Early prediction of in-hospital mortality utilizing multivariate predictive modelling of electronic medical records and socio-determinants of health of the first day of hospitalization

BACKGROUND

In France an average of 4% of hospitalized patients die during their hospital stay. To aid medical decision making and the attribution of resources, within a few days of admission the identification of patients at high risk of dying in hospital is essential.

METHODS

We used de-identified routine patient data available in the first 2 days of hospitalization in a French University Hospital (between 2016 and 2018) to build models predicting in-hospital mortality (at ≥ 2 and ≤ 30 days after admission). We tested nine different machine learning algorithms with repeated 10-fold cross-validation. Models were trained with 283 variables including age, sex, socio-determinants of health, laboratory test results, procedures (Classification of Medical Acts), medications (Anatomical Therapeutic Chemical code), hospital department/unit and home address (urban, rural etc.). The models were evaluated using various performance metrics. The dataset contained 123,729 admissions, of which the outcome for 3542 was all-cause in-hospital mortality and 120,187 admissions (no death reported within 30 days) were controls.

RESULTS

The support vector machine, logistic regression and Xgboost algorithms demonstrated high discrimination with a balanced accuracy of 0.81 (95%CI 0.80-0.82), 0.82 (95%CI 0.80-0.83) and 0.83 (95%CI 0.80-0.83) and AUC of 0.90 (95%CI 0.88-0.91), 0.90 (95%CI 0.89-0.91) and 0.90 (95%CI 0.89-0.91) respectively. The most predictive variables for in-hospital mortality in all three models were older age (greater risk), and admission with a confirmed appointment (reduced risk).

CONCLUSION

We propose three highly discriminating machine-learning models that could improve clinical and organizational decision making for adult patients at hospital admission.

Cohort profile for development of machine learning models to predict healthcare-related adverse events (Demeter): clinical objectives, data requirements for modelling and overview of data set for 2016-2018

PURPOSE

In-hospital health-related adverse events (HAEs) are a major concern for hospitals worldwide. In high-income countries, approximately 1 in 10 patients experience HAEs associated with their hospital stay. Estimating the risk of an HAE at the individual patient level as accurately as possible is one of the first steps towards improving patient outcomes. Risk assessment can enable healthcare providers to target resources to patients in greatest need through adaptations in processes and procedures. Electronic health data facilitates the application of machine-learning methods for risk analysis. We aim, first to reveal correlations between HAE occurrence and patients' characteristics and/or the procedures they undergo during their hospitalisation, and second, to build models that allow the early identification of patients at an elevated risk of HAE.

PARTICIPANTS

143 865 adult patients hospitalised at Grenoble Alpes University Hospital (France) between 1 January 2016 and 31 December 2018.

FINDINGS TO DATE

In this set-up phase of the project, we describe the preconditions for big data analysis using machine-learning methods. We present an overview of the retrospective de-identified multisource data for a 2-year period extracted from the hospital's Clinical Data Warehouse, along with social determinants of health data from the National Institute of Statistics and Economic Studies, to be used in machine learning (artificial intelligence) training and validation. No supplementary information or evaluation on the part of medical staff will be required by the information system for risk assessment.

FUTURE PLANS

We are using this data set to develop predictive models for several general HAEs including secondary intensive care admission, prolonged hospital stay, 7-day and 30-day re-hospitalisation, nosocomial bacterial infection, hospital-acquired venous thromboembolism, and in-hospital mortality.

The CAFA challenge reports improved protein function prediction and new functional annotations for hundreds of genes through experimental screens

BACKGROUND

The Critical Assessment of Functional Annotation (CAFA) is an ongoing, global, community-driven effort to evaluate and improve the computational annotation of protein function.

RESULTS

Here, we report on the results of the third CAFA challenge, CAFA3, that featured an expanded analysis over the previous CAFA rounds, both in terms of volume of data analyzed and the types of analysis performed. In a novel and major new development, computational predictions and assessment goals drove some of the experimental assays, resulting in new functional annotations for more than 1000 genes. Specifically, we performed experimental whole-genome mutation screening in Candida albicans and Pseudomonas aureginosa genomes, which provided us with genome-wide experimental data for genes associated with biofilm formation and motility. We further performed targeted assays on selected genes in Drosophila melanogaster, which we suspected of being involved in long-term memory.

CONCLUSION

We conclude that while predictions of the molecular function and biological process annotations have slightly improved over time, those of the cellular component have not. Term-centric prediction of experimental annotations remains equally challenging; although the performance of the top methods is significantly better than the expectations set by baseline methods in C. albicans and D. melanogaster, it leaves considerable room and need for improvement. Finally, we report that the CAFA community now involves a broad range of participants with expertise in bioinformatics, biological experimentation, biocuration, and bio-ontologies, working together to improve functional annotation, computational function prediction, and our ability to manage big data in the era of large experimental screens.

Using deep maxout neural networks to improve the accuracy of function prediction from protein interaction networks

Protein-protein interaction network data provides valuable information that infers direct links between genes and their biological roles. This information brings a fundamental hypothesis for protein function prediction that interacting proteins tend to have similar functions. With the help of recently-developed network embedding feature generation methods and deep maxout neural networks, it is possible to extract functional representations that encode direct links between protein-protein interactions information and protein function. Our novel method, STRING2GO, successfully adopts deep maxout neural networks to learn functional representations simultaneously encoding both protein-protein interactions and functional predictive information. The experimental results show that STRING2GO outperforms other protein-protein interaction network-based prediction methods and one benchmark method adopted in a recent large scale protein function prediction competition.

Predicting human protein function with multi-task deep neural networks

Machine learning methods for protein function prediction are urgently needed, especially now that a substantial fraction of known sequences remains unannotated despite the extensive use of functional assignments based on sequence similarity. One major bottleneck supervised learning faces in protein function prediction is the structured, multi-label nature of the problem, because biological roles are represented by lists of terms from hierarchically organised controlled vocabularies such as the Gene Ontology. In this work, we build on recent developments in the area of deep learning and investigate the usefulness of multi-task deep neural networks (MTDNN), which consist of upstream shared layers upon which are stacked in parallel as many independent modules (additional hidden layers with their own output units) as the number of output GO terms (the tasks). MTDNN learns individual tasks partially using shared representations and partially from task-specific characteristics. When no close homologues with experimentally validated functions can be identified, MTDNN gives more accurate predictions than baseline methods based on annotation frequencies in public databases or homology transfers. More importantly, the results show that MTDNN binary classification accuracy is higher than alternative machine learning-based methods that do not exploit commonalities and differences among prediction tasks. Interestingly, compared with a single-task predictor, the performance improvement is not linearly correlated with the number of tasks in MTDNN, but medium size models provide more improvement in our case. One of advantages of MTDNN is that given a set of features, there is no requirement for MTDNN to have a bootstrap feature selection procedure as what traditional machine learning algorithms do. Overall, the results indicate that the proposed MTDNN algorithm improves the performance of protein function prediction. On the other hand, there is still large room for deep learning techniques to further enhance prediction ability.

Distinct gene expression program dynamics during erythropoiesis from human induced pluripotent stem cells compared with adult and cord blood progenitors

BACKGROUND

Human-induced pluripotent stem cells (hiPSCs) are a potentially invaluable resource for regenerative medicine, including the in vitro manufacture of blood products. HiPSC-derived red blood cells are an attractive therapeutic option in hematology, yet exhibit unexplained proliferation and enucleation defects that presently preclude such applications. We hypothesised that substantial differential regulation of gene expression during erythroid development accounts for these important differences between hiPSC-derived cells and those from adult or cord-blood progenitors. We thus cultured erythroblasts from each source for transcriptomic analysis to investigate differential gene expression underlying these functional defects.

RESULTS

Our high resolution transcriptional view of definitive erythropoiesis captures the regulation of genes relevant to cell-cycle control and confers statistical power to deploy novel bioinformatics methods. Whilst the dynamics of erythroid program elaboration from adult and cord blood progenitors were very similar, the emerging erythroid transcriptome in hiPSCs revealed radically different program elaboration compared to adult and cord blood cells. We explored the function of differentially expressed genes in hiPSC-specific clusters defined by our novel tunable clustering algorithms (SMART and Bi-CoPaM). HiPSCs show reduced expression of c-KIT and key erythroid transcription factors SOX6, MYB and BCL11A, strong HBZ-induction, and aberrant expression of genes involved in protein degradation, lysosomal clearance and cell-cycle regulation.

CONCLUSIONS

Together, these data suggest that hiPSC-derived cells may be specified to a primitive erythroid fate, and implies that definitive specification may more accurately reflect adult development. We have therefore identified, for the first time, distinct gene expression dynamics during erythroblast differentiation from hiPSCs which may cause reduced proliferation and enucleation of hiPSC-derived erythroid cells. The data suggest several mechanistic defects which may partially explain the observed aberrant erythroid differentiation from hiPSCs.

UNCLES: method for the identification of genes differentially consistently co-expressed in a specific subset of datasets

Background

Collective analysis of the increasingly emerging gene expression datasets are required. The recently proposed binarisation of consensus partition matrices (Bi-CoPaM) method can combine clustering results from multiple datasets to identify the subsets of genes which are consistently co-expressed in all of the provided datasets in a tuneable manner. However, results validation and parameter setting are issues that complicate the design of such methods. Moreover, although it is a common practice to test methods by application to synthetic datasets, the mathematical models used to synthesise such datasets are usually based on approximations which may not always be sufficiently representative of real datasets.

Results

Here, we propose an unsupervised method for the unification of clustering results from multiple datasets using external specifications (UNCLES). This method has the ability to identify the subsets of genes consistently co-expressed in a subset of datasets while being poorly co-expressed in another subset of datasets, and to identify the subsets of genes consistently co-expressed in all given datasets. We also propose the M-N scatter plots validation technique and adopt it to set the parameters of UNCLES, such as the number of clusters, automatically. Additionally, we propose an approach for the synthesis of gene expression datasets using real data profiles in a way which combines the ground-truth-knowledge of synthetic data and the realistic expression values of real data, and therefore overcomes the problem of faithfulness of synthetic expression data modelling. By application to those datasets, we validate UNCLES while comparing it with other conventional clustering methods, and of particular relevance, biclustering methods. We further validate UNCLES by application to a set of 14 real genome-wide yeast datasets as it produces focused clusters that conform well to known biological facts. Furthermore, in-silico-based hypotheses regarding the function of a few previously unknown genes in those focused clusters are drawn.

Conclusions

The UNCLES method, the M-N scatter plots technique, and the expression data synthesis approach will have wide application for the comprehensive analysis of genomic and other sources of multiple complex biological datasets. Moreover, the derived in-silico-based biological hypotheses represent subjects for future functional studies.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-015-0614-0) contains supplementary material, which is available to authorized users.

Comprehensive analysis of forty yeast microarray datasets reveals a novel subset of genes (APha-RiB) consistently negatively associated with ribosome biogenesis

BACKGROUND

The scale and complexity of genomic data lend themselves to analysis using sophisticated mathematical techniques to yield information that can generate new hypotheses and so guide further experimental investigations. An ensemble clustering method has the ability to perform consensus clustering over the same set of genes from different microarray datasets by combining results from different clustering methods into a single consensus result.

RESULTS

In this paper we have performed comprehensive analysis of forty yeast microarray datasets. One recently described Bi-CoPaM method can analyse expressions of the same set of genes from various microarray datasets while using different clustering methods, and then combine these results into a single consensus result whose clusters' tightness is tunable from tight, specific clusters to wide, overlapping clusters. This has been adopted in a novel way over genome-wide data from forty yeast microarray datasets to discover two clusters of genes that are consistently co-expressed over all of these datasets from different biological contexts and various experimental conditions. Most strikingly, average expression profiles of those clusters are consistently negatively correlated in all of the forty datasets while neither profile leads or lags the other.

CONCLUSIONS

The first cluster is enriched with ribosomal biogenesis genes. The biological processes of most of the genes in the second cluster are either unknown or apparently unrelated although they show high connectivity in protein-protein and genetic interaction networks. Therefore, it is possible that this mostly uncharacterised cluster and the ribosomal biogenesis cluster are transcriptionally oppositely regulated by some common machinery. Moreover, we anticipate that the genes included in this previously unknown cluster participate in generic, in contrast to specific, stress response processes. These novel findings illuminate coordinated gene expression in yeast and suggest several hypotheses for future experimental functional work. Additionally, we have demonstrated the usefulness of the Bi-CoPaM-based approach, which may be helpful for the analysis of other groups of (microarray) datasets from other species and systems for the exploration of global genetic co-expression.

Noise Resistant Generalized Parametric Validity Index of Clustering for Gene Expression Data

Validity indices have been investigated for decades. However, since there is no study of noise-resistance performance of these indices in the literature, there is no guideline for determining the best clustering in noisy data sets, especially microarray data sets. In this paper, we propose a generalized parametric validity (GPV) index which employs two tunable parameters α and β to control the proportions of objects being considered to calculate the dissimilarities. The greatest advantage of the proposed GPV index is its noise-resistance ability, which results from the flexibility of tuning the parameters. Several rules are set to guide the selection of parameter values. To illustrate the noise-resistance performance of the proposed index, we evaluate the GPV index for assessing five clustering algorithms in two gene expression data simulation models with different noise levels and compare the ability of determining the number of clusters with eight existing indices. We also test the GPV in three groups of real gene expression data sets. The experimental results suggest that the proposed GPV index has superior noise-resistance ability and provides fairly accurate judgements.

SMART: unique splitting-while-merging framework for gene clustering

Successful clustering algorithms are highly dependent on parameter settings. The clustering performance degrades significantly unless parameters are properly set, and yet, it is difficult to set these parameters a priori. To address this issue, in this paper, we propose a unique splitting-while-merging clustering framework, named "splitting merging awareness tactics" (SMART), which does not require any a priori knowledge of either the number of clusters or even the possible range of this number. Unlike existing self-splitting algorithms, which over-cluster the dataset to a large number of clusters and then merge some similar clusters, our framework has the ability to split and merge clusters automatically during the process and produces the the most reliable clustering results, by intrinsically integrating many clustering techniques and tasks. The SMART framework is implemented with two distinct clustering paradigms in two algorithms: competitive learning and finite mixture model. Nevertheless, within the proposed SMART framework, many other algorithms can be derived for different clustering paradigms. The minimum message length algorithm is integrated into the framework as the clustering selection criterion. The usefulness of the SMART framework and its algorithms is tested in demonstration datasets and simulated gene expression datasets. Moreover, two real microarray gene expression datasets are studied using this approach. Based on the performance of many metrics, all numerical results show that SMART is superior to compared existing self-splitting algorithms and traditional algorithms. Three main properties of the proposed SMART framework are summarized as: (1) needing no parameters dependent on the respective dataset or a priori knowledge about the datasets, (2) extendible to many different applications, (3) offering superior performance compared with counterpart algorithms.

Paradigm of tunable clustering using Binarization of Consensus Partition Matrices (Bi-CoPaM) for gene discovery

Clustering analysis has a growing role in the study of co-expressed genes for gene discovery. Conventional binary and fuzzy clustering do not embrace the biological reality that some genes may be irrelevant for a problem and not be assigned to a cluster, while other genes may participate in several biological functions and should simultaneously belong to multiple clusters. Also, these algorithms cannot generate tight clusters that focus on their cores or wide clusters that overlap and contain all possibly relevant genes. In this paper, a new clustering paradigm is proposed. In this paradigm, all three eventualities of a gene being exclusively assigned to a single cluster, being assigned to multiple clusters, and being not assigned to any cluster are possible. These possibilities are realised through the primary novelty of the introduction of tunable binarization techniques. Results from multiple clustering experiments are aggregated to generate one fuzzy consensus partition matrix (CoPaM), which is then binarized to obtain the final binary partitions. This is referred to as Binarization of Consensus Partition Matrices (Bi-CoPaM). The method has been tested with a set of synthetic datasets and a set of five real yeast cell-cycle datasets. The results demonstrate its validity in generating relevant tight, wide, and complementary clusters that can meet requirements of different gene discovery studies.

Yeast gene CMR1/YDL156W is consistently co-expressed with genes participating in DNA-metabolic processes in a variety of stringent clustering experiments

The binarization of consensus partition matrices (Bi-CoPaM) method has, among its unique features, the ability to perform ensemble clustering over the same set of genes from multiple microarray datasets by using various clustering methods in order to generate tunable tight clusters. Therefore, we have used the Bi-CoPaM method to the most synchronized 500 cell-cycle-regulated yeast genes from different microarray datasets to produce four tight, specific and exclusive clusters of co-expressed genes. We found 19 genes formed the tightest of the four clusters and this included the gene CMR1/YDL156W, which was an uncharacterized gene at the time of our investigations. Two very recent proteomic and biochemical studies have independently revealed many facets of CMR1 protein, although the precise functions of the protein remain to be elucidated. Our computational results complement these biological results and add more evidence to their recent findings of CMR1 as potentially participating in many of the DNA-metabolism processes such as replication, repair and transcription. Interestingly, our results demonstrate the close co-expressions of CMR1 and the replication protein A (RPA), the cohesion complex and the DNA polymerases α, δ and ε, as well as suggest functional relationships between CMR1 and the respective proteins. In addition, the analysis provides further substantial evidence that the expression of the CMR1 gene could be regulated by the MBF complex. In summary, the application of a novel analytic technique in large biological datasets has provided supporting evidence for a gene of previously unknown function, further hypotheses to test, and a more general demonstration of the value of sophisticated methods to explore new large datasets now so readily generated in biological experiments.