Descriptor-augmented machine learning for enzyme-chemical interaction predictions
Ontology highlight
ABSTRACT:
SUBMITTER: Han Y
PROVIDER: S-EPMC10915406 | biostudies-literature | 2024 Feb
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:The QM9 dataset has become the golden standard for Machine Learning (ML) predictions of various chemical properties. QM9 is based on the GDB, which is a combinatorial exploration of the chemical space. ML molecular predictions have been recently published with an accuracy on par with Density Functional Theory calculations. Such ML models need to be tested and generalized on real data. PC9, a new QM9 equivalent dataset (only H, C, N, O and F and up to 9 "heavy" atoms) of the PubChemQC project is presented in this article. A statistical study of bonding distances and chemical functions shows that this new dataset encompasses more chemical diversity. Kernel Ridge Regression, Elastic Net and the Neural Network model provided by SchNet have been used on both datasets. The overall accuracy in energy prediction is higher for the QM9 subset. However, a model trained on PC9 shows a stronger ability to predict energies of the other dataset.
| S-EPMC6852905 | biostudies-literature
Project description:In the quest for advanced superconducting materials, the accurate prediction of critical temperatures (Tc) poses a formidable challenge, largely due to the complex interdependencies between superconducting properties and the chemical and structural characteristics of a given material. To address this challenges, we have developed a machine-learning framework that aims to elucidate these complicated and hitherto poorly understood structure-property and property-property relationships. This study introduces a novel machine-learning-based workflow, termed the Gradient Boosted Feature Selection (GBFS), which has been tailored to predict Tc for superconductors by employing a distributed gradient-boosting framework. This approach integrates exploratory data analyses, statistical evaluations, and multicollinearity reduction techniques to select highly relevant features from a high-dimensional feature space, derived solely from the chemical composition of materials. Our methodology was rigorously tested on a data set comprising approximately 16,400 chemical compounds with around 12,000 unique chemical compositions. The GBFS workflow enabled the development of a classification model that distinguishes compositions likely to exhibit Tc values greater than 10 K. This model achieved a weighted average F1-score of 0.912, an AUC-ROC of 0.986, and an average precision score of 0.919. Additionally, the GBFS workflow underpinned a regression model that predicted Tc values with an R2 of 0.945, an MAE of 3.54 K, and an RMSE of 6.57 K on a test set obtained via random splitting. Further exploration was conducted through out-of-sample Tc predictions, particularly those exceeding the liquid nitrogen temperature, and out-of-distribution predictions for (Ca1-xLax)FeAs2 based on varying lanthanum content. The outcome of our study underscores the significance of systematic feature analysis and selection in enhancing predictive model performance, offering various advantages over models that rely primarily on algorithmic complexity. This research not only advances the field of superconductivity but also sets a precedent for the application of machine learning in materials science.
| S-EPMC11481088 | biostudies-literature
Project description:The theory and computational tools developed to interpret and explore energy landscapes in molecular science are applied to the landscapes defined by local minima for neural networks. These machine learning landscapes correspond to fits of training data, where the inputs are vital signs and laboratory measurements for a database of patients, and the objective is to predict a clinical outcome. In this contribution, we test the predictions obtained by fitting to single measurements, and then to combinations of between 2 and 10 different patient medical data items. The effect of including measurements over different time intervals from the 48 h period in question is analysed, and the most recent values are found to be the most important. We also compare results obtained for neural networks as a function of the number of hidden nodes, and for different values of a regularization parameter. The predictions are compared with an alternative convex fitting function, and a strong correlation is observed. The dependence of these results on the patients randomly selected for training and testing decreases systematically with the size of the database available. The machine learning landscapes defined by neural network fits in this investigation have single-funnel character, which probably explains why it is relatively straightforward to obtain the global minimum solution, or a fit that behaves similarly to this optimal parameterization.
| S-EPMC5541539 | biostudies-other
Project description:Despite tremendous progress in understanding and engineering enzymes, knowledge of how enzyme structures and their dynamics induce observed catalytic properties is incomplete, and capabilities to engineer enzymes fall far short of industrial needs. Here, we investigate the structural and dynamic drivers of enzyme catalysis for the rate-limiting step of the industrially important enzyme ketol-acid reductoisomerase (KARI) and identify a region of the conformational space of the bound enzyme-substrate complex that, when populated, leads to large increases in reactivity. We apply computational statistical mechanical methods that implement transition interface sampling to simulate the kinetics of the reaction and combine this with machine learning techniques from artificial intelligence to select features relevant to reactivity and to build predictive models for reactive trajectories. We find that conformational descriptors alone, without the need for dynamic ones, are sufficient to predict reactivity with greater than 85% accuracy (90% AUC). Key descriptors distinguishing reactive from almost-reactive trajectories quantify substrate conformation, substrate bond polarization, and metal coordination geometry and suggest their role in promoting substrate reactivity. Moreover, trajectories constrained to visit a region of the reactant well, separated from the rest by a simple hyperplane defined by ten conformational parameters, show increases in computed reactivity by many orders of magnitude. This study provides evidence for the existence of reactivity promoting regions within the conformational space of the enzyme-substrate complex and develops methodology for identifying and validating these particularly reactive regions of phase space. We suggest that identification of reactivity promoting regions and re-engineering enzymes to preferentially populate them may lead to significant rate enhancements.
| S-EPMC6407039 | biostudies-literature
Project description:Machine learning is proving invaluable across disciplines. However, its success is often limited by the quality and quantity of available data, while its adoption is limited by the level of trust afforded by given models. Human vs. machine performance is commonly compared empirically to decide whether a certain task should be performed by a computer or an expert. In reality, the optimal learning strategy may involve combining the complementary strengths of humans and machines. Here, we present expert-augmented machine learning (EAML), an automated method that guides the extraction of expert knowledge and its integration into machine-learned models. We used a large dataset of intensive-care patient data to derive 126 decision rules that predict hospital mortality. Using an online platform, we asked 15 clinicians to assess the relative risk of the subpopulation defined by each rule compared to the total sample. We compared the clinician-assessed risk to the empirical risk and found that, while clinicians agreed with the data in most cases, there were notable exceptions where they overestimated or underestimated the true risk. Studying the rules with greatest disagreement, we identified problems with the training data, including one miscoded variable and one hidden confounder. Filtering the rules based on the extent of disagreement between clinician-assessed risk and empirical risk, we improved performance on out-of-sample data and were able to train with less data. EAML provides a platform for automated creation of problem-specific priors, which help build robust and dependable machine-learning models in critical applications.
| S-EPMC7060733 | biostudies-literature
Project description:There are continuous efforts to elucidate the structure and biological functions of short hydrogen bonds (SHBs), whose donor and acceptor heteroatoms reside more than 0.3 A closer than the sum of their van der Waals radii. In this work, we evaluate 1070 atomic-resolution protein structures and characterize the common chemical features of SHBs formed between the side chains of amino acids and small molecule ligands. We then develop a machine learning assisted prediction of protein-ligand SHBs (MAPSHB-Ligand) model and reveal that the types of amino acids and ligand functional groups as well as the sequence of neighboring residues are essential factors that determine the class of protein-ligand hydrogen bonds. The MAPSHB-Ligand model and its implementation on our web server enable the effective identification of protein-ligand SHBs in proteins, which will facilitate the design of biomolecules and ligands that exploit these close contacts for enhanced functions.
| S-EPMC10246099 | biostudies-literature
Project description:There are continuous efforts to elucidate the structure and biological functions of short hydrogen bonds (SHBs), whose donor and acceptor heteroatoms reside more than 0.3 Å closer than the sum of their van der Waals radii. In this work, we evaluate 1070 atomic-resolution protein structures and characterize the common chemical features of SHBs formed between the side chains of amino acids and small molecule ligands. We then develop a machine learning assisted prediction of protein-ligand SHBs (MAPSHB-Ligand) model and reveal that the types of amino acids and ligand functional groups as well as the sequence of neighboring residues are essential factors that determine the class of protein-ligand hydrogen bonds. The MAPSHB-Ligand model and its implementation on our web server enable the effective identification of protein-ligand SHBs in proteins, which will facilitate the design of biomolecules and ligands that exploit these close contacts for enhanced functions.
| S-EPMC10447522 | biostudies-literature
Project description:Protein-protein interactions (PPIs) may represent one of the next major classes of therapeutic targets. So far, only a minute fraction of the estimated 650,000 PPIs that comprise the human interactome are known with a tiny number of complexes being drugged. Such intricate biological systems cannot be cost-efficiently tackled using conventional high-throughput screening methods. Rather, time has come for designing new strategies that will maximize the chance for hit identification through a rationalization of the PPI inhibitor chemical space and the design of PPI-focused compound libraries (global or target-specific). Here, we train machine-learning-based models, mainly decision trees, using a dataset of known PPI inhibitors and of regular drugs in order to determine a global physico-chemical profile for putative PPI inhibitors. This statistical analysis unravels two important molecular descriptors for PPI inhibitors characterizing specific molecular shapes and the presence of a privileged number of aromatic bonds. The best model has been transposed into a computer program, PPI-HitProfiler, that can output from any drug-like compound collection a focused chemical library enriched in putative PPI inhibitors. Our PPI inhibitor profiler is challenged on the experimental screening results of 11 different PPIs among which the p53/MDM2 interaction screened within our own CDithem platform, that in addition to the validation of our concept led to the identification of 4 novel p53/MDM2 inhibitors. Collectively, our tool shows a robust behavior on the 11 experimental datasets by correctly profiling 70% of the experimentally identified hits while removing 52% of the inactive compounds from the initial compound collections. We strongly believe that this new tool can be used as a global PPI inhibitor profiler prior to screening assays to reduce the size of the compound collections to be experimentally screened while keeping most of the true PPI inhibitors. PPI-HitProfiler is freely available on request from our CDithem platform website, www.CDithem.com.
| S-EPMC2832677 | biostudies-other
Project description:Research in physiology and sports science has shown that fatigue, a complex psychophysiological phenomenon, has a relevant impact in performance and in the correct functioning of our motricity system, potentially being a cause of damage to the human organism. Fatigue can be seen as a subjective or objective phenomenon. Subjective fatigue corresponds to a mental and cognitive event, while fatigue referred as objective is a physical phenomenon. Despite the fact that subjective fatigue is often undervalued, only a physically and mentally healthy athlete is able to achieve top performance in a discipline. Therefore, we argue that physical training programs should address the preventive assessment of both subjective and objective fatigue mechanisms in order to minimize the risk of injuries. In this context, our paper presents a machine-learning system capable of extracting individual fatigue descriptors (IFDs) from electromyographic (EMG) and heart rate variability (HRV) measurements. Our novel approach, using two types of biosignals so that a global (mental and physical) fatigue assessment is taken into account, reflects the onset of fatigue by implementing a combination of a dimensionless (0-1) global fatigue descriptor (GFD) and a support vector machine (SVM) classifier. The system, based on 9 main combined features, achieves fatigue regime classification performances of 0.82 ± 0.24, ensuring a successful preventive assessment when dangerous fatigue levels are reached. Training data were acquired in a constant work rate test (executed by 14 subjects using a cycloergometry device), where the variable under study (fatigue) gradually increased until the volunteer reached an objective exhaustion state.
| S-EPMC6969995 | biostudies-literature
Project description:Machine learning models have emerged as powerful tools for drug discovery of lead compounds. Nevertheless, despite notable advances in model architectures, research on more reliable and physicochemical-based descriptors for molecules and proteins remains limited. To address this gap, we introduce the Fragment Integral Spectrum Descriptor (FISD), aimed at utilizing the spatial configuration and electronic structure information of molecules and proteins, as a novel physicochemical descriptor for virtual screening models. Validation demonstrates that the combination of FISD and a classical neural network model achieves performance comparable to that of complex models paired with conventional structural descriptors. Furthermore, we successfully predict and screen potential binding ligands for two given protein targets, showcasing the broad applicability and practicality of FISD. This research enriches the molecular and protein representation strategies of machine learning and accelerates the process of drug discovery.
| S-EPMC11905448 | biostudies-literature