A5060076719- Protein Structure and Dynamics
- Computational Drug Discovery Methods
- Monoclonal and Polyclonal Antibodies Research
- Machine Learning and Data Classification
- Machine Learning in Bioinformatics
- Infrastructure Maintenance and Monitoring
- Sepsis Diagnosis and Treatment
- Traffic Prediction and Management Techniques
- Machine Learning in Materials Science
- Clinical Reasoning and Diagnostic Skills
- Music and Audio Processing
- Enzyme Structure and Function
- Machine Learning in Healthcare
- Time Series Analysis and Forecasting
- Generative Adversarial Networks and Image Synthesis
University College London
2022-2024
Abstract Early detection of sepsis is key to ensure timely clinical intervention. Since very few end-to-end pipelines are publicly available, fair comparisons between methodologies difficult if not impossible. Progress further limited by discrepancies in the reconstruction onset time. This retrospective cohort study highlights variation performance predictive models under three subtly different interpretations from sepsis-III definition and compares this against inter-model differences. The...
The goal of Protein Structure Prediction (PSP) problem is to predict a protein's 3D structure (confirmation) from its amino acid sequence. has been 'holy grail' science since the Noble prize-winning work Anfinsen demonstrated that protein conformation was determined by A recent and important step towards this development AlphaFold2, currently best PSP method. AlphaFold2 probably highest profile application AI science. Both RoseTTAFold (another impressive method) have published placed in...
The representation of the protein-ligand complexes used in building machine learning models play an important role accuracy binding affinity prediction. Extended Connectivity Interaction Features (ECIF) is one such representation. We report that (i) including discretized distances between atom pairs ECIF scheme improves predictive accuracy, and (ii) evaluation using gradient boosted trees, we found resampling method selecting best hyperparameters has a strong effect on performance,...
Generating high-fidelity time series data using generative adversarial networks (GANs) remains a challenging task, as it is difficult to capture the temporal dependence of joint probability distributions induced by time-series data. Towards this goal, key step development an effective discriminator distinguish between distributions. We propose so-called PCF-GAN, novel GAN that incorporates path characteristic function (PCF) principled representation distribution into enhance its performance....
The path signature, a mathematically principled and universal feature of sequential data, leads to performance boost deep learning-based models in various data tasks as complimentary feature. However, it suffers from the curse dimensionality when dimension is high. To tackle this problem, we propose novel, trainable development layer, which exploits representations with help finite-dimensional matrix Lie groups. We also design backpropagation algorithm layer via an optimisation method on...