LAD: Layer-Wise Adaptive Distillation for BERT Model Compression.

Recent advances with large-scale pre-trained language models (e.g., BERT) have brought significant potential to natural language processing. However, the large model size hinders their use in IoT and edge devices. Several studies have utilized task-specific knowledge distillation to compress the pre-trained language models. However, to reduce the number of layers in a large model, a sound strategy for distilling knowledge to a student model with fewer layers than the teacher model is lacking. In this work, we present Layer-wise Adaptive Distillation (LAD), a task-specific distillation framework that can be used to reduce the model size of BERT. We design an iterative aggregation mechanism with multiple gate blocks in LAD to adaptively distill layer-wise internal knowledge from the teacher model to the student model. The proposed method enables an effective knowledge transfer process for a student model, without skipping any teacher layers. The experimental results show that both the six-layer and four-layer LAD student models outperform previous task-specific distillation approaches during GLUE tasks.

HAT4RD: Hierarchical Adversarial Training for Rumor Detection in Social Media.

With the development of social media, social communication has changed. While this facilitates people's communication and access to information, it also provides an ideal platform for spreading rumors. In normal or critical situations, rumors can affect people's judgment and even endanger social security. However, natural language is high-dimensional and sparse, and the same rumor may be expressed in hundreds of ways on social media. As such, the robustness and generalization of the current rumor detection model are in question. We proposed a novel hierarchical adversarial training method for rumor detection (HAT4RD) on social media. Specifically, HAT4RD is based on gradient ascent by adding adversarial perturbations to the embedding layers of post-level and event-level modules to deceive the detector. At the same time, the detector uses stochastic gradient descent to minimize the adversarial risk to learn a more robust model. In this way, the post-level and event-level sample spaces are enhanced, and we verified the robustness of our model under a variety of adversarial attacks. Moreover, visual experiments indicate that the proposed model drifts into an area with a flat loss landscape, thereby, leading to better generalization. We evaluate our proposed method on three public rumor datasets from two commonly used social platforms (Twitter and Weibo). Our experimental results demonstrate that our model achieved better results compared with the state-of-the-art methods.

Lithium concentration and recurrence risk during maintenance treatment of bipolar disorder: Multicenter cohort and meta-analysis.

OBJECTIVE: To compare differences in efficacy during maintenance treatment for bipolar disorder (BD) according to lithium serum levels. A multicenter retrospective cohort study and a dose-response meta-analysis were conducted. METHODS: The cohort study was conducted in Taiwan from 2001 to 2019 to identify patients with euthymic BD according to different serum levels (<0.4, 0.4-0.8, and 0.8-1.2 mmol/L). We adopted adjusted hazard ratios (aHRs) with 95% confidence intervals (CIs) for time to the recurrence of mood episodes having the <0.4 mmol/L group as the reference group. Moreover, we systematically searched for related articles in major databases before January 31, 2021 (PROSPERO: CRD42021235812). We used random-effects modeling to estimate the dose-response relationships between lithium serum levels and recurrence of mood episodes, which were depicted as odds ratios (ORs) with 95% CIs. RESULTS: A total of 1406 participants (cohort: 466; meta-analysis: 940) were included. In the cohort study, the 0.4-0.8 mmol/L group was associated with a significantly lower risk of recurrences (aHR: 0.75), while the 0.8-1.2 mmol/L group had a lower risk without statistical significance (aHR: 0.77). The dose-response meta-analysis showed that with the increase in lithium serum levels, the risk decreased (linear model OR: 0.85, for every 0.1 mmol/L increase; non-linear model OR: 1.00 at 0.0 mmol/L, 0.42 at 0.4 mmol/L, and 0.27 at 0.8 mmol/L). CONCLUSION: Although confounding by indication cannot be excluded, the combined results suggest a significant preventative effect on the recurrence of major affective episodes among those with serum levels of 0.4-0.8 mmol/L.

Month of birth and mental disorders: A population-based study and validation using global meta-analysis.

OBJECTIVE: Month of birth (MOB) is associated with specified mental disorders (MDs). However, whether these relationships extend to all MDs remains unclear. We investigate the association using a population-based cohort study and a meta-analysis. METHODS: First, we examined patients with 34 DSM-5-classified MDs in the Taiwan national database. We estimated the relative risk ratios (RR) of each illness in each MOB relative to that in the general population and assessed the periodicity, with six further sensitivity analyses. Second, we searched PubMed, Embase, and Cochrane for related articles through 31 December 2020. We used a random-effects model, pooled RRs with 95% confidence intervals of each MOB from the identified studies, and transformed them from MOB to relative age in a year or season. RESULTS: The cohort included 1,951,777 patients. Except for posttraumatic stress disorder, dissociative disorders, feeding/eating disorders, gender dysphoria, and paraphilic disorders, the other MDs had significant MOB periodicity. The meta-analysis included 51 studies investigating 10 MDs. The youngest age at the start of school owing to MOB was associated with the highest RRs of intellectual disability (1.13), autism (1.05), attention-deficit/hyperactivity disorder (1.13). Winter births had significant risks of schizophrenia (1.04), bipolar I disorder (1.02), and major depressive disorder (1.01), and autumn births had a significant risk of alcohol use disorder (1.02). No significant associations between season of birth and Alzheimer's disease, or eating disorders were found. CONCLUSIONS: MOB is related to the risks of certain MDs. This finding provides a reference for future research on the etiology of MDs.

PubTator: a web-based text mining tool for assisting biocuration.

Manually curating knowledge from biomedical literature into structured databases is highly expensive and time-consuming, making it difficult to keep pace with the rapid growth of the literature. There is therefore a pressing need to assist biocuration with automated text mining tools. Here, we describe PubTator, a web-based system for assisting biocuration. PubTator is different from the few existing tools by featuring a PubMed-like interface, which many biocurators find familiar, and being equipped with multiple challenge-winning text mining algorithms to ensure the quality of its automatic results. Through a formal evaluation with two external user groups, PubTator was shown to be capable of improving both the efficiency and accuracy of manual curation. PubTator is publicly available at http://www.ncbi.nlm.nih.gov/CBBresearch/Lu/Demo/PubTator/.

tmVar: a text mining approach for extracting sequence variants in biomedical literature.

MOTIVATION: Text-mining mutation information from the literature becomes a critical part of the bioinformatics approach for the analysis and interpretation of sequence variations in complex diseases in the post-genomic era. It has also been used for assisting the creation of disease-related mutation databases. Most of existing approaches are rule-based and focus on limited types of sequence variations, such as protein point mutations. Thus, extending their extraction scope requires significant manual efforts in examining new instances and developing corresponding rules. As such, new automatic approaches are greatly needed for extracting different kinds of mutations with high accuracy. RESULTS: Here, we report tmVar, a text-mining approach based on conditional random field (CRF) for extracting a wide range of sequence variants described at protein, DNA and RNA levels according to a standard nomenclature developed by the Human Genome Variation Society. By doing so, we cover several important types of mutations that were not considered in past studies. Using a novel CRF label model and feature set, our method achieves higher performance than a state-of-the-art method on both our corpus (91.4 versus 78.1% in F-measure) and their own gold standard (93.9 versus 89.4% in F-measure). These results suggest that tmVar is a high-performance method for mutation extraction from biomedical literature. AVAILABILITY: tmVar software and its corpus of 500 manually curated abstracts are available for download at http://www.ncbi.nlm.nih.gov/CBBresearch/Lu/pub/tmVar

Valproic acid monitoring: Serum prediction using a machine learning framework from multicenter real-world data.

BACKGROUND: Our study employs machine learning to predict serum valproic acid (VPA) concentrations, aiming to contribute to the development of non-invasive assays for therapeutic drug monitoring. METHODS: Medical records from 2002 to 2019 were obtained from the Taiwan Chang Gung Research Database. Using various machine learning algorithms, we developed predictive models to classify serum VPA concentrations into two categories (1-50 µg/ml or 51-100 µg/ml) and predicted the exact concentration value. The models were trained on 5142 samples and tested on 644 independent samples. Accuracy was the main metric used to evaluate model performance, with a tolerance of 20 µg/ml for continuous variables. Furthermore, we identified important features and developed simplified models with fewer features. RESULTS: The models achieved an average accuracy of 0.80-0.86 for binary outcomes and 0.72-0.88 for continuous outcome. Ten top features associated with higher serum VPA levels included higher VPA last and daily doses, bipolar disorder or schizophrenia spectrum disorder diagnoses, elevated levels of serum albumin, calcium, and creatinine, low platelet count, low percentage of segmented white blood cells, and low red cell distribution width-coefficient of variation. The simplified models had an average accuracy of 0.82-0.86 for binary outcome and 0.70-0.86 for continuous outcome. LIMITATIONS: The study's predictive model lacked external test data from outside the hospital for validation. CONCLUSIONS: Machine learning models have the potential to integrate real-world data and predict VPA concentrations, providing a promising tool for reducing the need for frequent monitoring of serum levels in clinical practice.

Augmented transitive relationships with high impact protein distillation in protein interaction prediction.

Predicting new protein-protein interactions is important for discovering novel functions of various biological pathways. Predicting these interactions is a crucial and challenging task. Moreover, discovering new protein-protein interactions through biological experiments is still difficult. Therefore, it is increasingly important to discover new protein interactions. Many studies have predicted protein-protein interactions, using biological features such as Gene Ontology (GO) functional annotations and structural domains of two proteins. In this paper, we propose an augmented transitive relationships predictor (ATRP), a new method of predicting potential protein interactions using transitive relationships and annotations of protein interactions. In addition, a distillation of virtual direct protein-protein interactions is proposed to deal with unbalanced distribution of different types of interactions in the existing protein-protein interaction databases. Our results demonstrate that ATRP can effectively predict protein-protein interactions. ATRP achieves an 81% precision, a 74% recall and a 77% F-measure in average rate in the prediction of direct protein-protein interactions. Using the generated benchmark datasets from KUPS to evaluate of all types of the protein-protein interaction, ATRP achieved a 93% precision, a 49% recall and a 64% F-measure in average rate. This article is part of a Special Issue entitled: Computational Methods for Protein Interaction and Structural Prediction.

Predicting new-onset post-stroke depression from real-world data using machine learning algorithm.

Introduction: Post-stroke depression (PSD) is a serious mental disorder after ischemic stroke. Early detection is important for clinical practice. This research aims to develop machine learning models to predict new-onset PSD using real-world data. Methods: We collected data for ischemic stroke patients from multiple medical institutions in Taiwan between 2001 and 2019. We developed models from 61,460 patients and used 15,366 independent patients to test the models' performance by evaluating their specificities and sensitivities. The predicted targets were whether PSD occurred at 30, 90, 180, and 365 days post-stroke. We ranked the important clinical features in these models. Results: In the study's database sample, 1.3% of patients were diagnosed with PSD. The average specificity and sensitivity of these four models were 0.83-0.91 and 0.30-0.48, respectively. Ten features were listed as important features related to PSD at different time points, namely old age, high height, low weight post-stroke, higher diastolic blood pressure after stroke, no pre-stroke hypertension but post-stroke hypertension (new-onset hypertension), post-stroke sleep-wake disorders, post-stroke anxiety disorders, post-stroke hemiplegia, and lower blood urea nitrogen during stroke. Discussion: Machine learning models can provide as potential predictive tools for PSD and important factors are identified to alert clinicians for early detection of depression in high-risk stroke patients.

Inference of transcriptional regulatory network by bootstrapping patterns.

MOTIVATION: Transcriptional regulatory networks, which consist of linkages between transcription factors (TF) and target genes (TGene), control the expression of a genome and play important roles in all aspects of an organism's life cycle. Accurate prediction of transcriptional regulatory networks is critical in providing useful information for biologists to determine what to do next. Currently, there is a substantial amount of fragmented gene regulation information described in the medical literature. However, current related text analysis methods designed to identify protein-protein interactions are not entirely suitable for finding transcriptional regulatory networks. RESULT: In this article, we propose an automatic regulatory network inference method that uses bootstrapping of description patterns to predict the relationship between a TF and its TGenes. The proposed method differs from other regulatory network generators in that it makes use of both positive and negative patterns for different vector combinations in a sentence. Moreover, the positive pattern learning process can be fully automatic. Furthermore, patterns for active and passive voice sentences are learned separately. The experiments use 609 HIF-1 expert-tagged articles from PubMed as the gold standard. The results show that the proposed method can automatically generate a predicted regulatory network for a transcription factor. Our system achieves an F-measure of 72.60%. AVAILABILITY: The software, training/test datasets and learned patterns are available at http://140.116.99.138/∼hcw0901/PubMedSearch.php.

Comparative effectiveness of antidepressants on geriatric depression: Real-world evidence from a population-based study.

BACKGROUND: There is little real-world evidence about effectiveness of different antidepressants on geriatric depression. METHODS: We used population-based claims data in Taiwan between 1997 and 2013 to include older patients (≥ 60 years of age) who were diagnosed with depression and started to use antidepressants. All patients were followed up until discontinuation of antidepressant use or the end of the study period. Treatment outcomes were set as the risk of switching to another antidepressant, receiving augmentation therapy, and psychiatric hospitalization. We used cox proportional hazards regression models to calculate hazard ratios with 95% confidence intervals (CIs) and adjust for several confounding factors (aHRs). RESULTS: During the study period, a total of 207,946 elderly patients with depression received one of the following 11 antidepressants: sertraline, fluoxetine, paroxetine, escitalopram, citalopram, fluvoxamine, venlafaxine, duloxetine, moclobemide, mirtazapine, and bupropion. Compared to the patients treated with sertraline, those treated with fluvoxamine / venlafaxine had significantly but modestly higher risks of switching (aHR [95% CI]: 1.16 [1.11-1.21] / 1.10 [1.06-1.14]), augmentation (1.06 [1.02-1.10] / 1.08 [1.05-1.12]), and hospitalization (1.28 [1.03-1.58] / 1.37 [1.16-1.62]). Otherwise, the remaining 9 antidepressants yielded no consistent result in the three outcomes. LIMITATIONS: This study is a multi-arm and active controlled trial, lacking a placebo group. CONCLUSION: As treating geriatric depression, no individual antidepressant posed consistently better effectiveness in the outcomes of switching antidepressant, receiving augmentation, and psychiatric hospitalization than any other one, whereas clinicians should be cautious when prescribing fluvoxamine and venlafaxine.

Differences in the prophylactic effect of serum lithium levels on depression and mania in bipolar disorder: A dose-response meta-analysis.

The optimal serum lithium levels for preventing the recurrence of mood episodes in bipolar disorder are controversial, especially when polarity is considered. The present study aimed to comprehensively examine the dose-response relationship between lithium concentration and risk of recurrence of mood episodes. We conducted a systematic search of major databases before January 2021 (PROSPERO: CRD42021235812). A one-stage, random-effects, restricted cubic splines model was used to estimate the dose-response relationship between lithium concentration and risk of recurrence of any or specific mood episodes (depression or mania). The effect size is shown as odds ratio (OR). Our meta-analysis included six randomised controlled trials with a total of 975 participants. The dose-response curve showed that increased serum concentrations were associated with a gradual decrease in the risk of any mood episodes (OR 0.50 at 0.60 mmol/l, OR 0.15 at 1.20 mmol/l). The risk of depression decreased slightly with a concentration of 0.60 mmol/l (OR 0.83) but dropped rapidly as the concentration increased to 1.20 mmol/l (OR 0.39). By contrast, the risk for mania initially decreased steadily (OR 0.44), but decreased only marginally (OR 0.30) as the concentration increased. To reduce the recurrence risk to 56%, prevention of depression required a higher concentration than that required for mania (1.13 mmol/l vs. 0.60 mmol/l). Our results suggest a negative dose-response relationship between serum lithium levels and risk of recurrence. In particular, the different preventive effects of serum concentration on depression and mania will be an important clinical reference.

Month of birth and the risk of narcolepsy: a systematic review and meta-analysis.

STUDY OBJECTIVES: The aim of this study is to evaluate the relationship between the month of birth (MOB) and the risk of narcolepsy. METHODS: We conducted a systematic review of the electronic databases PubMed, Embase, and Cochrane CENTRAL from their inception to September 30, 2021. We also added data on narcolepsy from the National Health Insurance Research Database in Taiwan. Then we extracted the relative risk (RR) ratios of narcolepsy in each month of birth to those of the general population and transformed them from MOB to season. A random-effects model was used to calculate pooled RR ratios from the meta-analysis and 95% confidence interval (CI). RESULTS: The meta-analysis analyzed 7 studies and included 3,776 patients from 8 areas (Canada, China, France, Germany, Hong Kong, Netherlands, Taiwan, and United States). The RR ratio was highest in March (1.11; 95% CI, 0.99-1.26) and August (1.11; 95% CI, 0.98-1.26) and lowest in April (0.90; 95% CI, 0.78-1.03). However, none of the MOBs reached statistical significance. Moreover, the narcolepsy risk patterns on the 3 continents (Asia, Europe, and North America) were different. In North America, the highest and lowest significant risks were found in March (1.47; 95% CI, 1.20-1.79) and September (0.75; 95% CI, 0.56-0.99). In Asia, the lowest notable risk was in April (0.80; 95% CI, 0.66-0.97). In Europe, the risk of narcolepsy was not significantly related to any MOB. In terms of seasons, only spring MOBs in North America had a significantly higher risk (1.21; 95% CI, 1.06-1.38). CONCLUSIONS: The findings indicated that the risk of narcolepsy and MOB differed across the 3 continents. This study indicates the important role of environmental factors in narcolepsy. SYSTEMATIC REVIEW REGISTRATION: Registry: PROSPERO; Identifier: CRD42020186660. CITATION: Hsu C-W, Tseng P-T, Tu Y-K, et al. Month of birth and the risk of narcolepsy: a systematic review and meta-analysis. J Clin Sleep Med. 2022;18(4):1113-1120.

Cross-species gene normalization by species inference.

BACKGROUND: To access and utilize the rich information contained in the biomedical literature, the ability to recognize and normalize gene mentions referenced in the literature is crucial. In this paper, we focus on improvements to the accuracy of gene normalization in cases where species information is not provided. Gene names are often ambiguous, in that they can refer to the genes of many species. Therefore, gene normalization is a difficult challenge. METHODS: We define "gene normalization" as a series of tasks involving several issues, including gene name recognition, species assignation and species-specific gene normalization. We propose an integrated method, GenNorm, consisting of three modules to handle the issues of this task. Every issue can affect overall performance, though the most important is species assignation. Clearly, correct identification of the species can decrease the ambiguity of orthologous genes. RESULTS: In experiments, the proposed model attained the top-1 threshold average precision (TAP-k) scores of 0.3297 (k=5), 0.3538 (k=10), and 0.3535 (k=20) when tested against 50 articles that had been selected for their difficulty and the most divergent results from pooled team submissions. In the silver-standard-507 evaluation, our TAP-k scores are 0.4591 for k=5, 10, and 20 and were ranked 2nd, 2nd, and 3rd respectively. AVAILABILITY: A web service and input, output formats of GenNorm are available at http://ikmbio.csie.ncku.edu.tw/GN/.

The gene normalization task in BioCreative III.

BACKGROUND: We report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers of the genes detected in full-text articles. For training, 32 fully and 500 partially annotated articles were prepared. A total of 507 articles were selected as the test set. Due to the high annotation cost, it was not feasible to obtain gold-standard human annotations for all test articles. Instead, we developed an Expectation Maximization (EM) algorithm approach for choosing a small number of test articles for manual annotation that were most capable of differentiating team performance. Moreover, the same algorithm was subsequently used for inferring ground truth based solely on team submissions. We report team performance on both gold standard and inferred ground truth using a newly proposed metric called Threshold Average Precision (TAP-k). RESULTS: We received a total of 37 runs from 14 different teams for the task. When evaluated using the gold-standard annotations of the 50 articles, the highest TAP-k scores were 0.3297 (k=5), 0.3538 (k=10), and 0.3535 (k=20), respectively. Higher TAP-k scores of 0.4916 (k=5, 10, 20) were observed when evaluated using the inferred ground truth over the full test set. When combining team results using machine learning, the best composite system achieved TAP-k scores of 0.3707 (k=5), 0.4311 (k=10), and 0.4477 (k=20) on the gold standard, representing improvements of 12.4%, 21.8%, and 26.6% over the best team results, respectively. CONCLUSIONS: By using full text and being species non-specific, the GN task in BioCreative III has moved closer to a real literature curation task than similar tasks in the past and presents additional challenges for the text mining community, as revealed in the overall team results. By evaluating teams using the gold standard, we show that the EM algorithm allows team submissions to be differentiated while keeping the manual annotation effort feasible. Using the inferred ground truth we show measures of comparative performance between teams. Finally, by comparing team rankings on gold standard vs. inferred ground truth, we further demonstrate that the inferred ground truth is as effective as the gold standard for detecting good team performance.

Differences in Psychiatric Comorbidities and Gender Distribution among Three Clusters of Personality Disorders: A Nationwide Population-Based Study.

Personality disorders (PDs) are grouped into clusters A, B, and C. However, whether the three clusters of PDs have differences in comorbid mental disorders or gender distribution is still lacking sufficient evidence. We aim to investigate the distribution pattern across the three clusters of PDs with a population-based cohort study. This study used the Taiwan national database between 1995 and 2013 to examine the data of patients with cluster A PDs, cluster B PDs, or cluster C PDs. We compared the differences of psychiatric comorbidities classified in the Diagnostic and Statistical Manual of Mental Disorders, fifth edition across the three clusters of PDs. Moreover, we formed gender subgroups of the three PDs to observe the discrepancy between male and female. Among the 9845 patients, those with cluster A PDs had the highest proportion of neurodevelopmental disorders, schizophrenia and neurocognitive disorders, those with cluster B PDs demonstrated the largest percentage of bipolar disorders, trauma and stressor disorders, feeding and eating disorders, and substance and addictive disorders, and those with cluster C PDs had the greatest proportion of depressive disorders, anxiety disorders, obsessive-compulsive disorders, somatic symptom disorders, and sleep-wake disorders. The gender subgroups revealed significant male predominance in neurodevelopmental disorders and female predominance in sleep-wake disorders across all three clusters of PDs. Our findings support that some psychiatric comorbidities are more prevalent in specified cluster PDs and that gender differences exist across the three clusters of PDs. These results are an important reference for clinicians who are developing services that target real-world patients with PDs.

An Innovative Wearable Device For Monitoring Continuous Body Surface Temperature (HEARThermo): Instrument Validation Study.

BACKGROUND: Variations in body temperature are highly informative during an illness. To date, there are not many adequate studies that have investigated the feasibility of a wearable wrist device for the continuous monitoring of body surface temperatures in humans. OBJECTIVE: The objective of this study was to validate the performance of HEARThermo, an innovative wearable device, which was developed to continuously monitor the body surface temperature in humans. METHODS: We implemented a multi-method research design in this study, which included 2 validation studies-one in the laboratory and one with human subjects. In validation study I, we evaluated the test-retest reliability of HEARThermo in the laboratory to measure the temperature and to correct the values recorded by each HEARThermo by using linear regression models. We conducted validation study II on human subjects who wore HEARThermo for the measurement of their body surface temperatures. Additionally, we compared the HEARThermo temperature recordings with those recorded by the infrared skin thermometer simultaneously. We used intraclass correlation coefficients (ICCs) and Bland-Altman plots to analyze the criterion validity and agreement between the 2 measurement tools. RESULTS: A total of 66 participants (age range, 10-77 years) were recruited, and 152,881 completed data were analyzed in this study. The 2 validation studies in the laboratory and on human skin indicated that HEARThermo showed a good test-retest reliability (ICC 0.96-0.98) and adequate criterion validity with the infrared skin thermometer at room temperatures of 20°C-27.9°C (ICC 0.72, P<.001). The corrected measurement bias averaged -0.02°C, which was calibrated using a water bath ranging in temperature from 16°C to 40°C. The values of each HEARThermo improved by the regression models were not significantly different from the temperature of the water bath (P=.19). Bland-Altman plots showed no visualized systematic bias. HEARThermo had a bias of 1.51°C with a 95% limit of agreement between -1.34°C and 4.35°C. CONCLUSIONS: The findings of our study show the validation of HEARThermo for the continuous monitoring of body surface temperatures in humans.

Predicting Serum Levels of Lithium-Treated Patients: A Supervised Machine Learning Approach.

Routine monitoring of lithium levels is common clinical practice. This is because the lithium prediction strategies available developed by previous studies are still limited due to insufficient prediction performance. Thus, we used machine learning approaches to predict lithium concentration in a large real-world dataset. Real-world data from multicenter electronic medical records were used in different machine learning algorithms to predict: (1) whether the serum level was 0.6-1.2 mmol/L or 0.0-0.6 mmol/L (binary prediction), and (2) its concentration value (continuous prediction). We developed models from 1505 samples through 5-fold cross-validation and used 204 independent samples to test their performance by evaluating their accuracy. Moreover, we ranked the most important clinical features in different models and reconstructed three reduced models with fewer clinical features. For binary and continuous predictions, the average accuracy of these models was 0.70-0.73 and 0.68-0.75, respectively. Seven features were listed as important features related to serum lithium levels of 0.6-1.2 mmol/L or higher lithium concentration, namely older age, lower systolic blood pressure, higher daily and last doses of lithium prescription, concomitant psychotropic drugs with valproic acid and -pine drugs, and comorbid substance-related disorders. After reducing the features in the three new predictive models, the binary or continuous models still had an average accuracy of 0.67-0.74. Machine learning processes complex clinical data and provides a potential tool for predicting lithium concentration. This may help in clinical decision-making and reduce the frequency of serum level monitoring.

AuDis: an automatic CRF-enhanced disease normalization in biomedical text.

Diseases play central roles in many areas of biomedical research and healthcare. Consequently, aggregating the disease knowledge and treatment research reports becomes an extremely critical issue, especially in rapid-growth knowledge bases (e.g. PubMed). We therefore developed a system, AuDis, for disease mention recognition and normalization in biomedical texts. Our system utilizes an order two conditional random fields model. To optimize the results, we customize several post-processing steps, including abbreviation resolution, consistency improvement and stopwords filtering. As the official evaluation on the CDR task in BioCreative V, AuDis obtained the best performance (86.46% of F-score) among 40 runs (16 unique teams) on disease normalization of the DNER sub task. These results suggest that AuDis is a high-performance recognition system for disease recognition and normalization from biomedical literature.Database URL: http://ikmlab.csie.ncku.edu.tw/CDR2015/AuDis.html.

Curatable Named-Entity Recognition Using Semantic Relations.

Named-entity recognition (NER) plays an important role in the development of biomedical databases. However, the existing NER tools produce multifarious named-entities which may result in both curatable and non-curatable markers. To facilitate biocuration with a straightforward approach, classifying curatable named-entities is helpful with regard to accelerating the biocuration workflow. Co-occurrence Interaction Nexus with Named-entity Recognition (CoINNER) is a web-based tool that allows users to identify genes, chemicals, diseases, and action term mentions in the Comparative Toxicogenomic Database (CTD). To further discover interactions, CoINNER uses multiple advanced algorithms to recognize the mentions in the BioCreative IV CTD Track. CoINNER is developed based on a prototype system that annotated gene, chemical, and disease mentions in PubMed abstracts at BioCreative 2012 Track I (literature triage). We extended our previous system in developing CoINNER. The pre-tagging results of CoINNER were developed based on the state-of-the-art named entity recognition tools in BioCreative III. Next, a method based on conditional random fields (CRFs) is proposed to predict chemical and disease mentions in the articles. Finally, action term mentions were collected by latent Dirichlet allocation (LDA). At the BioCreative IV CTD Track, the best F-measures reached for gene/protein, chemical/drug and disease NER were 54 percent while CoINNER achieved a 61.5 percent F-measure. System URL: http://ikmbio.csie.ncku.edu.tw/coinner/ introduction.htm.