Using topic modeling via non-negative matrix factorization to identify relationships between genetic variants and disease phenotypes: A case study of Lipoprotein(a) (LPA)

Pretreatment Lipoprotein(a) as a Biomarker for EGFR Mutation and Prognosis in Lung Adenocarcinoma

Purpose

This study aims to investigate the correlation between pretreatment serum lipoprotein(a) [Lp(a)] and epidermal growth factor receptor (EGFR) gene mutations, as well as its predictive value for progression-free survival (PFS) in advanced lung adenocarcinoma patients receiving epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKIs) therapy.

Patients and Methods

We determined the optimal cutoff value for Lp(a) by receiver operating characteristic (ROC) curves and Youden's index to categorize Lp(a) into high and low groups. Logistic regression was used to analyze the EGFR mutation rate in different groups. Additionally, the relationship between pretreatment Lp(a) levels and prognostic PFS in patients with advanced (TNM stage IIIB-IV) lung adenocarcinoma treated with EGFR-TKIs was retrospectively analyzed by Cox regression, survival and stratified analysis methods.

Results

We included 338 advanced lung adenocarcinoma patients, with median age of 64 years, and slightly more female patients (51.8%), most of whom had no smoking history (70.7%), no history of chronic lung disease (87.9%), and stage IV (81.1%) patients. The EGFR gene mutation rate was 55.3% and 123 patients were included in the prognostic evaluation through screening. The optimal cutoff value for Lp(a) was 20.48 mg/L. The mutation rate in the high Lp(a) group was significantly lower than the low Lp(a) group (48.0% vs 65.5%, p = 0.001). Multivariate logistic regression analysis indicated that Lp(a) is an independent predictor of EGFR mutations (OR = 0.41, 95% CI: 0.25-0.66, p<0.001). Survival analysis showed that the median PFS was significantly longer in the high Lp(a) level group compared to the low level group (16.1 months, 95% CI: 11.9-23.8 months vs 9.6 months, 95% CI: 8.9-13.3 months, p=0.015). Multivariate analysis confirmed that Lp(a) is an independent predictor of PFS in advanced lung adenocarcinoma patients receiving EGFR-TKIs treatment (HR = 0.42, 95% CI: 0.26-0.68, p<0.001).

Conclusion

Pretreatment Lp(a) may be a biomarker for EGFR mutations and the PFS in advanced lung adenocarcinoma patients undergoing EGFR-TKIs treatment.

Patient Subtyping via Learning Hidden Markov Models from Pairwise Co-occurrences in EHR Data

Patient subtyping is an effective tool to investigate clinically relevant characteristics of medical services, from which useful insights can be drawn for categorizing patients' conditions and informing disease progressions. Here we propose a hidden Markov model (HMM) based treatment of extracting patient subtypes from electronic health records (EHR). Using real-world clinical data, we show that the HMM based model can effectively identify the latent Markovian structure underlying the EHR data, and derive clinically or medically plausible patient subtypes, which can be used to categorize patients of various conditions.

Topic modeling identifies novel genetic loci associated with multimorbidities in UK Biobank

Many diseases show patterns of co-occurrence, possibly driven by systemic dysregulation of underlying processes affecting multiple traits. We have developed a method (treeLFA) for identifying such multimorbidities from routine health-care data, which combines topic modeling with an informative prior derived from medical ontology. We apply treeLFA to UK Biobank data and identify a variety of topics representing multimorbidity clusters, including a healthy topic. We find that loci identified using topic weights as traits in a genome-wide association study (GWAS) analysis, which we validated with a range of approaches, only partially overlap with loci from GWASs on constituent single diseases. We also show that treeLFA improves upon existing methods like latent Dirichlet allocation in various ways. Overall, our findings indicate that topic models can characterize multimorbidity patterns and that genetic analysis of these patterns can provide insight into the etiology of complex traits that cannot be determined from the analysis of constituent traits alone.

Improving Diagnostics with Deep Forest Applied to Electronic Health Records

An electronic health record (EHR) is a vital high-dimensional part of medical concepts. Discovering implicit correlations in the information of this data set and the research and informative aspects can improve the treatment and management process. The challenge of concern is the data sources' limitations in finding a stable model to relate medical concepts and use these existing connections. This paper presents Patient Forest, a novel end-to-end approach for learning patient representations from tree-structured data for readmission and mortality prediction tasks. By leveraging statistical features, the proposed model is able to provide an accurate and reliable classifier for predicting readmission and mortality. Experiments on MIMIC-III and eICU datasets demonstrate Patient Forest outperforms existing machine learning models, especially when the training data are limited. Additionally, a qualitative evaluation of Patient Forest is conducted by visualising the learnt representations in 2D space using the t-SNE, which further confirms the effectiveness of the proposed model in learning EHR representations.

Lack of significant associations between single nucleotide polymorphisms in LPAL2-LPA genetic region and all cancer incidence and mortality in Japanese population: The Japan public health center-based prospective study

BACKGROUND

High lipoprotein (a) level is an established cardiovascular risk, but its association with non-cardiovascular diseases, especially cancer, is controversial. Serum lipoprotein (a) levels vary widely by genetic backgrounds and are largely determined by the genetic variations of apolipoprotein (a) gene, LPA. In this study, we investigate the association between SNPs in LPA region and cancer incidence and mortality in Japanese.

METHODS

A genetic cohort study was conducted utilizing the data from 9923 participants in the Japan Public Health Center-based Prospective Study (JPHC Study). Twenty-five SNPs in the LPAL2-LPA region were selected from the genome-wide genotyped data. Cox regression analysis adjusted for the covariates and competing risks of death from other causes, were used to estimate the relative risk (hazard ratios (HR) with 95% confidence intervals (CI)) of overall and site-specific cancer incidence and mortality, for each SNP.

RESULTS

No significant association was found between SNPs in the LPAL2-LPA region and cancer incidence or mortality (overall/site-specific cancer). In men, however, HRs for stomach cancer incidence of 18SNPs were estimated higher than 1.5 (e.g., 2.15 for rs13202636, model free, 95%CI: 1.28-3.62) and those for stomach cancer mortality of 2SNPs (rs9365171, rs1367211) were estimated 2.13 (recessive, 95%CI:1.04-4.37) and 1.61 (additive, 95%CI: 1.00-2.59). Additionally, the minor allele for SNP rs3798220 showed increased death risk from colorectal cancer (CRC) in men (HR: 3.29, 95% CI:1.59 - 6.81) and decreased CRC incidence risk in women (HR: 0.46, 95%CI: 0.22-0.94). Minor allele carrier of any of 4SNPs could have risk of prostate cancer incidence (e.g., rs9365171 dominant, HR: 1.71, 95%CI: 1.06-2.77).

CONCLUSIONS

None of the 25 SNPs in the LPAL2-LPA region was found to be significantly associated with cancer incidence or mortality. Considering the possible association between SNPs in LPAL2-LPA region and colorectal, prostate and stomach cancer incidence or mortality, further analysis using different cohorts is warranted.

Polygenic Risk Score in African populations: progress and challenges

Polygenic Risk Score (PRS) analysis is a method that predicts the genetic risk of an individual towards targeted traits. Even when there are no significant markers, it gives evidence of a genetic effect beyond the results of Genome-Wide Association Studies (GWAS). Moreover, it selects  single nucleotide polymorphisms (SNPs) that  contribute to the disease with low effect size  making it more precise at individual level risk prediction. PRS  analysis addresses the shortfall of GWAS by taking into account the SNPs/alleles with  low effect size but play an indispensable role to the observed phenotypic/trait variance.  PRS analysis has  applications that investigate the genetic basis of several traits, which includes rare diseases. However, the accuracy of PRS analysis depends on the genomic data of the underlying population. For instance, several studies  show   that obtaining higher prediction power of PRS analysis is challenging for non-Europeans. In this manuscript, we review the conventional PRS methods and their application to sub-Saharan African communities. We conclude that  lack of sufficient GWAS data and tools is  the limiting factor of applying PRS analysis to sub-Saharan populations.   We recommend developing Africa-specific PRS methods and tools for estimating and analyzing  African population data   for clinical  evaluation of PRSs of interest and predicting  rare diseases.

Polygenic Risk Score in African populations: progress and challenges

Polygenic Risk Score (PRS) analysis is a method that predicts the genetic risk of an individual towards targeted traits. Even when there are no significant markers, it gives evidence of a genetic effect beyond the results of Genome-Wide Association Studies (GWAS). Moreover, it selects  single nucleotide polymorphisms (SNPs) that  contribute to the disease with low effect size  making it more precise at individual level risk prediction. PRS  analysis addresses the shortfall of GWAS by taking into account the SNPs/alleles with  low effect size but play an indispensable role to the observed phenotypic/trait variance.  PRS analysis has  applications that investigate the genetic basis of several traits, which includes rare diseases. However, the accuracy of PRS analysis depends on the genomic data of the underlying population. For instance, several studies  show   that obtaining higher prediction power of PRS analysis is challenging for non-Europeans. In this manuscript, we review the conventional PRS methods and their application to sub-Saharan African communities. We conclude that  lack of sufficient GWAS data and tools is  the limiting factor of applying PRS analysis to sub-Saharan populations.   We recommend developing Africa-specific PRS methods and tools for estimating and analyzing  African population data   for clinical  evaluation of PRSs of interest and predicting  rare diseases.

Identification of Functional Microbial Modules Through Network-Based Analysis of Meta-Microbial Features Using Matrix Factorization

As the microbiome is composed of a variety of microbial interactions, it is imperative in microbiome research to identify a microbial sub-community that collectively conducts a specific function. However, current methodologies have been highly limited to analyzing conditional abundance changes of individual microorganisms without considering group-wise collective microbial features. To overcome this limitation, we developed a network-based method using nonnegative matrix factorization (NMF) to identify functional meta-microbial features (MMFs) that, as a group, better discriminate specific environmental conditions of samples using microbiome data. As proof of concept, large-scale human microbiome data collected from different body sites were used to identify body site-specific MMFs by applying NMF. The statistical test for MMFs led us to identify highly discriminative MMFs on sample classes, called synergistic MMFs (SYMMFs). Finally, we constructed a SYMMF-based microbial interaction network (SYMMF-net) by integrating all of the SYMMF information. Network analysis revealed core microbial modules closely related to critical sample properties. Similar results were also found when the method was applied to various disease-associated microbiome data. The developed method interprets high-dimensional microbiome data by identifying functional microbial modules on sample properties and intuitively representing their systematic relationships via a microbial network.

Common genetic variation associated with Mendelian disease severity revealed through cryptic phenotype analysis

Clinical heterogeneity is common in Mendelian disease, but small sample sizes make it difficult to identify specific contributing factors. However, if a disease represents the severely affected extreme of a spectrum of phenotypic variation, then modifier effects may be apparent within a larger subset of the population. Analyses that take advantage of this full spectrum could have substantially increased power. To test this, we developed cryptic phenotype analysis, a model-based approach that infers quantitative traits that capture disease-related phenotypic variability using qualitative symptom data. By applying this approach to 50 Mendelian diseases in two cohorts, we identify traits that reliably quantify disease severity. We then conduct genome-wide association analyses for five of the inferred cryptic phenotypes, uncovering common variation that is predictive of Mendelian disease-related diagnoses and outcomes. Overall, this study highlights the utility of computationally-derived phenotypes and biobank-scale cohorts for investigating the complex genetic architecture of Mendelian diseases.

Integration of Omics and Phenotypic Data for Precision Medicine

Over the past two decades, biomedical research is moving toward a big-data-driven approach. The underlying causes of this transition include the ability to gather genetic or molecular profiles of humans faster, the increasing adoption of electronic health record (EHR) system, and the growing interest in linking omics and phenotypic data for analysis. The integration of individual's biology data (e.g., genomics, proteomics, metabolomics), and health-care data has created unprecedented opportunities for precision medicine, that is, a medical model that uses a patient's unique information, mainly genetic, to prevent, diagnose, or treat disease. This chapter reviewed the research opportunities and applications of integrating omics and phenotypic data for precision medicine, such as understanding the relationship between genotype and phenotype, disease subtyping, and diagnosis or prediction of adverse outcomes. We reviewed the recent advanced methods, particularly the machine learning and deep learning-based approaches used for harnessing and harmonizing the multiomics and phenotypic data to address these applications. We finally discussed the challenges and future directions.

Predicting Discharge Disposition Following Meningioma Resection Using a Multi-Institutional Natural Language Processing Model

BACKGROUND

Machine learning (ML)-based predictive models are increasingly common in neurosurgery, but typically require large databases of discrete variables for training. Natural language processing (NLP) can extract meaningful data from unstructured text.

OBJECTIVE

To present an NLP model that predicts nonhome discharge and a point-of-care implementation.

METHODS

We retrospectively collected age, preoperative notes, and radiology reports from 595 adults who underwent meningioma resection in an academic center from 1995 to 2015. A total of 32 algorithms were trained with the data; the 3 best performing algorithms were combined to form an ensemble. Predictive ability, assessed by area under the receiver operating characteristic curve (AUC) and calibration, was compared to a previously published model utilizing 52 neurosurgeon-selected variables. We then built a multi-institutional model by incorporating notes from 693 patients at another center into algorithm training. Permutation importance was used to analyze the relative importance of each input to model performance. Word clouds and non-negative matrix factorization were used to analyze predictive features of text.

RESULTS

The single-institution NLP model predicted nonhome discharge with AUC of 0.80 (95% CI = 0.74-0.86) on internal and 0.76 on holdout validation compared to AUC of 0.77 (95% CI = 0.73-0.81) and 0.74 for the 52-variable ensemble. The multi-institutional model performed similarly well with AUC = 0.78 (95% CI = 0.74-0.81) on internal and 0.76 on holdout validation. Preoperative notes most influenced predictions. The model is available at http://nlp-home.insds.org.

CONCLUSION

ML and NLP are underutilized in neurosurgery. Here, we construct a multi-institutional NLP model that predicts nonhome discharge.

TASTE: Temporal and Static Tensor Factorization for Phenotyping Electronic Health Records

Phenotyping electronic health records (EHR) focuses on defining meaningful patient groups (e.g., heart failure group and diabetes group) and identifying the temporal evolution of patients in those groups. Tensor factorization has been an effective tool for phenotyping. Most of the existing works assume either a static patient representation with aggregate data or only model temporal data. However, real EHR data contain both temporal (e.g., longitudinal clinical visits) and static information (e.g., patient demographics), which are difficult to model simultaneously. In this paper, we propose Temporal And Static TEnsor factorization (TASTE) that jointly models both static and temporal information to extract phenotypes. TASTE combines the PARAFAC2 model with non-negative matrix factorization to model a temporal and a static tensor. To fit the proposed model, we transform the original problem into simpler ones which are optimally solved in an alternating fashion. For each of the sub-problems, our proposed mathematical re-formulations lead to efficient sub-problem solvers. Comprehensive experiments on large EHR data from a heart failure (HF) study confirmed that TASTE is up to 14× faster than several baselines and the resulting phenotypes were confirmed to be clinically meaningful by a cardiologist. Using 60 phenotypes extracted by TASTE, a simple logistic regression can achieve the same level of area under the curve (AUC) for HF prediction compared to a deep learning model using recurrent neural networks (RNN) with 345 features.

Artificial Intelligence Pipeline to Bridge the Gap between Bench Researchers and Clinical Researchers in Precision Medicine

Precision medicine informatics is a field of research that incorporates learning systems that generate new knowledge to improve individualized treatments using integrated data sets and models. Given the ever-increasing volumes of data that are relevant to patient care, artificial intelligence (AI) pipelines need to be a central component of such research to speed discovery. Applying AI methodology to complex multidisciplinary information retrieval can support efforts to discover bridging concepts within collaborating communities. This dovetails with precision medicine research, given the information rich multi-omic data that are used in precision medicine analysis pipelines. In this perspective article we define a prototype AI pipeline to facilitate discovering research connections between bioinformatics and clinical researchers. We propose building knowledge representations that are iteratively improved through AI and human-informed learning feedback loops supported through crowdsourcing. To illustrate this, we will explore the specific use case of nonalcoholic fatty liver disease, a growing health care problem. We will examine AI pipeline construction and utilization in relation to bench-to-bedside bridging concepts with interconnecting knowledge representations applicable to bioinformatics researchers and clinicians.

Mapping ICD-10 and ICD-10-CM Codes to Phecodes: Workflow Development and Initial Evaluation

BACKGROUND

The phecode system was built upon the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) for phenome-wide association studies (PheWAS) using the electronic health record (EHR).

OBJECTIVE

The goal of this paper was to develop and perform an initial evaluation of maps from the International Classification of Diseases, 10th Revision (ICD-10) and the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes to phecodes.

METHODS

We mapped ICD-10 and ICD-10-CM codes to phecodes using a number of methods and resources, such as concept relationships and explicit mappings from the Centers for Medicare & Medicaid Services, the Unified Medical Language System, Observational Health Data Sciences and Informatics, Systematized Nomenclature of Medicine-Clinical Terms, and the National Library of Medicine. We assessed the coverage of the maps in two databases: Vanderbilt University Medical Center (VUMC) using ICD-10-CM and the UK Biobank (UKBB) using ICD-10. We assessed the fidelity of the ICD-10-CM map in comparison to the gold-standard ICD-9-CM phecode map by investigating phenotype reproducibility and conducting a PheWAS.

RESULTS

We mapped >75% of ICD-10 and ICD-10-CM codes to phecodes. Of the unique codes observed in the UKBB (ICD-10) and VUMC (ICD-10-CM) cohorts, >90% were mapped to phecodes. We observed 70-75% reproducibility for chronic diseases and <10% for an acute disease for phenotypes sourced from the ICD-10-CM phecode map. Using the ICD-9-CM and ICD-10-CM maps, we conducted a PheWAS with a Lipoprotein(a) genetic variant, rs10455872, which replicated two known genotype-phenotype associations with similar effect sizes: coronary atherosclerosis (ICD-9-CM: P<.001; odds ratio (OR) 1.60 [95% CI 1.43-1.80] vs ICD-10-CM: P<.001; OR 1.60 [95% CI 1.43-1.80]) and chronic ischemic heart disease (ICD-9-CM: P<.001; OR 1.56 [95% CI 1.35-1.79] vs ICD-10-CM: P<.001; OR 1.47 [95% CI 1.22-1.77]).

CONCLUSIONS

This study introduces the beta versions of ICD-10 and ICD-10-CM to phecode maps that enable researchers to leverage accumulated ICD-10 and ICD-10-CM data for PheWAS in the EHR.

Detecting time-evolving phenotypic topics via tensor factorization on electronic health records: Cardiovascular disease case study

OBJECTIVE

Discovering subphenotypes of complex diseases can help characterize disease cohorts for investigative studies aimed at developing better diagnoses and treatments. Recent advances in unsupervised machine learning on electronic health record (EHR) data have enabled researchers to discover phenotypes without input from domain experts. However, most existing studies have ignored time and modeled diseases as discrete events. Uncovering the evolution of phenotypes - how they emerge, evolve and contribute to health outcomes - is essential to define more precise phenotypes and refine the understanding of disease progression. Our objective was to assess the benefits of an unsupervised approach that incorporates time to model diseases as dynamic processes in phenotype discovery.

METHODS

In this study, we applied a constrained non-negative tensor-factorization approach to characterize the complexity of cardiovascular disease (CVD) patient cohort based on longitudinal EHR data. Through tensor-factorization, we identified a set of phenotypic topics (i.e., subphenotypes) that these patients established over the 10 years prior to the diagnosis of CVD, and showed the progress pattern. For each identified subphenotype, we examined its association with the risk for adverse cardiovascular outcomes estimated by the American College of Cardiology/American Heart Association Pooled Cohort Risk Equations, a conventional CVD-risk assessment tool frequently used in clinical practice. Furthermore, we compared the subsequent myocardial infarction (MI) rates among the six most prevalent subphenotypes using survival analysis.

RESULTS

From a cohort of 12,380 adult CVD individuals with 1068 unique PheCodes, we successfully identified 14 subphenotypes. Through the association analysis with estimated CVD risk for each subtype, we found some phenotypic topics such as Vitamin D deficiency and depression, Urinary infections cannot be explained by the conventional risk factors. Through a survival analysis, we found markedly different risks of subsequent MI following the diagnosis of CVD among the six most prevalent topics (p < 0.0001), indicating these topics may capture clinically meaningful subphenotypes of CVD.

CONCLUSION

This study demonstrates the potential benefits of using tensor-decomposition to model diseases as dynamic processes from longitudinal EHR data. Our results suggest that this data-driven approach may potentially help researchers identify complex and chronic disease subphenotypes in precision medicine research.