A Clinical Decision Support System for Diabetes Patients with Deep Learning: Experience of a Taiwan Medical Center.

Background: Diabetes mellitus (DM) is a major public health problem worldwide. It involves dysfunction of blood sugar regulation resulting from insulin resistance, inadequate insulin secretion, or excessive glucagon secretion. Methods: This study collated 971,401 drug usage records of 51,009 DM patients. These data include patient identification code, age, gender, outpatient visiting dates, visiting code, medication features (included items, doses, and frequencies of drugs), HbA1c results, and testing time. We apply a random forest (RF) model for feature selection and implement a regression model with the bidirectional long short-term memory (Bi-LSTM) deep learning architecture. Finally, we use the root mean square error (RMSE) as the evaluation index for the prediction model. Results: After data cleaning, the data included 8,729 male and 9,115 female cases. Metformin was the most important feature suggested by the RF model, followed by glimepiride, acarbose, pioglitazone, glibenclamide, gliclazide, repaglinide, nateglinide, sitagliptin, and vildagliptin. The model performed better with the past two seasons in the training data than with additional seasons. Further, the Bi-LSTM architecture model performed better than support vector machines (SVMs). Discussion & Conclusion: This study found that Bi-LSTM models is a well kernel in a CDSS which help physicians' decision-making, and the increasing the number of seasons will negative impact the performance. In addition, this study found that the most important drug is metformin, which is recommended as first-line treatment OHA in various situations for DM patients.

Liver cancer prediction in a viral hepatitis cohort: A deep learning approach.

Viral hepatitis is the primary cause of liver diseases, among which liver cancer is the leading cause of death from cancer. However, this cancer is often diagnosed in the later stages, which makes treatment difficult or even impossible. This study applied deep learning (DL) models for the early prediction of liver cancer in a hepatitis cohort. In this study, we surveyed 1 million random samples from the National Health Insurance Research Database (NHIRD) to analyze viral hepatitis patients from 2002 to 2010. Then, we used DL models to predict liver cancer cases based on the history of diseases of the hepatitis cohort. Our results revealed the annual prevalence of hepatitis in Taiwan increased from 2002 to 2010, with an average annual percentage change (AAPC) of 5.8% (95% CI: 4.2-7.4). However, young people (aged 16-30 years) exhibited a decreasing trend, with an AAPC of -5.6 (95% CI: -8.1 to -2.9). The results of applying DL models showed that the convolution neural network (CNN) model yielded the best performance in terms of predicting liver cancer cases, with an accuracy of 0.980 (AUC: 0.886). In conclusion, this study showed an increasing trend in the annual prevalence of hepatitis, but a decreasing trend in young people from 2002 to 2010 in Taiwan. The CNN model may be applied to predict liver cancer in a hepatitis cohort with high accuracy.

Crowdsourced estimation of cognitive decline and resilience in Alzheimer's disease.

Identifying accurate biomarkers of cognitive decline is essential for advancing early diagnosis and prevention therapies in Alzheimer's disease. The Alzheimer's disease DREAM Challenge was designed as a computational crowdsourced project to benchmark the current state-of-the-art in predicting cognitive outcomes in Alzheimer's disease based on high dimensional, publicly available genetic and structural imaging data. This meta-analysis failed to identify a meaningful predictor developed from either data modality, suggesting that alternate approaches should be considered for prediction of cognitive performance.

Identification of cucurbitacins and assembly of a draft genome for Aquilaria agallocha.

BACKGROUND: Agarwood is derived from Aquilaria trees, the trade of which has come under strict control with a listing in Appendix II of the Convention on International Trade in Endangered Species of Wild Fauna and Flora. Many secondary metabolites of agarwood are known to have medicinal value to humans, including compounds that have been shown to elicit sedative effects and exhibit anti-cancer properties. However, little is known about the genome, transcriptome, and the biosynthetic pathways responsible for producing such secondary metabolites in agarwood. RESULTS: In this study, we present a draft genome and a putative pathway for cucurbitacin E and I, compounds with known medicinal value, from in vitro Aquilaria agallocha agarwood. DNA and RNA data are utilized to annotate many genes and protein functions in the draft genome. The expression changes for cucurbitacin E and I are shown to be consistent with known responses of A. agallocha to biotic stress and a set of homologous genes in Arabidopsis thaliana related to cucurbitacin bio-synthesis is presented and validated through qRT-PCR. CONCLUSIONS: This study is the first attempt to identify cucurbitacin E and I from in vitro agarwood and the first draft genome for any species of Aquilaria. The results of this study will aid in future investigations of secondary metabolite pathways in Aquilaria and other non-model medicinal plants.

DBD2BS: connecting a DNA-binding protein with its binding sites.

By binding to short and highly conserved DNA sequences in genomes, DNA-binding proteins initiate, enhance or repress biological processes. Accurately identifying such binding sites, often represented by position weight matrices (PWMs), is an important step in understanding the control mechanisms of cells. When given coordinates of a DNA-binding domain (DBD) bound with DNA, a potential function can be used to estimate the change of binding affinity after base substitutions, where the changes can be summarized as a PWM. This technique provides an effective alternative when the chromatin immunoprecipitation data are unavailable for PWM inference. To facilitate the procedure of predicting PWMs based on protein-DNA complexes or even structures of the unbound state, the web server, DBD2BS, is presented in this study. The DBD2BS uses an atom-level knowledge-based potential function to predict PWMs characterizing the sequences to which the query DBD structure can bind. For unbound queries, a list of 1066 DBD-DNA complexes (including 1813 protein chains) is compiled for use as templates for synthesizing bound structures. The DBD2BS provides users with an easy-to-use interface for visualizing the PWMs predicted based on different templates and the spatial relationships of the query protein, the DBDs and the DNAs. The DBD2BS is the first attempt to predict PWMs of DBDs from unbound structures rather than from bound ones. This approach increases the number of existing protein structures that can be exploited when analyzing protein-DNA interactions. In a recent study, the authors showed that the kernel adopted by the DBD2BS can generate PWMs consistent with those obtained from the experimental data. The use of DBD2BS to predict PWMs can be incorporated with sequence-based methods to discover binding sites in genome-wide studies. Available at: http://dbd2bs.csie.ntu.edu.tw/, http://dbd2bs.csbb.ntu.edu.tw/, and http://dbd2bs.ee.ncku.edu.tw.

Diseases That Occur Prior to Spontaneous Intracerebral Hemorrhage: Identification of Predisposing and Risk Factors Using Lag Sequential Analysis.

Spontaneous intracerebral hemorrhage (sICH) has many predisposing/risk factors. Lag sequential analysis (LSA) is a method of analyzing sequential patterns and their associations within categorical data in different system states. The results of this study will assist in preventing sICH and improving the patient outcome after sICH. The correlations between a first sICH and previous clinic visits were examined using LSA with data obtained from the Taiwan National Health Insurance Research Database (NHIRD). In this study, LSA was employed to examine the data in the Taiwan NHIRD in order to identify predisposing and risk factors related to sICH, and the results increased our knowledge of the temporal relationships between diseases. This study employed LSA to identify predisposing/risk factors prior to the first occurrence of sICH using a healthcare administrative database in Taiwan. The data were managed using the clinical classification software (CCS). All cases of traumatic ICH were excluded. Ten disease groups were identified using CCS. Hypertension and dizziness/vertigo were identified as two important predisposing/risk factors for sICH, and early treatment of hypertension resulted in a greater survival rate. Five disease groups were found to have occurred prior to other diseases and affected mostly the elderly, resulting in subsequent sICH. The results of this study also showed that nutritional status and tooth health were highly associated with the occurrence of sICH owing to a poor state of the digestive system. In conclusion, there are many diseases that influence the risk of a subsequent sICH. This study demonstrated that LSA is a very useful tool for future study of healthcare administrative databases.

seeMotif: exploring and visualizing sequence motifs in 3D structures.

Sequence motifs are important in the study of molecular biology. Motif discovery tools efficiently deliver many function related signatures of proteins and largely facilitate sequence annotation. As increasing numbers of motifs are detected experimentally or predicted computationally, characterizing the functional roles of motifs and identifying the potential synergetic relationships between them are important next steps. A good way to investigate novel motifs is to utilize the abundant 3D structures that have also been accumulated at an astounding rate in recent years. This article reports the development of the web service seeMotif, which provides users with an interactive interface for visualizing sequence motifs on protein structures from the Protein Data Bank (PDB). Researchers can quickly see the locations and conformation of multiple motifs among a number of related structures simultaneously. Considering the fact that PDB sequences are usually shorter than those in sequence databases and/or may have missing residues, seeMotif has two complementary approaches for selecting structures and mapping motifs to protein chains in structures. As more and more structures belonging to previously uncharacterized protein families become available, combining sequence and structure information gives good opportunities to facilitate understanding of protein functions in large-scale genome projects. Available at: http://seemotif.csie.ntu.edu.tw,http://seemotif.ee.ncku.edu.tw or http://seemotif.csbb.ntu.edu.tw.

E1DS: catalytic site prediction based on 1D signatures of concurrent conservation.

Large-scale automatic annotation of protein sequences remains challenging in postgenomics era. E1DS is designed for annotating enzyme sequences based on a repository of 1D signatures. The employed sequence signatures are derived using a novel pattern mining approach that discovers long motifs consisted of several sequential blocks (conserved segments). Each of the sequential blocks is considerably conserved among the protein members of an EC group. Moreover, a signature includes at least three sequential blocks that are concurrently conserved, i.e. frequently observed together in sequences. In other words, a sequence signature is consisted of residues from multiple regions of the protein sequence, which echoes the observation that an enzyme catalytic site is usually constituted of residues that are largely separated in the sequence. E1DS currently contains 5421 sequence signatures that in total cover 932 4-digital EC numbers. E1DS is evaluated based on a collection of enzymes with catalytic sites annotated in Catalytic Site Atlas. When compared to the famous pattern database PROSITE, predictions based on E1DS signatures are considered more sensitive in identifying catalytic sites and the involved residues. E1DS is available at http://e1ds.ee.ncku.edu.tw/ and a mirror site can be found at http://e1ds.csbb.ntu.edu.tw/.

Exploration of Medical Trajectories of Stroke Patients Based on Group-Based Trajectory Modeling.

A high mortality rate is an issue with acute cerebrovascular disease (ACVD), as it often leads to a high medical expenditure, and in particular to high costs of treatment for emergency medical conditions and critical care. In this study, we used group-based trajectory modeling (GBTM) to study the characteristics of various groups of patients hospitalized with ACVD. In this research, the patient data were derived from the 1 million sampled cases in the National Health Insurance Research Database (NHIRD) in Taiwan. Cases who had been admitted to hospitals fewer than four times or more than eight times were excluded. Characteristics of the ACVD patients were collected, including age, mortality rate, medical expenditure, and length of hospital stay for each admission. We then performed GBTM to examine hospitalization patterns in patients who had been hospitalized more than four times and fewer than or equal to eight times. The patients were divided into three groups according to medical expenditure: high, medium, and low groups, split at the 33rd and 66th percentiles. After exclusion of unqualified patients, a total of 27,264 cases (male/female = 15,972/11,392) were included. Analysis of the characteristics of the ACVD patients showed that there were significant differences between the two gender groups in terms of age, mortality rate, medical expenditure, and total length of hospital stay. In addition, the data were compared between two admissions, which included interval, outpatient department (OPD) visit after discharge, OPD visit after hospital discharge, and OPD cost. Finally, the differences in medical expenditure between genders and between patients with different types of stroke-ischemic stroke, spontaneous intracerebral hemorrhage (sICH), and subarachnoid hemorrhage (SAH)-were examined using GBTM. Overall, this study employed GBTM to examine the trends in medical expenditure for different groups of stroke patients at different admissions, and some important results were obtained. Our results demonstrated that the time interval between subsequent hospitalizations decreased in the ACVD patients, and there were significant differences between genders and between patients with different types of stroke. It is often difficult to decide when the time has been reached at which further treatment will not improve the condition of ACVD patients, and the findings of our study may be used as a reference for assessing outcomes and quality of care for stroke patients. Because of the characteristics of NHIRD, this study had some limitations; for example, the number of cases for some diseases was not sufficient for effective statistical analysis.

Differences in Spontaneous Intracerebral Hemorrhage Cases between Urban and Rural Regions of Taiwan: Big Data Analytics of Government Open Data.

This study evaluated the differences in spontaneous intracerebral hemorrhage (sICH) between rural and urban areas of Taiwan with big data analysis. We used big data analytics and visualization tools to examine government open data, which included the residents' health medical administrative data, economic status, educational status, and relevant information. The study subjects included sICH patients of Taipei region (29,741 cases) and Eastern Taiwan (4565 cases). The incidence of sICH per 100,000 population per year in Eastern Taiwan (71.3 cases) was significantly higher than that of the Taipei region (42.3 cases). The mean coverage area per hospital in Eastern Taiwan (452.4 km²) was significantly larger than the Taipei region (24 km²). The residents educational level in the Taipei region was significantly higher than that in Eastern Taiwan. The mean hospital length of stay in the Taipei region (17.9 days) was significantly greater than that in Eastern Taiwan (16.3 days) (p < 0.001). There were no significant differences in other medical profiles between two areas. Distance and educational barriers were two possible reasons for the higher incidence of sICH in the rural area of Eastern Taiwan. Further studies are necessary in order to understand these phenomena in greater depth.

Does the Short-Term Effect of Air Pollution Influence the Incidence of Spontaneous Intracerebral Hemorrhage in Different Patient Groups? Big Data Analysis in Taiwan.

Spontaneous intracerebral hemorrhage (sICH) has a high mortality rate. Research has demonstrated that the occurrence of sICH is related to air pollution. This study used big data analysis to explore the impact of air pollution on the risk of sICH in patients of differing age and geographic location. 39,053 cases were included in this study; 14,041 in the Taipei region (Taipei City and New Taipei City), 5537 in Taoyuan City, 7654 in Taichung City, 4739 in Tainan City, and 7082 in Kaohsiung City. The results of correlation analysis indicated that there were two pollutants groups, the CO and NO2 group and the PM2.5 and PM10 group. Furthermore, variations in the correlations of sICH with air pollutants were identified in different age groups. The co-factors of the influence of air pollutants in the different age groups were explored using regression analysis. This study integrated Taiwan National Health Insurance data and air pollution data to explore the risk factors of sICH using big data analytics. We found that PM2.5 and PM10 are very important risk factors for sICH, and age is an important modulating factor that allows air pollutants to influence the incidence of sICH.

Analysis of new functional profiles of protein isoforms yielded by ds exonization in rice.

Insertion of transposable elements (TEs) into introns can lead to their activation as alternatively spliced cassette exons, an event called exonization. Exonization can enrich the complexity of transcriptomes and proteomes. Previously, we performed a genome-wide computational analysis of Ds exonization events in the monocot Oryza sativa (rice). The insertion patterns of Ds increased the number of transcripts and subsequent protein isoforms, which were determined as interior and C-terminal variants. In this study, these variants were scanned with the PROSITE database in order to identify new functional profiles (domains) that were referred to their reference proteins. The new profiles of the variants were expected to be beneficial for a selective advantage and more than 70% variants achieved this. The new functional profiles could be contributed by an exon-intron junction, an intron alone, an intron-TE junction, or a TE alone. A Ds-inserted intron may yield 167 new profiles on average, while some cases can yield thousands of new profiles, of which C-terminal variants were in major. Additionally, more than 90% of the TE-inserted genes were found to gain novel functional profiles in each intron via exonization. Therefore, new functional profiles yielded by the exonization may occur in many local regions of the reference protein.

Predicting target DNA sequences of DNA-binding proteins based on unbound structures.

DNA-binding proteins such as transcription factors use DNA-binding domains (DBDs) to bind to specific sequences in the genome to initiate many important biological functions. Accurate prediction of such target sequences, often represented by position weight matrices (PWMs), is an important step to understand many biological processes. Recent studies have shown that knowledge-based potential functions can be applied on protein-DNA co-crystallized structures to generate PWMs that are considerably consistent with experimental data. However, this success has not been extended to DNA-binding proteins lacking co-crystallized structures. This study aims at investigating the possibility of predicting the DNA sequences bound by DNA-binding proteins from the proteins' unbound structures (structures of the unbound state). Given an unbound query protein and a template complex, the proposed method first employs structure alignment to generate synthetic protein-DNA complexes for the query protein. Once a complex is available, an atomic-level knowledge-based potential function is employed to predict PWMs characterizing the sequences to which the query protein can bind. The evaluation of the proposed method is based on seven DNA-binding proteins, which have structures of both DNA-bound and unbound forms for prediction as well as annotated PWMs for validation. Since this work is the first attempt to predict target sequences of DNA-binding proteins from their unbound structures, three types of structural variations that presumably influence the prediction accuracy were examined and discussed. Based on the analyses conducted in this study, the conformational change of proteins upon binding DNA was shown to be the key factor. This study sheds light on the challenge of predicting the target DNA sequences of a protein lacking co-crystallized structures, which encourages more efforts on the structure alignment-based approaches in addition to docking- and homology modeling-based approaches for generating synthetic complexes.

Discovery of genes related to insecticide resistance in Bactrocera dorsalis by functional genomic analysis of a de novo assembled transcriptome.

Insecticide resistance has recently become a critical concern for control of many insect pest species. Genome sequencing and global quantization of gene expression through analysis of the transcriptome can provide useful information relevant to this challenging problem. The oriental fruit fly, Bactrocera dorsalis, is one of the world's most destructive agricultural pests, and recently it has been used as a target for studies of genetic mechanisms related to insecticide resistance. However, prior to this study, the molecular data available for this species was largely limited to genes identified through homology. To provide a broader pool of gene sequences of potential interest with regard to insecticide resistance, this study uses whole transcriptome analysis developed through de novo assembly of short reads generated by next-generation sequencing (NGS). The transcriptome of B. dorsalis was initially constructed using Illumina's Solexa sequencing technology. Qualified reads were assembled into contigs and potential splicing variants (isotigs). A total of 29,067 isotigs have putative homologues in the non-redundant (nr) protein database from NCBI, and 11,073 of these correspond to distinct D. melanogaster proteins in the RefSeq database. Approximately 5,546 isotigs contain coding sequences that are at least 80% complete and appear to represent B. dorsalis genes. We observed a strong correlation between the completeness of the assembled sequences and the expression intensity of the transcripts. The assembled sequences were also used to identify large numbers of genes potentially belonging to families related to insecticide resistance. A total of 90 P450-, 42 GST-and 37 COE-related genes, representing three major enzyme families involved in insecticide metabolism and resistance, were identified. In addition, 36 isotigs were discovered to contain target site sequences related to four classes of resistance genes. Identified sequence motifs were also analyzed to characterize putative polypeptide translational products and associate them with specific genes and protein functions.