A5100700305- Computational Drug Discovery Methods
- Machine Learning in Materials Science
- Protein Structure and Dynamics
- Media and Digital Communication
- Various Chemistry Research Topics
- Genetics, Bioinformatics, and Biomedical Research
- Cosmology and Gravitation Theories
- Black Holes and Theoretical Physics
- Microbial Natural Products and Biosynthesis
- Click Chemistry and Applications
- Analytical Chemistry and Chromatography
- Advanced Differential Geometry Research
- Receptor Mechanisms and Signaling
- Explainable Artificial Intelligence (XAI)
- Chemistry and Chemical Engineering
- Cell Image Analysis Techniques
- Metabolomics and Mass Spectrometry Studies
- Wireless Communication Networks Research
- Gaussian Processes and Bayesian Inference
- Bluetooth and Wireless Communication Technologies
- Advanced Graph Neural Networks
- Scientific Computing and Data Management
- Clostridium difficile and Clostridium perfringens research
- Satellite Communication Systems
- Advanced Bandit Algorithms Research
Microsoft Research (United Kingdom)
2022-2024
ETH Zurich
2020-2022
Boehringer Ingelheim (Germany)
2021-2022
Universitat Pompeu Fabra
2017-2022
Barcelona Biomedical Research Park
2017-2022
Applied Mathematics (United States)
2020
Birkbeck, University of London
1999
Central University of Venezuela
1979-1987
Accurately predicting protein–ligand binding affinities is an important problem in computational chemistry since it can substantially accelerate drug discovery for virtual screening and lead optimization. We propose here a fast machine-learning approach using state-of-the-art 3D-convolutional neural networks compare this to other scoring methods several diverse data sets. The results the standard PDBbind (v.2016) core test-set are with Pearson's correlation coefficient of 0.82 RMSE 1.27 pK...
Abstract Motivation An important step in structure-based drug design consists the prediction of druggable binding sites. Several algorithms for detecting cavities, those likely to bind a small compound, have been developed over years by clever exploitation geometric, chemical and evolutionary features protein. Results Here we present novel knowledge-based approach that uses state-of-the-art convolutional neural networks, where algorithm is learned examples. In total, 7622 proteins from scPDB...
In this work, we propose a machine learning approach to generate novel molecules starting from seed compound, its three-dimensional (3D) shape, and pharmacophoric features. The pipeline draws inspiration generative models used in image analysis represents first example of the de novo design lead-like guided by shape-based A variational autoencoder is perturb 3D representation followed system convolutional recurrent neural networks that sequence SMILES tokens. scaffolds functional groups can...
Abstract Machine learning approaches in drug discovery, as well other areas of the chemical sciences, benefit from curated datasets physical molecular properties. However, there currently is a lack data collections featuring large bioactive molecules alongside first-principle quantum information. The open-access QMugs (Quantum-Mechanical Properties Drug-like Molecules) dataset fills this void. collection comprises mechanical properties more than 665 k biologically and pharmacologically...
Graph neural networks are able to solve certain drug discovery tasks such as molecular property prediction and de novo molecule generation. However, these models considered "black-box" "hard-to-debug". This study aimed improve modeling transparency for rational design by applying the integrated gradients explainable artificial intelligence (XAI) approach graph network models. Models were trained predicting plasma protein binding, hERG channel inhibition, passive permeability, cytochrome P450...
Many molecular design tasks benefit from fast and accurate calculations of quantum-mechanical (QM) properties. However, the computational cost QM methods applied to drug-like molecules currently renders large-scale applications quantum chemistry challenging. Aiming mitigate this problem, we developed DelFTa, an open-source toolbox for prediction electronic properties at density functional (DFT) level theory, using Δ-machine-learning. Δ-Learning corrects error (Δ) a but inaccurate property...
Structure-based drug discovery methods exploit protein structural information to design small molecules binding given pockets. This work proposes a purely data driven, structure-based approach for imaging ligands as spatial fields in target We use an end-to-end deep learning framework trained on experimental protein-ligand complexes with the intention of mimicking chemist's intuition at manually placing atoms when designing new compound. show that these models can generate images ligand...
The capability to rank different potential drug molecules against a protein target for potency has always been fundamental challenge in computational chemistry due its importance design. While several simulation-based methodologies exist, they are hard use prospectively and thus predicting lead optimization campaigns remains an open challenge. Here we present the first machine learning approach specifically tailored ranking congeneric series based on deep 3D-convolutional neural networks....
The lead optimization process in drug discovery campaigns is an arduous endeavour where the input of many medicinal chemists weighed order to reach a desired molecular property profile. Building expertise successfully drive such projects collaboratively very time-consuming that typically spans years within chemist's career. In this work we aim replicate by applying artificial intelligence learning-to-rank techniques on feedback was obtained from 35 at Novartis over course several months. We...
Following the sequence and structure revolutions, predicting dynamical mechanisms of proteins that implement biological function remains an outstanding scientific challenge. Several experimental techniques molecular dynamics (MD) simulations can, in principle, determine conformational states, binding configurations their probabilities, but suffer from low throughput. Here we develop a Biomolecular Emulator (BioEmu), generative deep learning system can generate thousands statistically...
Abstract Summary Virtual screening pipelines are one of the most popular used tools in structure-based drug discovery, since they can reduce both time and cost associated with experimental assays. Recent advances deep learning methodologies have shown that these outperform classical scoring functions at discriminating binder protein-ligand complexes. Here, we present BindScope, a web application for large-scale active-inactive classification compounds based on convolutional neural networks....
Feature attribution techniques are popular choices within the explainable artificial intelligence toolbox, as they can help elucidate which parts of provided inputs used by an underlying supervised-learning method considered relevant for a specific prediction. In context molecular design, these approaches typically involve coloring graphs, whose presentation to medicinal chemists be useful making decision compounds synthesize or prioritize. The consistency highlighted moieties alongside...
Protein design often begins with knowledge of a desired function from motif which motif-scaffolding aims to construct functional protein around. Recently, generative models have achieved breakthrough success in designing scaffolds for diverse range motifs. However, the generated tend lack structural diversity, can hinder wet-lab validation. In this work, we extend FrameFlow, an SE(3) flow matching model backbone generation, perform two complementary approaches. The first is amortization,...
Bayesian optimization has risen over the last few years as a very attractive method to optimize expensive evaluate, black box, derivative-free and possibly noisy functions (Shahriari et al. 2016).This framework uses surrogate models, such likes of Gaussian Process (Rasmussen Williams 2004) which describe prior belief possible objective in order approximate them.The procedure itself is inherently sequential: our function first evaluated times, model then fit with this information, will later...
We present FrameFlow, a method for fast protein backbone generation using SE(3) flow matching. Specifically, we adapt FrameDiff, state-of-the-art diffusion model, to the flow-matching generative modeling paradigm. show how matching can be applied on and propose modifications during training effectively learn vector field. Compared FrameFlow requires five times fewer sampling timesteps while achieving two fold better designability. The ability generate high quality samples at fraction of cost...
Drug discovery suffers from high attrition because compounds initially deemed as promising can later show ineffectiveness or toxicity resulting a poor understanding of their activity profile. In this work, we describe deep self-normalizing neural network model for the prediction molecular pathway association and evaluate its performance, showing an AUC ranging 0.69 to 0.91 on set extracted ChEMBL 0.81 0.83 external data provided by Novartis. We finally discuss applicability proposed in...
A procedure given by Newman and Janis, to obtain the exterior Kerr metric from Schwarzschild performing a complex coordinate transformation, is applied an interior spherically symmetric metric. The resulting can be matched on boundary of source which chosen oblate spheroid. specific example solution for energy density positive in detail.
Deep learning has been successfully applied to structure-based protein–ligand affinity prediction, yet the black box nature of these models raises some questions. In a previous study, we presented KDEEP, convolutional neural network that predicted binding given complex while reaching state-of-the-art performance. However, it was unclear what this model learning. work, present new application visualize contribution each input atom prediction made by network, aiding in interpretability such...
While a plethora of different protein–ligand docking protocols have been developed over the past twenty years, their performances greatly depend on provided input pair. In this study, we machine-learning model that uses combination convolutional and fully connected neural networks for task predicting performance several popular given protein structure small compound. We also rigorously evaluated our using widely available database complexes types data splits. further open-source all code...
A set of junction conditions is stated in terms the Newman-Penrose variables (tetrad vectors and spin coefficients). It shown that these are equivalent to those Darmois Lichnerowicz. As an example we study matching Schwarzschild metric with axially reflection-symmetric metric. For this particular propagation Killing show how conditioned by fulfillment conditions.
Abstract Explainable machine learning is increasingly used in drug discovery to help rationalize compound property predictions. Feature attribution techniques are popular choices identify which molecular substructures responsible for a predicted change. However, established feature methods have so far displayed low performance deep algorithms such as graph neural networks (GNNs), especially when compared with simpler modeling alternatives random forests coupled atom masking. To mitigate this...
Graph neural networks are able to solve certain drug discovery tasks such as molecular property prediction and de novo molecule generation. However, these models considered 'black-box' 'hard-to-debug'. This study aimed improve modeling transparency for rational design by applying the integrated gradients explainable artificial intelligence (XAI) approach graph network models. Models were trained predicting plasma protein binding, cardiac potassium channel inhibition, passive permeability,...