Can machine learning predict late seizures after intracerebral hemorrhages? Evidence from real-world data.

INTRODUCTION: Intracerebral hemorrhage represents 15 % of all strokes and it is associated with a high risk of post-stroke epilepsy. However, there are no reliable methods to accurately predict those at higher risk for developing seizures despite their importance in planning treatments, allocating resources, and advancing post-stroke seizure research. Existing risk models have limitations and have not taken advantage of readily available real-world data and artificial intelligence. This study aims to evaluate the performance of Machine-learning-based models to predict post-stroke seizures at 1 year and 5 years after an intracerebral hemorrhage in unselected patients across multiple healthcare organizations. DESIGN/METHODS: We identified patients with intracerebral hemorrhage (ICH) without a prior diagnosis of seizures from 2015 until inception (11/01/22) in the TriNetX Diamond Network, using the International Classification of Diseases, Tenth Revision (ICD-10) I61 (I61.0, I61.1, I61.2, I61.3, I61.4, I61.5, I61.6, I61.8, and I61.9). The outcome of interest was any ICD-10 diagnosis of seizures (G40/G41) at 1 year and 5 years following the first occurrence of the diagnosis of intracerebral hemorrhage. We applied a conventional logistic regression and a Light Gradient Boosted Machine (LGBM) algorithm, and the performance of the model was assessed using the area under the receiver operating characteristics (AUROC), the area under the precision-recall curve (AUPRC), the F1 statistic, model accuracy, balanced-accuracy, precision, and recall, with and without seizure medication use in the models. RESULTS: A total of 85,679 patients had an ICD-10 code of intracerebral hemorrhage and no prior diagnosis of seizures, constituting our study cohort. Seizures were present in 4.57 % and 6.27 % of patients within 1 and 5 years after ICH, respectively. At 1-year, the AUROC, AUPRC, F1 statistic, accuracy, balanced-accuracy, precision, and recall were respectively 0.7051 (standard error: 0.0132), 0.1143 (0.0068), 0.1479 (0.0055), 0.6708 (0.0076), 0.6491 (0.0114), 0.0839 (0.0032), and 0.6253 (0.0216). Corresponding metrics at 5 years were 0.694 (0.009), 0.1431 (0.0039), 0.1859 (0.0064), 0.6603 (0.0059), 0.6408 (0.0119), 0.1094 (0.0037) and 0.6186 (0.0264). These numerical values indicate that the statistical models fit the data very well. CONCLUSION: Machine learning models applied to electronic health records can improve the prediction of post-hemorrhagic stroke epilepsy, presenting a real opportunity to incorporate risk assessments into clinical decision-making in post-stroke care clinical care and improve patients' selection for post-stroke epilepsy research.

Large datasets from Electronic Health Records predict seizures after ischemic strokes: A Machine Learning approach.

Objective: To develop an artificial intelligence, machine learning prediction model for estimating the risk of seizures 1 year and 5 years after ischemic stroke (IS) using a large dataset from Electronic Health Records. Background: Seizures are frequent after ischemic strokes and are associated with increased mortality, poor functional outcomes, and lower quality of life. Separating patients at high risk of seizures from those at low risk of seizures is needed for treatment and clinical trial planning, but remains challenging. Machine learning (ML) is a potential approach to solve this paradigm. Design/Methods: We identified patients (aged ≥18 years) with IS without a prior diagnosis of seizures from 2015 until inception (08/09/22) in the TriNetX Research Network, using the International Classification of Diseases, Tenth Revision (ICD-10) I63, excluding I63.6 (venous infarction). The outcome of interest was any ICD-10 diagnosis of seizures (G40/G41) at 1 year and 5 years following the index IS. We applied a conventional logistic regression and a Light Gradient Boosted Machine algorithm to predict the risk of seizures at 1 year and 5 years. The performance of the model was assessed using the area under the receiver operating characteristics (AUROC), the area under the precision-recall curve (AUPRC), F1 statistic, model accuracy, balanced accuracy, precision, and recall, with and without anti-seizure medication use in the models. Results: Our study cohort included 430,254 IS patients. Seizures were present in 18,502 (4.3%) and (5.3%) patients within 1 and 5 years after IS, respectively. At 1-year, the AUROC, AUPRC, F1 statistic, accuracy, balanced-accuracy, precision, and recall were respectively 0.7854 (standard error: 0.0038), 0.2426 (0.0048), 0.2299 (0.0034), 0.8236 (0.001), 0.7226 (0.0049), 0.1415 (0.0021), and 0.6122, (0.0095). Corresponding metrics at 5 years were 0.7607 (0.0031), 0.247 (0.0064), 0.2441 (0.0032), 0.8125 (0.0013), 0.7001 (0.0045), 0.155 (0.002) and 0.5745 (0.0095). Conclusion: Our findings suggest that ML models show good model performance for predicting seizures after IS.

Non-covalent Lasso Entanglements in Folded Proteins: Prevalence, Functional Implications, and Evolutionary Significance.

One-third of protein domains in the CATH database contain a recently discovered tertiary topological motif: non-covalent lasso entanglements, in which a segment of the protein backbone forms a loop closed by non-covalent interactions between residues and is threaded one or more times by the N- or C-terminal backbone segment. Unknown is how frequently this structural motif appears across the proteomes of organisms. And the correlation of these motifs with various classes of protein function and biological processes have not been quantified. Here, using a combination of protein crystal structures, AlphaFold2 predictions, and Gene Ontology terms we show that in E. coli, S. cerevisiae and H. sapiens that 71%, 52% and 49% of globular proteins contain one-or-more non-covalent lasso entanglements in their native fold, and that some of these are highly complex with multiple threading events. Further, proteins containing these tertiary motifs are consistently enriched in certain functions and biological processes across these organisms and depleted in others, strongly indicating an influence of evolutionary selection pressures acting positively and negatively on the distribution of these motifs. Together, these results demonstrate that non-covalent lasso entanglements are widespread and indicate they may be extensively utilized for protein function and subcellular processes, thus impacting phenotype.

Ethane adsorption on aggregates of dahlia-like nanohorns: experiments and computer simulations.

This is a report on a study of the adsorption characteristics of ethane on aggregates of unopened dahlia-like carbon nanohorns. This sorbent presents two main groups of adsorption sites: the outside surface of individual nanohorns and deep, interstitial spaces between neighbouring nanohorns towards the interior of the aggregates. We have explored the equilibrium properties of the adsorbed ethane films by determining the adsorption isotherms and isosteric heat of adsorption. Computer simulations performed on different model structures indicate that the majority of ethane adsorption occurs on the outer region of the aggregates, near the ends of the nanohorns. We have also measured the kinetics of adsorption of ethane on this sorbent. The measurements and simulations were conducted along several isotherms spanning the range between 120 K and 220 K.

Diffusion, adsorption, and desorption of molecular hydrogen on graphene and in graphite.

The diffusion of molecular hydrogen (H2) on a layer of graphene and in the interlayer space between the layers of graphite is studied using molecular dynamics computer simulations. The interatomic interactions were modeled by an Adaptive Intermolecular Reactive Empirical Bond Order (AIREBO) potential. Molecular statics calculations of H2 on graphene indicate binding energies ranging from 41 meV to 54 meV and migration barriers ranging from 3 meV to 12 meV. The potential energy surface of an H2 molecule on graphene, with the full relaxations of molecular hydrogen and carbon atoms is calculated. Barriers for the formation of H2 through the Langmuir-Hinshelwood mechanism are calculated. Molecular dynamics calculations of mean square displacements and average surface lifetimes of H2 on graphene at various temperatures indicate a diffusion barrier of 9.8 meV and a desorption barrier of 28.7 meV. Similar calculations for the diffusion of H2 in the interlayer space between the graphite sheets indicate high and low temperature regimes for the diffusion with barriers of 51.2 meV and 11.5 meV. Our results are compared with those of first principles.