Evaluation of crowdsourced mortality prediction models as a framework for assessing artificial intelligence in medicine.

OBJECTIVE: Applications of machine learning in healthcare are of high interest and have the potential to improve patient care. Yet, the real-world accuracy of these models in clinical practice and on different patient subpopulations remains unclear. To address these important questions, we hosted a community challenge to evaluate methods that predict healthcare outcomes. We focused on the prediction of all-cause mortality as the community challenge question. MATERIALS AND METHODS: Using a Model-to-Data framework, 345 registered participants, coalescing into 25 independent teams, spread over 3 continents and 10 countries, generated 25 accurate models all trained on a dataset of over 1.1 million patients and evaluated on patients prospectively collected over a 1-year observation of a large health system. RESULTS: The top performing team achieved a final area under the receiver operator curve of 0.947 (95% CI, 0.942-0.951) and an area under the precision-recall curve of 0.487 (95% CI, 0.458-0.499) on a prospectively collected patient cohort. DISCUSSION: Post hoc analysis after the challenge revealed that models differ in accuracy on subpopulations, delineated by race or gender, even when they are trained on the same data. CONCLUSION: This is the largest community challenge focused on the evaluation of state-of-the-art machine learning methods in a healthcare system performed to date, revealing both opportunities and pitfalls of clinical AI.

Exploring Covalent Docking Mechanisms of Boron-Based Inhibitors to Class A, C and D ß-Lactamases Using Time-dependent Hybrid QM/MM Simulations.

Recently, molecular covalent docking has been extensively developed to design new classes of inhibitors that form chemical bonds with their biological targets. This strategy for the design of such inhibitors, in particular boron-based inhibitors, holds great promise for the vast family of ß-lactamases produced, inter alia, by Gram-negative antibiotic-resistant bacteria. However, the description of covalent docking processes requires a quantum-mechanical approach, and so far, only a few studies of this type have been presented. This study accurately describes the covalent docking process between two model inhibitors - representing two large families of inhibitors based on boronic-acid and bicyclic boronate scaffolds, and three ß-lactamases which belong to the A, C, and D classes. Molecular fragments containing boron can be converted from a neutral, trigonal, planar state with sp2 hybridization to the anionic, tetrahedral sp3 state in a process sometimes referred to as morphing. This study applies multi-scale modeling methods, in particular, the hybrid QM/MM approach which has predictive power reaching well beyond conventional molecular modeling. Time-dependent QM/MM simulations indicated several structural changes and geometric preferences, ultimately leading to covalent docking processes. With current computing technologies, this approach is not computationally expensive, can be used in standard molecular modeling and molecular design works, and can effectively support experimental research which should allow for a detailed understanding of complex processes important to molecular medicine. In particular, it can support the rational design of covalent boron-based inhibitors for ß-lactamases as well as for many other enzyme systems of clinical relevance, including SARS-CoV-2 proteins.

Structural characterisation of inhibitory and non-inhibitory MMP-9-TIMP-1 complexes and implications for regulatory mechanisms of MMP-9.

MMP-9 plays a number of important physiological functions but is also responsible for many pathological processes, including cancer invasion, metastasis, and angiogenesis. It is, therefore, crucial to understand its enzymatic activity, including activation and inhibition mechanisms. This enzyme may also be partially involved in the "cytokine storm" that is characteristic of COVID-19 disease (SARS-CoV-2), as well as in the molecular mechanisms responsible for lung fibrosis. Due to the variety of processing pathways involving MMP-9 in biological systems and its uniqueness due to the O-glycosylated domain (OGD) and fibronectin-like (FBN) domain, specific interactions with its natural TIMP-1 inhibitor should be carefully studied, because they differ significantly from other homologous systems. In particular, earlier experimental studies have indicated that the newly characterised circular form of a proMMP-9 homotrimer exhibits stronger binding properties to TIMP-1 compared to its monomeric form. However, molecular structures of the complexes and the binding mechanisms remain unknown. The purpose of this study is to fill in the gaps in knowledge. Molecular modelling methods are applied to build the inhibitory and non-inhibitory MMP-9-TIMP-1 complexes, which allows for a detailed description of these structures and should allow for a better understanding of the regulatory processes in which MMP-9 is involved.

Chronic fluoxetine treatment impairs motivation and reward learning by affecting neuronal plasticity in the central amygdala.

BACKGROUND AND PURPOSE: The therapeutic effects of fluoxetine are believed to be due to increasing neuronal plasticity and reversing some learning deficits. Nevertheless, a growing amount of evidence shows adverse effects of this drug on cognition and some forms of neuronal plasticity. EXPERIMENTAL APPROACH: To study the effects of chronic fluoxetine treatment, we combine an automated assessment of motivation and learning in mice with an investigation of neuronal plasticity in the central amygdala and basolateral amygdala. We use immunohistochemistry to visualize neuronal types and perineuronal nets, along with DI staining to assess dendritic spine morphology. Gel zymography is used to test fluoxetine's impact on matrix metalloproteinase-9, an enzyme involved in synaptic plasticity. KEY RESULTS: We show that chronic fluoxetine treatment in non-stressed mice increases perineuronal nets-dependent plasticity in the basolateral amygdala, while impairing MMP-9-dependent plasticity in the central amygdala. Further, we illustrate how the latter contributes to anhedonia and deficits of reward learning. Behavioural impairments are accompanied by alterations in morphology of dendritic spines in the central amygdala towards an immature state, most likely reflecting animals' inability to adapt. We strengthen the link between the adverse effects of fluoxetine and its influence on MMP-9 by showing that behaviour of MMP-9 knockout animals remains unaffected by the drug. CONCLUSION AND IMPLICATIONS: Chronic fluoxetine treatment differentially affects various forms of neuronal plasticity, possibly explaining its opposing effects on brain and behaviour. These findings are of immediate clinical relevance since reported side effects of fluoxetine pose a potential threat to patients.

Alterations in 5hmC level and genomic distribution in aging-related epigenetic drift in human adipose stem cells.

Aim: To clarify mechanisms affecting the level and distribution of 5-hydroxymethylcytosine (5hmC) during aging. Materials & methods: We examined levels and genomic distribution of 5hmC along with the expression of ten-eleven translocation methylcytosine dioxygenases (TETs) in adipose stem cells in young and age-advanced individuals. Results: 5hmC levels were higher in adipose stem cells of age-advanced than young individuals (p = 0.0003), but were not associated with age-related changes in expression of TETs. 5hmC levels correlated with population doubling time (r = 0.62; p = 0.01). We identified 58 differentially hydroxymethylated regions. Hypo-hydroxymethylated differentially hydroxymethylated regions were approximately twofold enriched in CCCTC-binding factor binding sites. Conclusion: Accumulation of 5hmC in aged cells can result from inefficient active demethylation due to altered TETs activity and reduced passive demethylation due to slower proliferation.

A Molecular Dynamics Study of Vasoactive Intestinal Peptide Receptor 1 and the Basis of Its Therapeutic Antagonism.

Vasoactive intestinal peptide receptor 1 (VPAC1) is a member of a secretin-like subfamily of G protein-coupled receptors. Its endogenous neuropeptide (VIP), secreted by neurons and immune cells, modulates various physiological functions such as exocrine and endocrine secretions, immune response, smooth muscles relaxation, vasodilation, and fetal development. As a drug target, VPAC1 has been selected for therapy of inflammatory diseases but drug discovery is still hampered by lack of its crystal structure. In this study we presented the homology model of this receptor constructed with the well-known web service GPCRM. The VPAC1 model is composed of extracellular and transmembrane domains that form a complex with an endogenous hormone VIP. Using the homology model of VPAC1 the mechanism of action of potential drug candidates for VPAC1 was described. Only two series of small-molecule antagonists of confirmed biological activity for VPAC1 have been described thus far. Molecular docking and a series of molecular dynamics simulations were performed to elucidate their binding to VPAC1 and resulting antagonist effect. The presented work provides the basis for the possible binding mode of VPAC1 antagonists and determinants of their molecular recognition in the context of other class B GPCRs. Until the crystal structure of VPAC1 will be released, the presented homology model of VPAC1 can serve as a scaffold for drug discovery studies and is available from the author upon request.

Antimicrobial and KPC/AmpC inhibitory activity of functionalized benzosiloxaboroles.

A series of 22 benzosiloxaboroles, silicon analogues of strong antimicrobial agents - benzoxaboroles, have been synthesized and tested against ß-lactamases KPC- and pAmpC-producing strains of Gram-negative rods. Comprehensive structural-property relationship studies supported by molecular modelling as well as biological studies reveal that 6-B(OH)2-substituted derivative 27 strongly inhibits the activity of cephalosporinases (chromosomally encoded AmpC and plasmid encoded CMY-2) and KPC carbapenemases. It also shows strong ability to inhibit growth of the strains producing KPC-3 when combined with meropenem. In addition, halogen-substituted (mono-, di- or tetra-) benzosiloxaboroles demonstrate high antifungal activity (MIC 1.56-6.25 mg/L) against C. tropicalis, C. guilliermondii and S. cerevisiae. The highest activity against pathogenic yeasts (C. albicans, C. krusei and C. parapsilosis - MICs 12.5 mg/L) and against Gram-positive cocci (S. aureus and E. faecalis - 6.25 mg/L and 25 mg/L respectively) was displayed by 6,7-dichloro-substituted benzosiloxaborole. The studied systems exhibit low cytotoxity toward human lung fibroblasts.

Molecular modelling of mitofusin 2 for a prediction for Charcot-Marie-Tooth 2A clinical severity.

Charcot-Marie-Tooth disease type 2A (CMT2A) is an autosomal dominant neuropathy caused by mutations in the mitofusin 2 gene (MFN2). More than 100 MFN2 gene mutations have been reported so far, with majority located within the GTPase domain encoding region. These domain-specific mutations present wide range of symptoms with differences associated with distinct amino acid substitutions in the same position. Due to the lack of conclusive phenotype-genotype correlation the predictive value of genetic results remains still limited. We have explored whether changes in the protein structure caused by MFN2 mutations can help to explain diseases phenotypes. Using a stable protein model, we evaluated the effect of 26 substitutions on the MFN2 structure and predicted the molecular consequences of such alterations. The observed changes were correlated with clinical features associated with a given mutation. Of all tested mutations positive correlation of molecular modelling with the clinical features reached 73%. Our analysis revealed that molecular modelling of mitofusin 2 mutations is a powerful tool, which predicts associated pathogenic impacts and that these correlate with clinical outcomes. This approach may aid an early diagnosis and prediction of symptoms severity in CMT2A patients.

The Effect of a Novel c.820C>T (Arg274Trp) Mutation in the Mitofusin 2 Gene on Fibroblast Metabolism and Clinical Manifestation in a Patient.

Charcot-Marie-Tooth disease type 2A (CMT2A) is an autosomal dominant axonal peripheral neuropathy caused by mutations in the mitofusin 2 gene (MFN2). Mitofusin 2 is a GTPase protein present in the outer mitochondrial membrane and responsible for regulation of mitochondrial network architecture via the fusion of mitochondria. As that fusion process is known to be strongly dependent on the GTPase activity of mitofusin 2, it is postulated that the MFN2 mutation within the GTPase domain may lead to impaired GTPase activity, and in turn to mitochondrial dysfunction. The work described here has therefore sought to verify the effects of MFN2 mutation within its GTPase domain on mitochondrial and endoplasmic reticulum morphology, as well as the mtDNA content in a cultured primary fibroblast obtained from a CMT2A patient harboring a de novo Arg274Trp mutation. In fact, all the parameters studied were affected significantly by the presence of the mutant MFN2 protein. However, using the stable model for mitofusin 2 obtained by us, we were next able to determine that the Arg274Trp mutation does not impact directly upon GTP binding. Such results were also confirmed for GTP-hydrolysis activity of MFN2 protein in patient fibroblast. We therefore suggest that the biological malfunctions observable with the disease are not consequences of impaired GTPase activity, but rather reflect an impaired contribution of the GTPase domain to other MFN2 activities involving that region, for example protein-protein interactions.