Application of unsupervised deep learning algorithms for identification of specific clusters of chronic cough patients from EMR data.

BACKGROUND: Chronic cough affects approximately 10% of adults. The lack of ICD codes for chronic cough makes it challenging to apply supervised learning methods to predict the characteristics of chronic cough patients, thereby requiring the identification of chronic cough patients by other mechanisms. We developed a deep clustering algorithm with auto-encoder embedding (DCAE) to identify clusters of chronic cough patients based on data from a large cohort of 264,146 patients from the Electronic Medical Records (EMR) system. We constructed features using the diagnosis within the EMR, then built a clustering-oriented loss function directly on embedded features of the deep autoencoder to jointly perform feature refinement and cluster assignment. Lastly, we performed statistical analysis on the identified clusters to characterize the chronic cough patients compared to the non-chronic cough patients. RESULTS: The experimental results show that the DCAE model generated three chronic cough clusters and one non-chronic cough patient cluster. We found various diagnoses, medications, and lab tests highly associated with chronic cough patients by comparing the chronic cough cluster with the non-chronic cough cluster. Comparison of chronic cough clusters demonstrated that certain combinations of medications and diagnoses characterize some chronic cough clusters. CONCLUSIONS: To the best of our knowledge, this study is the first to test the potential of unsupervised deep learning methods for chronic cough investigation, which also shows a great advantage over existing algorithms for patient data clustering.

Applying interpretable deep learning models to identify chronic cough patients using EHR data.

BACKGROUND AND OBJECTIVE: Chronic cough (CC) affects approximately 10% of adults. Many disease states are associated with chronic cough, such as asthma, upper airway cough syndrome, bronchitis, and gastroesophageal reflux disease. The lack of an ICD code specific for chronic cough makes it challenging to identify such patients from electronic health records (EHRs). For clinical and research purposes, computational methods using EHR data are urgently needed to identify chronic cough cases. This research aims to investigate the data representations and deep learning algorithms for chronic cough prediction. METHODS: Utilizing real-world EHR data from a large academic healthcare system from October 2005 to September 2015, we investigated Natural Language Representation of the EHR data and systematically evaluated deep learning and traditional machine learning models to predict chronic cough patients. We built these machine learning models using structured data (medication and diagnosis) and unstructured data (clinical notes). RESULTS: The sensitivity and specificity of a transformer-based deep learning algorithm, specifically BERT with attention model, was 0.856 and 0.866, respectively, using structured data (medication and diagnosis). Sensitivity and specificity improved to 0.952 and 0.930 when we combined structured data with symptoms extracted from clinical notes. We further found that the attention mechanism of deep learning models can be used to extract important features that drive the prediction decisions. Compared with our previously published rule-based algorithm, the deep learning algorithm can identify more chronic cough patients with structured data. CONCLUSIONS: By applying deep learning models, chronic cough patients can be reliably identified for prospective or retrospective research through medication and diagnosis data, widely available in EHR and electronic claims data, thus improving the generalizability of the patient identification algorithm. Deep learning models can identify chronic cough patients with even higher sensitivity and specificity when structured and unstructured EHR data are utilized. We anticipate language-based data representation and deep learning models developed in this research could also be productively used for other disease prediction and case identification.

Predicting dementia with routine care EMR data.

Our aim is to develop a machine learning (ML) model that can predict dementia in a general patient population from multiple health care institutions one year and three years prior to the onset of the disease without any additional monitoring or screening. The purpose of the model is to automate the cost-effective, non-invasive, digital pre-screening of patients at risk for dementia. Towards this purpose, routine care data, which is widely available through Electronic Medical Record (EMR) systems is used as a data source. These data embody a rich knowledge and make related medical applications easy to deploy at scale in a cost-effective manner. Specifically, the model is trained by using structured and unstructured data from three EMR data sets: diagnosis, prescriptions, and medical notes. Each of these three data sets is used to construct an individual model along with a combined model which is derived by using all three data sets. Human-interpretable data processing and ML techniques are selected in order to facilitate adoption of the proposed model by health care providers from multiple institutions. The results show that the combined model is generalizable across multiple institutions and is able to predict dementia within one year of its onset with an accuracy of nearly 80% despite the fact that it was trained using routine care data. Moreover, the analysis of the models identified important predictors for dementia. Some of these predictors (e.g., age and hypertensive disorders) are already confirmed by the literature while others, especially the ones derived from the unstructured medical notes, require further clinical analysis.

Passive Digital Signature for Early Identification of Alzheimer's Disease and Related Dementia.

OBJECTIVES: Developing scalable strategies for the early identification of Alzheimer's disease and related dementia (ADRD) is important. We aimed to develop a passive digital signature for early identification of ADRD using electronic medical record (EMR) data. DESIGN: A case-control study. SETTING: The Indiana Network for Patient Care (INPC), a regional health information exchange in Indiana. PARTICIPANTS: Patients identified with ADRD and matched controls. MEASUREMENTS: We used data from the INPC that includes structured and unstructured (visit notes, progress notes, medication notes) EMR data. Cases and controls were matched on age, race, and sex. The derivation sample consisted of 10 504 cases and 39 510 controls; the validation sample included 4500 cases and 16 952 controls. We constructed models to identify early 1- to 10-year, 3- to 10-year, and 5- to 10-year ADRD signatures. The analyses included 14 diagnostic risk variables and 10 drug classes in addition to new variables produced from unstructured data (eg, disorientation, confusion, wandering, apraxia, etc). The area under the receiver operating characteristics (AUROC) curve was used to determine the best models. RESULTS: The AUROC curves for the validation samples for the 1- to 10-year, 3- to 10-year, and 5- to 10-year models that used only structured data were .689, .649, and .633, respectively. For the same samples and years, models that used both structured and unstructured data produced AUROC curves of .798, .748, and .704, respectively. Using a cutoff to maximize sensitivity and specificity, the 1- to 10-year, 3- to 10-year, and 5- to 10-year models had sensitivity that ranged from 51% to 62% and specificity that ranged from 80% to 89%. CONCLUSION: EMR-based data provide a targeted and scalable process for early identification of risk of ADRD as an alternative to traditional population screening. J Am Geriatr Soc 68:511-518, 2020.

Innovation at the Intersection of Clinical Trials and Real-World Data Science to Advance Patient Care.

While efficacy and safety data collected from randomized clinical trials are the evidentiary standard for determining market authorization, this alone may no longer be sufficient to address the needs of key stakeholders (regulators, providers, and payers) and guarantee long-term success of pharmaceutical products. There is a heightened interest from stakeholders on understanding the use of real-world evidence (RWE) to substantiate benefit-risk assessment and support the value of a new drug. This review provides an overview of real-world data (RWD) and related advances in the regulatory framework, and discusses their impact on clinical research and development. A framework for linking drug development decisions with the value proposition of the drug, utilizing pharmacokinetic-pharmacodynamic-pharmacoeconomic models, is introduced. The summary presented here is based on the presentations and discussion at the symposium entitled Innovation at the Intersection of Clinical Trials and Real-World Data to Advance Patient Care at the American Society for Clinical Pharmacology and Therapeutics (ASCPT) 2017 Annual Meeting.

Targeting Bacterial Cell Wall Peptidoglycan Synthesis by Inhibition of Glycosyltransferase Activity.

Synthesis of bacterial cell wall peptidoglycan requires glycosyltransferase enzymes that transfer the disaccharide-peptide from lipid II onto the growing glycan chain. The polymerization of the glycan chain precedes cross-linking by penicillin-binding proteins and is essential for growth for key bacterial pathogens. As such, bacterial cell wall glycosyltransferases are an attractive target for antibiotic drug discovery. However, significant challenges to the development of inhibitors for these targets include the development of suitable assays and chemical matter that is suited to the nature of the binding site. We developed glycosyltransferase enzymatic activity and binding assays using the natural products moenomycin and vancomycin as model inhibitors. In addition, we designed a library of disaccharide compounds based on the minimum moenomycin fragment with peptidoglycan glycosyltransferase inhibitory activity and based on a more drug-like and synthetically versatile disaccharide building block. A subset of these disaccharide compounds bound and inhibited the glycosyltransferase enzymes, and these compounds could serve as chemical entry points for antibiotic development.

Recent estimates of the structure of the factor VIIa (FVIIa)/tissue factor (TF) and factor Xa (FXa) ternary complex.

The putative structure of the Tissue Factor/Factor VIIa/Factor Xa (TF/FVIIa/FXa) ternary complex is reconsidered. Two independently derived docking models proposed in 2003 (one for our laboratory: CHeA and one from the Scripps laboratory: Ss) are dynamically equilibrated for over 10 ns in an electrically neutral solution using all-atom molecular dynamics. Although the dynamical models (CHeB and Se) differ in atomic detail, there are similarities in that TF is found to interact with the gamma-carboxyglutamic acid (Gla) and Epidermal Growth Factor-like 1 (EGF-1) domains of FXa, and FVIIa is found to interact with the Gla, EGF-2 and serine protease (SP) domains of FXa in both models. FVIIa does not interact with the FXa EGF-1 domain in Se and the EGF domains of FVIIa do not interact with FXa in the CHeB. Both models are consistent with experimentally suggested contacts between the SP domain of FVIIa with the EGF-2 and SP domains of FXa.

Computational study of the putative active form of protein Z (PZa): sequence design and structural modeling.

Although protein Z (PZ) has a domain arrangement similar to the essential coagulation proteins FVII, FIX, FX, and protein C, its serine protease (SP)-like domain is incomplete and does not exhibit proteolytic activity. We have generated a trial sequence of putative activated protein Z (PZa) by identifying amino acid mutations in the SP-like domain that might reasonably resurrect the serine protease catalytic activity of PZ. The structure of the activated form was then modeled based on the proposed sequence using homology modeling and solvent-equilibrated molecular dynamics simulations. In silico docking of inhibitors of FVIIa and FXa to the putative active site of equilibrated PZa, along with structural comparison with its homologous proteins, suggest that the designed PZa can possibly act as a serine protease.

Sequence- and target-independent angiogenesis suppression by siRNA via TLR3.

Clinical trials of small interfering RNA (siRNA) targeting vascular endothelial growth factor-A (VEGFA) or its receptor VEGFR1 (also called FLT1), in patients with blinding choroidal neovascularization (CNV) from age-related macular degeneration, are premised on gene silencing by means of intracellular RNA interference (RNAi). We show instead that CNV inhibition is a siRNA-class effect: 21-nucleotide or longer siRNAs targeting non-mammalian genes, non-expressed genes, non-genomic sequences, pro- and anti-angiogenic genes, and RNAi-incompetent siRNAs all suppressed CNV in mice comparably to siRNAs targeting Vegfa or Vegfr1 without off-target RNAi or interferon-alpha/beta activation. Non-targeted (against non-mammalian genes) and targeted (against Vegfa or Vegfr1) siRNA suppressed CNV via cell-surface toll-like receptor 3 (TLR3), its adaptor TRIF, and induction of interferon-gamma and interleukin-12. Non-targeted siRNA suppressed dermal neovascularization in mice as effectively as Vegfa siRNA. siRNA-induced inhibition of neovascularization required a minimum length of 21 nucleotides, a bridging necessity in a modelled 2:1 TLR3-RNA complex. Choroidal endothelial cells from people expressing the TLR3 coding variant 412FF were refractory to extracellular siRNA-induced cytotoxicity, facilitating individualized pharmacogenetic therapy. Multiple human endothelial cell types expressed surface TLR3, indicating that generic siRNAs might treat angiogenic disorders that affect 8% of the world's population, and that siRNAs might induce unanticipated vascular or immune effects.

Molecular modeling of the oxidized form of nuclear factor-kappa B suggests a mechanism for redox regulation of DNA binding and transcriptional activation.

NF-kappa B is an important transcriptional regulator of numerous cellular genes, as well as viruses such as HIV-1. Oxidative stimuli in the cytosol are associated with nuclear translocation of NF-kappa B, whereas in the nucleus, reductive activation by thioredoxin is required for NF-kappa B to bind to DNA and activate target genes. Experimental structures of the reduced form of NF-kappa B bound to its DNA targets are available, from which we have modeled the oxidized form of NF-kappa B homodimer by removal of bound DNA, and modification via a hinge movement of a linker between the dimerization and DNA-binding domains of each subunit. These torsional motions enabled the formation of an inter-subunit disulfide bond between the Cys62 residues of each monomer; the resulting structure was refined using molecular dynamics simulation. The final model of oxidized, disulfide-bridged NF-kappaB is more compact than the open, reduced form. This may facilitate its nuclear translocation through small pores in the nuclear envelope, in response to oxidative stimuli in the cytosol. Furthermore, the inter-subunit disulfide blocks DNA from entering the active site of the oxidized dimer, explaining why subsequent reduction to the thiol form in the nucleus is essential for DNA binding and transcriptional activation to occur.

Molecular docking and analysis of interactions between vascular endothelial growth factor (VEGF) and SPARC protein.

The extracellular module of SPARC/osteonectin binds to vascular endothelial growth factor (VEGF) and inhibits VEGF-stimulated proliferation of endothelial cells. In an attempt to identify the binding site for SPARC on VEGF, we hypothesized that this binding site could overlap at least partially the binding site of VEGF receptor 1 (VEGFR-1), as SPARC acts by preventing VEGF-induced phosphorylation of VEGFR-1. To this end, a docking simulation was carried out using a predictive docking tool to obtain modeled structures of the VEGF-SPARC complex. The predicted structure of VEGF-SPARC complex indicates that the extracellular domain of SPARC interacts with the VEGFR-1 binding site of VEGF, and is consistent with known biochemical data. Following molecular dynamics refinement, side-chain interactions at the protein interface were identified that were predicted to contribute substantially to the free energy of binding. These provide a detailed prediction of key amino acid side-chain interactions at the protein-protein interface. To validate the model further, the identified interactions will be used for designing mutagenesis studies to investigate their effect on binding activity. This model of the VEGF-SPARC complex should provide a basis for future studies aimed at identifying inhibitors of VEGF-induced angiogenesis.