                            Dr Vincent Zoete (Chair) 


                            Dr. Gonzalo Parra


                            Dr Neeladri Sen           




                             Swiss Institute of

                            Bioinformatics (SIB)  


                            Barcelona Supercomputing Center  (BSC)  

                        University College London (UCL)




            Programme: 

         

SILVR: Conditioning Diffusion Models for Fragment-Based Small Molecule
Generation

Dr. Antonia Mey
(The University of Edinburgh, UK) 

Diffusion models have proven to be a powerful tool in image generation and, more recently, in small molecule generation. Broadly, molecular diffusion models can generate random molecules, however, the generation of small molecules tailored to specific protein pockets is much more challenging.
In this talk, I will introduce Selective Iterative Latent Variable Refinement (SILVR), a novel method designed to condition existing equivariant diffusion models based on X-ray fragment hits. SILVR is a conditioning method that does not require additional training. The conditioning is achieved in the latent space of the trained equivariant diffusion model using the SILVR rate as a parameter to vary the level of conditioning.
This runtime modification in combination with X-ray fragments allows for the generation of new molecules that fit the binding site of the target protein. Furthermore, it is possible to link, merge, and extend fragments. I will show the capabilities of SILVR on a dataset of SARS-CoV-2 main protease fragments from the Diamond X-Chem COVID Moonshot dataset.
This novel method sits at the interface between experimental data and generative models, offering a direct tool for enhancing small molecule generation. Its advantage lies in its broad applicably to any protein target for which fragment hits are available and is a promising method in the arsenal of tools used in fragment-based drug design.

    Two-Step Covalent Docking with Attracting Cavities

    Dr. Mathilde Goullieux
        (Swiss Institute of Bioinformatics)



    Molecular docking is a computational approach used to predict the most probable pose of a ligand in a protein binding site. Recently, our docking code, Attracting Cavities1 (AC), has undergone significant enhancements aimed at improving its sampling procedure, robustness and flexibility2.

    Given the efficacy and advantages of covalent drugs, such as beta-lactam antibiotics or proton-pump inhibitors, understanding and predicting their interactions with biological targets is of utmost importance. Consequently, we implemented a covalent docking procedure into AC. This new feature mimics the two-step process of covalent ligand binding. First, non-bonded interactions drive ligand binding to the protein, and second, a chemical reaction leads to the formation of a new covalent bond between the ligand and the protein.

    AC 2.0 was rigorously tested on 285 complexes from the PDBbind Core set (2016 version) and achieved a success rate of 73%, surpassing the performance of the widely used docking codes GOLD (64%) and AutoDock Vina (58%) in non-covalent redocking experiments. Additionally, we evaluated the covalent docking algorithm using a benchmark set of 304 experimentally resolved covalent complexes. The results showed that our approach outperformed the two state-of-the-art covalent docking codes, AutoDock4 (66%) and GOLD (35%), with a success rate of 78%.

    In parallel, we developed a suite of tools designed to make docking calculations accessible to non-expert users. These tools are freely accessible through two web servers. SwissParam 2023 generates force field topologies and parameters for small molecules, both for non-covalent and covalent docking3. SwissDock 2024, which will be released soon, will host tools for target preparation and will enable docking with AC and Autodock Vina. Both web servers will be described during the present webinar.

    You can find previous webinars from the 3D-BioInfo Community on the Community webinars page.