A5075354673- Computational Drug Discovery Methods
- Machine Learning in Materials Science
- Machine Learning in Bioinformatics
- Protein Structure and Dynamics
- Chemical Reactions and Isotopes
- Analytical Chemistry and Chromatography
- Neural Networks and Applications
- Advanced Graph Neural Networks
- Chemical Synthesis and Analysis
- Innovative Microfluidic and Catalytic Techniques Innovation
- Digital Media Forensic Detection
- Target Tracking and Data Fusion in Sensor Networks
- Advanced Memory and Neural Computing
- Click Chemistry and Applications
- Industrial Vision Systems and Defect Detection
- Complex Network Analysis Techniques
- Fuzzy Logic and Control Systems
- Cholinesterase and Neurodegenerative Diseases
- Neural Networks and Reservoir Computing
- Advanced Proteomics Techniques and Applications
- Bioinformatics and Genomic Networks
- Advanced Adaptive Filtering Techniques
- Adversarial Robustness in Machine Learning
- Image Retrieval and Classification Techniques
- Head and Neck Cancer Studies
Bayer (Germany)
2018-2023
Freie Universität Berlin
2020
United States Air Force
1990
Stanford University
1988
The adaptive linear combiner (ALC) is described, and practical applications of the ALC in signal processing pattern recognition are presented. Six examples given, which system modeling, statistical prediction, noise canceling, echo universe channel equalization. Adaptive using neural nets then discussed. concept involves use an invariance net followed by a trainable classifier. It makes multilayer adaptation algorithm that descrambles output reproduces original patterns.< <ETX...
Translation between semantically equivalent but syntactically different line notations of molecular structures compresses meaningful information into a continuous descriptor.
A pattern recognition concept involving first an 'invariance net' and second a 'trainable classifier' is proposed. The invariance net can be trained or designed to produce set of outputs that are insensitive translation, rotation, scale change, perspective etc., the retinal input pattern. scrambled, however. When these fed trainable classifier, final descrambled original patterns reproduced in standard position, orientation, scale, etc. It expected same basic approach will effective for...
One of the main challenges in small molecule drug discovery is finding novel chemical compounds with desirable properties. In this work, we propose a method that combines silico prediction molecular properties such as biological activity or pharmacokinetics an optimization algorithm, namely Particle Swarm Optimization. Our takes starting compound input and proposes new molecules more (predicted) It navigates machine-learned continuous representation drug-like space guided by defined...
An analysis is made of the sensitivity feedforward layered networks Adaline elements (threshold logic units) to weight errors. approximation derived which expresses probability error for an output neuron a large network (a with many neurons per layer) as function percentage change in weights. As would be expected, increases number layers and The essentially independent weights layer, long these numbers are (on order 100 or more).
There has been a recent surge of interest in using machine learning across chemical space order to predict properties molecules or design and materials with desired properties. Most this work relies on defining clever feature representations, which the graph structure is encoded uniform way such that predictions can be made. In work, we propose exploit powerful ability deep neural networks learn representation from low-level encodings huge corpus structures. Our model borrows ideas...
In this work, we propose a novel method that combines in silico prediction of molecular properties such as biological activity or pharmacokinetics with an optimization algorithm, namely Particle Swarm Optimization. Our takes starting compound input and proposes new molecules more desirable (predicted) properties. It navigates machine-learned continuous representation drug-like chemical space guided by de fined objective function. The function multiple models, desirability ranges substructure...
The automatic recognition of the molecular content a molecule's graphical depiction is an extremely challenging problem that remains largely unsolved despite decades research. Recent advances in neural machine translation enable auto-encoding structures continuous vector space fixed size (latent representation) with low reconstruction errors. In this paper, we present fast and accurate model combining deep convolutional network learning from molecule depictions pre-trained decoder translates...
Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework collaborations, it common to exchange datasets by encoding into descriptors. Molecular fingerprints extended-connectivity (ECFPs) are frequently used an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs often considered be non-invertible due way...
The COVID-19 pandemic continues to pose a substantial threat human lives and is likely do so for years come. Despite the availability of vaccines, searching efficient small-molecule drugs that are widely available, including in low- middle-income countries, an ongoing challenge. In this work, we report results open science community effort, "Billion molecules against challenge", identify inhibitors SARS-CoV-2 or relevant receptors. Participating teams used wide variety computational methods...
In silico protein-ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as accurate predictive model could greatly reduce the time resources necessary for detection prioritization possible candidates. Proteochemometric modeling (PCM) attempts to create interaction space by combining explicit protein ligand descriptors. This requires creation information-rich, uniform computer interpretable representations proteins...
Abstract Summary Optimizing small molecules in a drug discovery project is notoriously difficult task as multiple molecular properties have to be considered and balanced at the same time. In this work, we present our novel interactive silico compound optimization platform termed grünifai support ideation of next generation compounds under constraints multiparameter objective. integrates adjustable models, continuous representation chemical space, scalable particle swarm algorithm possibility...
In-silico protein-ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as accurate predictive model could greatly reduce the time resources necessary for detection prioritization possible candidates. Proteochemometric modeling (PCM) attempts to make interaction space by combining explicit protein ligand descriptors. This requires creation information-rich, uniform computer interpretable representations proteins...
Automatic recognition of the molecular content a molecule’s graphical depiction is an extremely challenging problem that remains largely unsolved despite decades research. Recent advances in neural machine translation enable auto-encoding structures continuous vector space fixed size (latent representation) with low reconstruction errors. In this paper, we present fast and accurate model combining deep convolutional network learning from molecule depictions pre-trained decoder translates...
The COVID-19 pandemic continues to pose a substantial threat human lives and is likely do so for years come. Despite the availability of vaccines, searching efficient small-molecule drugs that are widely available, including in low- middle-income countries, an ongoing challenge. In this work, we report results community effort, “Billion molecules against Covid-19 challenge”, identify inhibitors SARS-CoV-2 or relevant receptors. Participating teams used wide variety computational methods...
In this work, we propose a novel method that combines in silico prediction of molecular properties such as biological activity or pharmacokinetics with an optimization algorithm, namely Particle Swarm Optimization. Our takes starting compound input and proposes new molecules more desirable (predicted) properties. It navigates machine-learned continuous representation drug-like chemical space guided by de�fined objective function. The function multiple models, desirability ranges substructure...
Generative adversarial networks (GANs) are a powerful framework for generative tasks. However, they difficult to train and tend miss modes of the true data generation process. Although GANs can learn rich representation covered in their latent space, misses an inverse mapping from this space. We propose Invariant Encoding Adversarial Networks (IVE-GANs), novel GAN that introduces such individual samples by utilizing features which invariant certain transformations. Since model maps it...
Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework collaborations, it common to exchange datasets by encoding into descriptors. Molecular fingerprints extended-connectivity are frequently used an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs often considered be non-invertible due way...
An essential part of computer-aided drug development is the prediction molecular properties for a specific chemical structure (e.g. interaction compound with protein) or design molecules desired properties. Molecular descriptors play crucial role building predictive models, since they allow representing information in computer-interpretable vector. In this talk, we present (i) neural machine translation approach learning robust and continuous feature representation structures, (ii) an...