Scalable and accurate deep learning with electronic health records.

Predictive modeling with electronic health record (EHR) data is anticipated to drive personalized medicine and improve healthcare quality. Constructing predictive statistical models typically requires extraction of curated predictor variables from normalized EHR data, a labor-intensive process that discards the vast majority of information in each patient's record. We propose a representation of patients' entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format. We demonstrate that deep learning methods using this representation are capable of accurately predicting multiple medical events from multiple centers without site-specific data harmonization. We validated our approach using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 h. In the sequential format we propose, this volume of EHR data unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality (area under the receiver operator curve [AUROC] across sites 0.93-0.94), 30-day unplanned readmission (AUROC 0.75-0.76), prolonged length of stay (AUROC 0.85-0.86), and all of a patient's final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios. In a case study of a particular prediction, we demonstrate that neural networks can be used to identify relevant information from the patient's chart.

An angiotensin II type 1 receptor activation switch patch revealed through evolutionary trace analysis.

Seven transmembrane (7TM) or G protein-coupled receptors constitute a large superfamily of cell surface receptors sharing a structural motif of seven transmembrane spanning alpha helices. Their activation mechanism most likely involves concerted movements of the transmembrane helices, but remains to be completely resolved. Evolutionary Trace (ET) analysis is a computational method, which identifies clusters of functionally important residues by integrating information on evolutionary important residue variations with receptor structure. Combined with known mutational data, ET predicted a patch of residues in the cytoplasmic parts of TM2, TM3, and TM6 to form an activation switch that is common to all family A 7TM receptors. We tested this hypothesis in the rat Angiotensin II (Ang II) type 1a (AT1a) receptor. The receptor has important roles in the cardiovascular system, but has also frequently been applied as a model for 7TM receptor activation and signaling. Six mutations: F66A, L67R, L70R, L119R, D125A, and I245F were targeted to the putative switch and assayed for changes in activation state by their ligand binding, signaling, and trafficking properties. All but one receptor mutant (that was not expressed well) displayed phenotypes associated with changed activation state, such as increased agonist affinity or basal activity, promiscuous activation, or constitutive internalization highlighting the importance of testing different signaling pathways. We conclude that this evolutionary important patch mediates interactions important for maintaining the inactive state. More broadly, these observations in the AT1 receptor are consistent with computational predictions of a generic role for this patch in 7TM receptor activation.

Essential helix interactions in the anion transporter domain of prestin revealed by evolutionary trace analysis.

Prestin, a member of the SLC26A family of anion transporters, is a polytopic membrane protein found in outer hair cells (OHCs) of the mammalian cochlea. Prestin is an essential component of the membrane-based motor that enhances electromotility of OHCs and contributes to frequency sensitivity and selectivity in mammalian hearing. Mammalian cells expressing prestin display a nonlinear capacitance (NLC), widely accepted as the electrical signature of electromotility. The associated charge movement requires intracellular anions reflecting the membership of prestin in the SLC26A family. We used the computational approach of evolutionary trace analysis to identify candidate functional (trace) residues in prestin for mutational studies. We created a panel of mutations at each trace residue and determined membrane expression and nonlinear capacitance associated with each mutant. We observe that several residue substitutions near the conserved sulfate transporter domain of prestin either greatly reduce or eliminate NLC, and the effect is dependent on the size of the substituted residue. These data suggest that packing of helices and interactions between residues surrounding the "sulfate transporter motif" is essential for normal prestin activity.

beta-arrestin-dependent, G protein-independent ERK1/2 activation by the beta2 adrenergic receptor.

Physiological effects of beta adrenergic receptor (beta2AR) stimulation have been classically shown to result from G(s)-dependent adenylyl cyclase activation. Here we demonstrate a novel signaling mechanism wherein beta-arrestins mediate beta2AR signaling to extracellular-signal regulated kinases 1/2 (ERK 1/2) independent of G protein activation. Activation of ERK1/2 by the beta2AR expressed in HEK-293 cells was resolved into two components dependent, respectively, on G(s)-G(i)/protein kinase A (PKA) or beta-arrestins. G protein-dependent activity was rapid, peaking within 2-5 min, was quite transient, was blocked by pertussis toxin (G(i) inhibitor) and H-89 (PKA inhibitor), and was insensitive to depletion of endogenous beta-arrestins by siRNA. beta-Arrestin-dependent activation was slower in onset (peak 5-10 min), less robust, but more sustained and showed little decrement over 30 min. It was insensitive to pertussis toxin and H-89 and sensitive to depletion of either beta-arrestin1 or -2 by small interfering RNA. In G(s) knock-out mouse embryonic fibroblasts, wild-type beta2AR recruited beta-arrestin2-green fluorescent protein and activated pertussis toxin-insensitive ERK1/2. Furthermore, a novel beta2AR mutant (beta2AR(T68F,Y132G,Y219A) or beta2AR(TYY)), rationally designed based on Evolutionary Trace analysis, was incapable of G protein activation but could recruit beta-arrestins, undergo beta-arrestin-dependent internalization, and activate beta-arrestin-dependent ERK. Interestingly, overexpression of GRK5 or -6 increased mutant receptor phosphorylation and beta-arrestin recruitment, led to the formation of stable receptor-beta-arrestin complexes on endosomes, and increased agonist-stimulated phospho-ERK1/2. In contrast, GRK2, membrane translocation of which requires Gbetagamma release upon G protein activation, was ineffective unless it was constitutively targeted to the plasma membrane by a prenylation signal (CAAX). These findings demonstrate that the beta2AR can signal to ERK via a GRK5/6-beta-arrestin-dependent pathway, which is independent of G protein coupling.

Evolutionary trace of G protein-coupled receptors reveals clusters of residues that determine global and class-specific functions.

G protein-coupled receptor (GPCR) activation mediated by ligand-induced structural reorganization of its helices is poorly understood. To determine the universal elements of this conformational switch, we used evolutionary tracing (ET) to identify residue positions commonly important in diverse GPCRs. When mapped onto the rhodopsin structure, these trace residues cluster into a network of contacts from the retinal binding site to the G protein-coupling loops. Their roles in a generic transduction mechanism were verified by 211 of 239 published mutations that caused functional defects. When grouped according to the nature of the defects, these residues sub-divided into three striking sub-clusters: a trigger region, where mutations mostly affect ligand binding, a coupling region near the cytoplasmic interface to the G protein, where mutations affect G protein activation, and a linking core in between where mutations cause constitutive activity and other defects. Differential ET analysis of the opsin family revealed an additional set of opsin-specific residues, several of which form part of the retinal binding pocket, and are known to cause functional defects upon mutation. To test the predictive power of ET, we introduced novel mutations in bovine rhodopsin at a globally important position, Leu-79, and at an opsin-specific position, Trp-175. Both were functionally critical, causing constitutive G protein activation of the mutants and rapid loss of regeneration after photobleaching. These results define in GPCRs a canonical signal transduction mechanism where ligand binding induces conformational changes propagated through adjacent trigger, linking core, and coupling regions.

Accurate and scalable identification of functional sites by evolutionary tracing.

A common difficulty in post genomics biology is that large-scale techniques of data collection often strip away information on the biological context of these data. The result is a massive number of disconnected observations on sequence, structure, and function from which underlying patterns and biological meaning are obscured. One solution is to build computational filters that pick out sufficiently few facts, relevant to a query, that their relationship is immediately apparent and experimentally testable. Typically, these filters rely on mathematics and statistics, and on first principles from physics and chemistry. We show here that evolution itself can be used to filter sequence and structure data in order to identify evolutionarily important amino acids. A general property of these residues is that they form clusters in native protein structures and point to regions where mutations have the greatest biological impact. The result is an accurate method of functional site annotation that is scalable for structural proteomics.

Conserved motifs in somatostatin, D2-dopamine, and alpha 2B-adrenergic receptors for inhibiting the Na-H exchanger, NHE1.

Receptor subtypes within families of G protein-coupled receptors that are activated by similar ligands can regulate distinct intracellular effectors. We identified conserved motifs within intracellular domains 2 and 3 of selective subtypes of several G protein-coupled receptor families that confer coupling to the Na-H exchanger, NHE1. A T(s,p)V motif within intracellular domain 2 and a QQ(r) motif within intracellular domain 3 are shared by the somatostatin receptor subtypes SSTR1, -3, and -4, which couple to the inhibition of NHE1, but not by SSTR2 and -5, which do not signal to NHE1. Only the collective substitution of cognate SSTR2 residues with these two motifs conferred the ability of mutant SSTR2 to inhibit NHE1. Both motifs are present in D(2)-dopamine receptors, which inhibit NHE1, and in alpha(2B)-adrenergic receptors, which couple to the inhibition of NHE1, but not in alpha(2A)-adrenergic receptors, which do not regulate NHE1. These findings indicate that motifs shared by different subfamilies of G protein-coupled receptors, but not necessarily by receptor subtypes within a subfamily, can confer coupling to a common effector.

Structural clusters of evolutionary trace residues are statistically significant and common in proteins.

Given the massive increase in the number of new sequences and structures, a critical problem is how to integrate these raw data into meaningful biological information. One approach, the Evolutionary Trace, or ET, uses phylogenetic information to rank the residues in a protein sequence by evolutionary importance and then maps those ranked at the top onto a representative structure. If these residues form structural clusters, they can identify functional surfaces such as those involved in molecular recognition. Now that a number of examples have shown that ET can identify binding sites and focus mutational studies on their relevant functional determinants, we ask whether the method can be improved so as to be applicable on a large scale. To address this question, we introduce a new treatment of gaps resulting from insertions and deletions, which streamlines the selection of sequences used as input. We also introduce objective statistics to assess the significance of the total number of clusters and of the size of the largest one. As a result of the novel treatment of gaps, ET performance improves measurably. We find evolutionarily privileged clusters that are significant at the 5% level in 45 out of 46 (98%) proteins drawn from a variety of structural classes and biological functions. In 37 of the 38 proteins for which a protein-ligand complex is available, the dominant cluster contacts the ligand. We conclude that spatial clustering of evolutionarily important residues is a general phenomenon, consistent with the cooperative nature of residues that determine structure and function. In practice, these results suggest that ET can be applied on a large scale to identify functional sites in a significant fraction of the structures in the protein databank (PDB). This approach to combining raw sequences and structure to obtain detailed insights into the molecular basis of function should prove valuable in the context of the Structural Genomics Initiative.