A public resource of single cell transcriptomes and multiscale networks from persons with and without Alzheimer's disease.

The emergence of technologies that can support high-throughput profiling of single cell transcriptomes offers to revolutionize the study of brain tissue from persons with and without Alzheimer's disease (AD). Integration of these data with additional complementary multiomics data such as genetics, proteomics and clinical data provides powerful opportunities to link observed cell subpopulations and molecular network features within a broader disease-relevant context. We report here single nucleus RNA sequencing (snRNA-seq) profiles generated from superior frontal gyrus cortical tissue samples from 101 exceptionally well characterized, aged subjects from the Banner Brain and Body Donation Program in combination with whole genome sequences. We report findings that link common AD risk variants with CR1 expression in oligodendrocytes as well as alterations in peripheral hematological lab parameters, with these observations replicated in an independent, prospective cohort study of ageing and dementia. We also observed an AD-associated CD83(+) microglial subtype with unique molecular networks that encompass many known regulators of AD-relevant microglial biology, and which are associated with immunoglobulin IgG4 production in the transverse colon. These findings illustrate the power of multi-tissue molecular profiling to contextualize snRNA-seq brain transcriptomics and reveal novel disease biology. The transcriptomic, genetic, phenotypic, and network data resources described within this study are available for access and utilization by the scientific community.

Validation of a Transcriptome-Based Assay for Classifying Cancers of Unknown Primary Origin.

INTRODUCTION: Cancers assume a variety of distinct histologies, and may originate from a myriad of sites including solid organs, hematopoietic cells, and connective tissue. Clinical decision-making based on consensus guidelines such as the National Comprehensive Cancer Network (NCCN) is often predicated on a specific histologic and anatomic diagnosis, supported by clinical features and pathologist interpretation of morphology and immunohistochemical (IHC) staining patterns. However, in patients with nonspecific morphologic and IHC findings-in addition to ambiguous clinical presentations such as recurrence versus new primary-a definitive diagnosis may not be possible, resulting in the patient being categorized as having a cancer of unknown primary (CUP). Therapeutic options and clinical outcomes are poor for patients with CUP, with a median survival of 8-11 months. METHODS: Here, we describe and validate the Tempus Tumor Origin (Tempus TO) assay, an RNA-sequencing-based machine learning classifier capable of discriminating between 68 clinically relevant cancer subtypes. Model accuracy was assessed using primary and/or metastatic samples with known subtype. RESULTS: We show that the Tempus TO model is 91% accurate when assessed on both a retrospectively held out cohort and a set of samples sequenced after model freeze that collectively contained 9210 total samples with known diagnoses. When evaluated on a cohort of CUPs, the model recapitulated established associations between genomic alterations and cancer subtype. DISCUSSION: Combining diagnostic prediction tests (e.g., Tempus TO) with sequencing-based variant reporting (e.g., Tempus xT) may expand therapeutic options for patients with cancers of unknown primary or uncertain histology.

Quantifying the impact of immune history and variant on SARS-CoV-2 viral kinetics and infection rebound: A retrospective cohort study.

Background: The combined impact of immunity and SARS-CoV-2 variants on viral kinetics during infections has been unclear. Methods: We characterized 1,280 infections from the National Basketball Association occupational health cohort identified between June 2020 and January 2022 using serial RT-qPCR testing. Logistic regression and semi-mechanistic viral RNA kinetics models were used to quantify the effect of age, variant, symptom status, infection history, vaccination status and antibody titer to the founder SARS-CoV-2 strain on the duration of potential infectiousness and overall viral kinetics. The frequency of viral rebounds was quantified under multiple cycle threshold (Ct) value-based definitions. Results: Among individuals detected partway through their infection, 51.0% (95% credible interval [CrI]: 48.3-53.6%) remained potentially infectious (Ct <30) 5 days post detection, with small differences across variants and vaccination status. Only seven viral rebounds (0.7%; N=999) were observed, with rebound defined as 3+days with Ct <30 following an initial clearance of 3+days with Ct ≥30. High antibody titers against the founder SARS-CoV-2 strain predicted lower peak viral loads and shorter durations of infection. Among Omicron BA.1 infections, boosted individuals had lower pre-booster antibody titers and longer clearance times than non-boosted individuals. Conclusions: SARS-CoV-2 viral kinetics are partly determined by immunity and variant but dominated by individual-level variation. Since booster vaccination protects against infection, longer clearance times for BA.1-infected, boosted individuals may reflect a less effective immune response, more common in older individuals, that increases infection risk and reduces viral RNA clearance rate. The shifting landscape of viral kinetics underscores the need for continued monitoring to optimize isolation policies and to contextualize the health impacts of therapeutics and vaccines. Funding: Supported in part by CDC contract #200-2016-91779, a sponsored research agreement to Yale University from the National Basketball Association contract #21-003529, and the National Basketball Players Association.

Saracatinib, a Selective Src Kinase Inhibitor, Blocks Fibrotic Responses in Preclinical Models of Pulmonary Fibrosis.

Rationale: Idiopathic pulmonary fibrosis (IPF) is a chronic, progressive, and often fatal disorder. Two U.S. Food and Drug Administration-approved antifibrotic drugs, nintedanib and pirfenidone, slow the rate of decline in lung function, but responses are variable and side effects are common. Objectives: Using an in silico data-driven approach, we identified a robust connection between the transcriptomic perturbations in IPF disease and those induced by saracatinib, a selective Src kinase inhibitor originally developed for oncological indications. Based on these observations, we hypothesized that saracatinib would be effective at attenuating pulmonary fibrosis. Methods: We investigated the antifibrotic efficacy of saracatinib relative to nintedanib and pirfenidone in three preclinical models: 1) in vitro in normal human lung fibroblasts; 2) in vivo in bleomycin and recombinant Ad-TGF-ß (adenovirus transforming growth factor-ß) murine models of pulmonary fibrosis; and 3) ex vivo in mice and human precision-cut lung slices from these two murine models as well as patients with IPF and healthy donors. Measurements and Main Results: In each model, the effectiveness of saracatinib in blocking fibrogenic responses was equal or superior to nintedanib and pirfenidone. Transcriptomic analyses of TGF-ß-stimulated normal human lung fibroblasts identified specific gene sets associated with fibrosis, including epithelial-mesenchymal transition, TGF-ß, and WNT signaling that was uniquely altered by saracatinib. Transcriptomic analysis of whole-lung extracts from the two animal models of pulmonary fibrosis revealed that saracatinib reverted many fibrogenic pathways, including epithelial-mesenchymal transition, immune responses, and extracellular matrix organization. Amelioration of fibrosis and inflammatory cascades in human precision-cut lung slices confirmed the potential therapeutic efficacy of saracatinib in human lung fibrosis. Conclusions: These studies identify novel Src-dependent fibrogenic pathways and support the study of the therapeutic effectiveness of saracatinib in IPF treatment.

A Wipe-Based Stool Collection and Preservation Kit for Microbiome Community Profiling.

While a range of methods for stool collection exist, many require complicated, self-directed protocols and stool transfer. In this study, we introduce and validate a novel, wipe-based approach to fecal sample collection and stabilization for metagenomics analysis. A total of 72 samples were collected across four different preservation types: freezing at -20°C, room temperature storage, a commercial DNA preservation kit, and a dissolvable wipe used with DESS (dimethyl sulfoxide, ethylenediaminetetraacetic acid, sodium chloride) solution. These samples were sequenced and analyzed for taxonomic abundance metrics, bacterial metabolic pathway classification, and diversity analysis. Overall, the DESS wipe results validated the use of a wipe-based capture method to collect stool samples for microbiome analysis, showing an R2 of 0.96 for species across all kingdoms, as well as exhibiting a maintenance of Shannon diversity (3.1-3.3) and species richness (151-159) compared to frozen samples. Moreover, DESS showed comparable performance to the commercially available preservation kit (R2 of 0.98), and samples consistently clustered by subject across each method. These data support that the DESS wipe method can be used for stable, room temperature collection and transport of human stool specimens.

Deep learning-based brain transcriptomic signatures associated with the neuropathological and clinical severity of Alzheimer's disease.

Brain tissue gene expression from donors with and without Alzheimer's disease has been used to help inform the molecular changes associated with the development and potential treatment of this disorder. Here, we use a deep learning method to analyse RNA-seq data from 1114 brain donors from the Accelerating Medicines Project for Alzheimer's Disease consortium to characterize post-mortem brain transcriptome signatures associated with amyloid-ß plaque, tau neurofibrillary tangles and clinical severity in multiple Alzheimer's disease dementia populations. Starting from the cross-sectional data in the Religious Orders Study and Memory and Aging Project cohort (n = 634), a deep learning framework was built to obtain a trajectory that mirrors Alzheimer's disease progression. A severity index was defined to quantitatively measure the progression based on the trajectory. Network analysis was then carried out to identify key gene (index gene) modules present in the model underlying the progression. Within this data set, severity indexes were found to be very closely correlated with all Alzheimer's disease neuropathology biomarkers (R ∼ 0.5, P < 1e-11) and global cognitive function (R = -0.68, P < 2.2e-16). We then applied the model to additional transcriptomic data sets from different brain regions (MAYO, n = 266; Mount Sinai Brain Bank, n = 214), and observed that the model remained significantly predictive (P < 1e-3) of neuropathology and clinical severity. The index genes that significantly contributed to the model were integrated with Alzheimer's disease co-expression regulatory networks, resolving four discrete gene modules that are implicated in vascular and metabolic dysfunction in different cell types, respectively. Our work demonstrates the generalizability of this signature to frontal and temporal cortex measurements and additional brain donors with Alzheimer's disease, other age-related neurological disorders and controls, and revealed that the transcriptomic network modules contribute to neuropathological and clinical disease severity. This study illustrates the promise of using deep learning methods to analyse heterogeneous omics data and discover potentially targetable molecular networks that can inform the development, treatment and prevention of neurodegenerative diseases like Alzheimer's disease.

Predictive Modelling of Susceptibility to Substance Abuse, Mortality and Drug-Drug Interactions in Opioid Patients.

Objective: Opioids are a class of drugs that are known for their use as pain relievers. They bind to opioid receptors on nerve cells in the brain and the nervous system to mitigate pain. Addiction is one of the chronic and primary adverse events of prolonged usage of opioids. They may also cause psychological disorders, muscle pain, depression, anxiety attacks etc. In this study, we present a collection of predictive models to identify patients at risk of opioid abuse and mortality by using their prescription histories. Also, we discover particularly threatening drug-drug interactions in the context of opioid usage. Methods and Materials: Using a publicly available dataset from MIMIC-III, two models were trained, Logistic Regression with L2 regularization (baseline) and Extreme Gradient Boosting (enhanced model), to classify the patients of interest into two categories based on their susceptibility to opioid abuse. We've also used K-Means clustering, an unsupervised algorithm, to explore drug-drug interactions that might be of concern. Results: The baseline model for classifying patients susceptible to opioid abuse has an F1 score of 76.64% (accuracy 77.16%) while the enhanced model has an F1 score of 94.45% (accuracy 94.35%). These models can be used as a preliminary step towards inferring the causal effect of opioid usage and can help monitor the prescription practices to minimize the opioid abuse. Discussion and Conclusion: Results suggest that the enhanced model provides a promising approach in preemptive identification of patients at risk for opioid abuse. By discovering and correlating the patterns contributing to opioid overdose or abuse among a variety of patients, machine learning models can be used as an efficient tool to help uncover the existing gaps and/or fraudulent practices in prescription writing. To quote an example of one such incidental finding, our study discovered that insulin might possibly be interacting with opioids in an unfavourable way leading to complications in diabetic patients. This indicates that diabetic patients under long term opioid usage might need to take increased amounts of insulin to make it more effective. This observation backs up prior research studies done on a similar aspect. To increase the translational value of our work, the predictive models and the associated software code are made available under the MIT License.

Patient similarity network of newly diagnosed multiple myeloma identifies patient subgroups with distinct genetic features and clinical implications.

The remarkable genetic heterogeneity of multiple myeloma poses a substantial challenge for proper prognostication and clinical management of patients. Here, we introduce MM-PSN, the first multiomics patient similarity network of myeloma. MM-PSN enabled accurate dissection of the genetic and molecular landscape of the disease and determined 12 distinct subgroups defined by five data types generated from genomic and transcriptomic profiling of 655 patients. MM-PSN identified patient subgroups not previously described defined by specific patterns of alterations, enriched for specific gene vulnerabilities, and associated with potential therapeutic options. Our analysis revealed that co-occurrence of t(4;14) and 1q gain identified patients at significantly higher risk of relapse and shorter survival as compared to t(4;14) as a single lesion. Furthermore, our results show that 1q gain is the most important single lesion conferring high risk of relapse and that it can improve on the current International Staging Systems (ISS and R-ISS).

Repositioning of a novel GABA-B receptor agonist, AZD3355 (Lesogaberan), for the treatment of non-alcoholic steatohepatitis.

Non-alcoholic steatohepatitis (NASH) is a rising health challenge, with no approved drugs. We used a computational drug repositioning strategy to uncover a novel therapy for NASH, identifying a GABA-B receptor agonist, AZD3355 (Lesogaberan) previously evaluated as a therapy for esophageal reflux. AZD3355's potential efficacy in NASH was tested in human stellate cells, human precision cut liver slices (hPCLS), and in vivo in a well-validated murine model of NASH. In human stellate cells AZD3355 significantly downregulated profibrotic gene and protein expression. Transcriptomic analysis of these responses identified key regulatory nodes impacted by AZD3355, including Myc, as well as MAP and ERK kinases. In PCLS, AZD3355 down-regulated collagen1α1, αSMA and TNF-α mRNAs as well as secreted collagen1α1. In vivo, the drug significantly improved histology, profibrogenic gene expression, and tumor development, which was comparable to activity of obeticholic acid in a robust mouse model of NASH, but awaits further testing to determine its relative efficacy in patients. These data identify a well-tolerated clinical stage asset as a novel candidate therapy for human NASH through its hepatoprotective, anti-inflammatory and antifibrotic mechanisms of action. The approach validates computational methods to identify novel therapies in NASH in uncovering new pathways of disease development that can be rapidly translated into clinical trials.

Phe2vec: Automated disease phenotyping based on unsupervised embeddings from electronic health records.

Robust phenotyping of patients from electronic health records (EHRs) at scale is a challenge in clinical informatics. Here, we introduce Phe2vec, an automated framework for disease phenotyping from EHRs based on unsupervised learning and assess its effectiveness against standard rule-based algorithms from Phenotype KnowledgeBase (PheKB). Phe2vec is based on pre-computing embeddings of medical concepts and patients' clinical history. Disease phenotypes are then derived from a seed concept and its neighbors in the embedding space. Patients are linked to a disease if their embedded representation is close to the disease phenotype. Comparing Phe2vec and PheKB cohorts head-to-head using chart review, Phe2vec performed on par or better in nine out of ten diseases. Differently from other approaches, it can scale to any condition and was validated against widely adopted expert-based standards. Phe2vec aims to optimize clinical informatics research by augmenting current frameworks to characterize patients by condition and derive reliable disease cohorts.

Integrative chemogenomic analysis identifies small molecules that partially rescue ΔF508-CFTR for cystic fibrosis.

Rare diseases affect 10% of the first-world population, yet over 95% lack even a single pharmaceutical treatment. In the present age of information, we need ways to leverage our vast data and knowledge to streamline therapeutic development and lessen this gap. Here, we develop and implement an innovative informatic approach to identify therapeutic molecules, using the Connectivity Map and LINCS L1000 databases and disease-associated transcriptional signatures and pathways. We apply this to cystic fibrosis (CF), the most common genetic disease in people of northern European ancestry leading to chronic lung disease and reduced lifespan. We selected and tested 120 small molecules in a CF cell line, finding 8 with activity, and confirmed 3 in primary CF airway epithelia. Although chemically diverse, the transcriptional profiles of the hits suggest a common mechanism associated with the unfolded protein response and/or TNFα signaling. This study highlights the power of informatics to help identify new therapies and reveal mechanistic insights while moving beyond target-centric drug discovery.

Multiplex qPCR discriminates variants of concern to enhance global surveillance of SARS-CoV-2.

With the emergence of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) variants that may increase transmissibility and/or cause escape from immune responses, there is an urgent need for the targeted surveillance of circulating lineages. It was found that the B.1.1.7 (also 501Y.V1) variant, first detected in the United Kingdom, could be serendipitously detected by the Thermo Fisher TaqPath COVID-19 PCR assay because a key deletion in these viruses, spike Δ69-70, would cause a "spike gene target failure" (SGTF) result. However, a SGTF result is not definitive for B.1.1.7, and this assay cannot detect other variants of concern (VOC) that lack spike Δ69-70, such as B.1.351 (also 501Y.V2), detected in South Africa, and P.1 (also 501Y.V3), recently detected in Brazil. We identified a deletion in the ORF1a gene (ORF1a Δ3675-3677) in all 3 variants, which has not yet been widely detected in other SARS-CoV-2 lineages. Using ORF1a Δ3675-3677 as the primary target and spike Δ69-70 to differentiate, we designed and validated an open-source PCR assay to detect SARS-CoV-2 VOC. Our assay can be rapidly deployed in laboratories around the world to enhance surveillance for the local emergence and spread of B.1.1.7, B.1.351, and P.1.

Early introductions and transmission of SARS-CoV-2 variant B.1.1.7 in the United States.

The emergence and spread of SARS-CoV-2 lineage B.1.1.7, first detected in the United Kingdom, has become a global public health concern because of its increased transmissibility. Over 2,500 COVID-19 cases associated with this variant have been detected in the United States (US) since December 2020, but the extent of establishment is relatively unknown. Using travel, genomic, and diagnostic data, we highlight that the primary ports of entry for B.1.1.7 in the US were in New York, California, and Florida. Furthermore, we found evidence for many independent B.1.1.7 establishments starting in early December 2020, followed by interstate spread by the end of the month. Finally, we project that B.1.1.7 will be the dominant lineage in many states by mid- to late March. Thus, genomic surveillance for B.1.1.7 and other variants urgently needs to be enhanced to better inform the public health response.

PCR assay to enhance global surveillance for SARS-CoV-2 variants of concern.

With the emergence of SARS-CoV-2 variants that may increase transmissibility and/or cause escape from immune responses 1-3 , there is an urgent need for the targeted surveillance of circulating lineages. It was found that the B.1.1.7 (also 501Y.V1) variant first detected in the UK 4,5 could be serendipitously detected by the ThermoFisher TaqPath COVID-19 PCR assay because a key deletion in these viruses, spike Δ69-70, would cause a "spike gene target failure" (SGTF) result. However, a SGTF result is not definitive for B.1.1.7, and this assay cannot detect other variants of concern that lack spike Δ69-70, such as B.1.351 (also 501Y.V2) detected in South Africa 6 and P.1 (also 501Y.V3) recently detected in Brazil 7 . We identified a deletion in the ORF1a gene (ORF1a Δ3675-3677) in all three variants, which has not yet been widely detected in other SARS-CoV-2 lineages. Using ORF1a Δ3675-3677 as the primary target and spike Δ69-70 to differentiate, we designed and validated an open source PCR assay to detect SARS-CoV-2 variants of concern 8 . Our assay can be rapidly deployed in laboratories around the world to enhance surveillance for the local emergence spread of B.1.1.7, B.1.351, and P.1.

Early introductions and community transmission of SARS-CoV-2 variant B.1.1.7 in the United States.

The emergence and spread of SARS-CoV-2 lineage B.1.1.7, first detected in the United Kingdom, has become a global public health concern because of its increased transmissibility. Over 2500 COVID-19 cases associated with this variant have been detected in the US since December 2020, but the extent of establishment is relatively unknown. Using travel, genomic, and diagnostic data, we highlight the primary ports of entry for B.1.1.7 in the US and locations of possible underreporting of B.1.1.7 cases. Furthermore, we found evidence for many independent B.1.1.7 establishments starting in early December 2020, followed by interstate spread by the end of the month. Finally, we project that B.1.1.7 will be the dominant lineage in many states by mid to late March. Thus, genomic surveillance for B.1.1.7 and other variants urgently needs to be enhanced to better inform the public health response.

Longitudinal data in peripheral blood confirm that PM20D1 is a quantitative trait locus (QTL) for Alzheimer's disease and implicate its dynamic role in disease progression.

BACKGROUND: While Alzheimer's disease (AD) remains one of the most challenging diseases to tackle, genome-wide genetic/epigenetic studies reveal many disease-associated risk loci, which sheds new light onto disease heritability, provides novel insights to understand its underlying mechanism and potentially offers easily measurable biomarkers for early diagnosis and intervention. METHODS: We analyzed whole-genome DNA methylation data collected from peripheral blood in a cohort (n = 649) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and compared the DNA methylation level at baseline among participants diagnosed with AD (n = 87), mild cognitive impairment (MCI, n = 175) and normal controls (n = 162), to identify differentially methylated regions (DMRs). We also leveraged up to 4 years of longitudinal DNA methylation data, sampled at approximately 1 year intervals to model alterations in methylation levels at DMRs to delineate methylation changes associated with aging and disease progression, by linear mixed-effects (LME) modeling for the unchanged diagnosis groups (AD, MCI and control, respectively) and U-shape testing for those with changed diagnosis (converters). RESULTS: When compared with controls, patients with MCI consistently displayed promoter hypomethylation at methylation QTL (mQTL) gene locus PM20D1. This promoter hypomethylation was even more prominent in patients with mild to moderate AD. This is in stark contrast with previously reported hypermethylation in hippocampal and frontal cortex brain tissues in patients with advanced-stage AD at this locus. From longitudinal data, we show that initial promoter hypomethylation of PM20D1 during MCI and early stage AD is reversed to eventual promoter hypermethylation in late stage AD, which helps to complete a fuller picture of methylation dynamics. We also confirm this observation in an independent cohort from the Religious Orders Study and Memory and Aging Project (ROSMAP) Study using DNA methylation and gene expression data from brain tissues as neuropathological staging (Braak score) advances. CONCLUSIONS: Our results confirm that PM20D1 is an mQTL in AD and demonstrate that it plays a dynamic role at different stages of the disease. Further in-depth study is thus warranted to fully decipher its role in the evolution of AD and potentially explore its utility as a blood-based biomarker for AD.

Integrative genomic meta-analysis reveals novel molecular insights into cystic fibrosis and ΔF508-CFTR rescue.

Cystic fibrosis (CF), caused by mutations to CFTR, leads to severe and progressive lung disease. The most common mutant, ΔF508-CFTR, undergoes proteasomal degradation, extinguishing its anion channel function. Numerous in vitro interventions have been identified to partially rescue ΔF508-CFTR function yet remain poorly understood. Improved understanding of both the altered state of CF cells and the mechanisms of existing rescue strategies could reveal novel therapeutic strategies. Toward this aim, we measured transcriptional profiles of established temperature, genetic, and chemical interventions that rescue ΔF508-CFTR and also re-analyzed public datasets characterizing transcription in human CF vs. non-CF samples from airway and whole blood. Meta-analysis yielded a core disease signature and two core rescue signatures. To interpret these through the lens of prior knowledge, we compiled a "CFTR Gene Set Library" from literature. The core disease signature revealed remarkably strong connections to genes with established effects on CFTR trafficking and function and suggested novel roles of EGR1 and SGK1 in the disease state. Our data also revealed an unexpected mechanistic link between several genetic rescue interventions and the unfolded protein response. Finally, we found that C18, an analog of the CFTR corrector compound Lumacaftor, induces almost no transcriptional perturbation despite its rescue activity.

MicroRNA-195 controls MICU1 expression and tumor growth in ovarian cancer.

MICU1 is a mitochondrial inner membrane protein that inhibits mitochondrial calcium entry; elevated MICU1 expression is characteristic of many cancers, including ovarian cancer. MICU1 induces both glycolysis and chemoresistance and is associated with poor clinical outcomes. However, there are currently no available interventions to normalize aberrant MICU1 expression. Here, we demonstrate that microRNA-195-5p (miR-195) directly targets the 3' UTR of the MICU1 mRNA and represses MICU1 expression. Additionally, miR-195 is under-expressed in ovarian cancer cell lines, and restoring miR-195 expression reestablishes native MICU1 levels and the associated phenotypes. Stable expression of miR-195 in a human xenograft model of ovarian cancer significantly reduces tumor growth, increases tumor doubling times, and enhances overall survival. In conclusion, miR-195 controls MICU1 levels in ovarian cancer and could be exploited to normalize aberrant MICU1 expression, thus reversing both glycolysis and chemoresistance and consequently improving patient outcomes.

Deep representation learning of electronic health records to unlock patient stratification at scale.

Deriving disease subtypes from electronic health records (EHRs) can guide next-generation personalized medicine. However, challenges in summarizing and representing patient data prevent widespread practice of scalable EHR-based stratification analysis. Here we present an unsupervised framework based on deep learning to process heterogeneous EHRs and derive patient representations that can efficiently and effectively enable patient stratification at scale. We considered EHRs of 1,608,741 patients from a diverse hospital cohort comprising a total of 57,464 clinical concepts. We introduce a representation learning model based on word embeddings, convolutional neural networks, and autoencoders (i.e., ConvAE) to transform patient trajectories into low-dimensional latent vectors. We evaluated these representations as broadly enabling patient stratification by applying hierarchical clustering to different multi-disease and disease-specific patient cohorts. ConvAE significantly outperformed several baselines in a clustering task to identify patients with different complex conditions, with 2.61 entropy and 0.31 purity average scores. When applied to stratify patients within a certain condition, ConvAE led to various clinically relevant subtypes for different disorders, including type 2 diabetes, Parkinson's disease, and Alzheimer's disease, largely related to comorbidities, disease progression, and symptom severity. With these results, we demonstrate that ConvAE can generate patient representations that lead to clinically meaningful insights. This scalable framework can help better understand varying etiologies in heterogeneous sub-populations and unlock patterns for EHR-based research in the realm of personalized medicine.