LCD Benchmark: Long Clinical Document Benchmark on Mortality Prediction.

Natural Language Processing (NLP) is a study of automated processing of text data. Application of NLP in the clinical domain is important due to the rich unstructured information implanted in clinical documents, which often remains inaccessible in structured data. Empowered by the recent advance of language models (LMs), there is a growing interest in their application within the clinical domain. When applying NLP methods to a certain domain, the role of benchmark datasets are crucial as benchmark datasets not only guide the selection of best-performing models but also enable assessing of the reliability of the generated outputs. Despite the recent availability of LMs capable of longer context, benchmark datasets targeting long clinical document classification tasks are absent. To address this issue, we propose LCD benchmark, a benchmark for the task of predicting 30-day out-of-hospital mortality using discharge notes of MIMIC-IV and statewide death data. Our notes have a median word count of 1687 and an interquartile range of 1308 to 2169. We evaluated this benchmark dataset using baseline models, from bag-of-words and CNN to Hierarchical Transformer and an open-source instruction-tuned large language model. Additionally, we provide a comprehensive analysis of the model outputs, including manual review and visualization of model weights, to offer insights into their predictive capabilities and limitations. We expect LCD benchmarks to become a resource for the development of advanced supervised models, prompting methods, or the foundation models themselves, tailored for clinical text. The benchmark dataset is available at https://github.com/Machine-Learning-for-Medical-Language/long-clinical-doc.

Coatable Negative Dispersion Retarder: Kinetically Controlled Self-Assembly Pathway of Butterfly-Shaped Molecular Building Blocks for the Construction of Nanocolumns.

Disc-shaped building blocks with columnar phases have attracted attention for their potential in optical applications, including a retarder. However, to achieve coatable high-performance optical films, it is essential to understand a subtle interaction balance between building blocks and relevant self-assembled behaviors during material processing. Herein, we studied a self-assembled nanocolumn evaluation of linear butterfly-shaped dendrons (T-A3D) consisting of thiophene-based conjugated core and flexible alkyl dendron. X-ray diffraction provided insight into the unique hexagonal columnar liquid crystal phase of T-A3D, driven by intermolecular hydrogen bonding and coplanarity of the thiophene-based conjugated core. The formation of a self-assembled nanocolumn with high mobility enabled the uniaxial orientation of butterfly-shaped T-A3D on the aligned rod-shaped nematic reactive mesogens, resulting in a transparent and colorless two-layered negative retarder. The self-assembled nanocolumn consisting of butterfly-shaped molecule would break a new ground for developing advanced optical thin films.

Biomedical relation extraction with knowledge base-refined weak supervision.

Biomedical relation extraction (BioRE) is the task of automatically extracting and classifying relations between two biomedical entities in biomedical literature. Recent advances in BioRE research have largely been powered by supervised learning and large language models (LLMs). However, training of LLMs for BioRE with supervised learning requires human-annotated data, and the annotation process often accompanies challenging and expensive work. As a result, the quantity and coverage of annotated data are limiting factors for BioRE systems. In this paper, we present our system for the BioCreative VII challenge-DrugProt track, a BioRE system that leverages a language model structure and weak supervision. Our system is trained on weakly labelled data and then fine-tuned using human-labelled data. To create the weakly labelled dataset, we combined two approaches. First, we trained a model on the original dataset to predict labels on external literature, which will become a model-labelled dataset. Then, we refined the model-labelled dataset using an external knowledge base. Based on our experiment, our approach using refined weak supervision showed significant performance gain over the model trained using standard human-labelled datasets. Our final model showed outstanding performance at the BioCreative VII challenge, achieving 3rd place (this paper focuses on our participating system in the BioCreative VII challenge). Database URL: http://wonjin.info/biore-yoon-et-al-2022.

Full-text chemical identification with improved generalizability and tagging consistency.

Chemical identification involves finding chemical entities in text (i.e. named entity recognition) and assigning unique identifiers to the entities (i.e. named entity normalization). While current models are developed and evaluated based on article titles and abstracts, their effectiveness has not been thoroughly verified in full text. In this paper, we identify two limitations of models in tagging full-text articles: (1) low generalizability to unseen mentions and (2) tagging inconsistency. We use simple training and post-processing methods to address the limitations such as transfer learning and mention-wise majority voting. We also present a hybrid model for the normalization task that utilizes the high recall of a neural model while maintaining the high precision of a dictionary model. In the BioCreative VII NLM-Chem track challenge, our best model achieves 86.72 and 78.31 F1 scores in named entity recognition and normalization, significantly outperforming the median (83.73 and 77.49 F1 scores) and taking first place in named entity recognition. In a post-challenge evaluation, we re-implement our model and obtain 84.70 F1 score in the normalization task, outperforming the best score in the challenge by 3.34 F1 score. Database URL: https://github.com/dmis-lab/bc7-chem-id.

Sequence tagging for biomedical extractive question answering.

MOTIVATION: Current studies in extractive question answering (EQA) have modeled the single-span extraction setting, where a single answer span is a label to predict for a given question-passage pair. This setting is natural for general domain EQA as the majority of the questions in the general domain can be answered with a single span. Following general domain EQA models, current biomedical EQA (BioEQA) models utilize the single-span extraction setting with post-processing steps. RESULTS: In this article, we investigate the question distribution across the general and biomedical domains and discover biomedical questions are more likely to require list-type answers (multiple answers) than factoid-type answers (single answer). This necessitates the models capable of producing multiple answers for a question. Based on this preliminary study, we propose a sequence tagging approach for BioEQA, which is a multi-span extraction setting. Our approach directly tackles questions with a variable number of phrases as their answer and can learn to decide the number of answers for a question from training data. Our experimental results on the BioASQ 7b and 8b list-type questions outperformed the best-performing existing models without requiring post-processing steps. AVAILABILITY AND IMPLEMENTATION: Source codes and resources are freely available for download at https://github.com/dmis-lab/SeqTagQA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Pandemics are catalysts of scientific novelty: Evidence from COVID-19.

Scientific novelty drives the efforts to invent new vaccines and solutions during the pandemic. First-time collaboration and international collaboration are two pivotal channels to expand teams' search activities for a broader scope of resources required to address the global challenge, which might facilitate the generation of novel ideas. Our analysis of 98,981 coronavirus papers suggests that scientific novelty measured by the BioBERT model that is pretrained on 29 million PubMed articles, and first-time collaboration increased after the outbreak of COVID-19, and international collaboration witnessed a sudden decrease. During COVID-19, papers with more first-time collaboration were found to be more novel and international collaboration did not hamper novelty as it had done in the normal periods. The findings suggest the necessity of reaching out for distant resources and the importance of maintaining a collaborative scientific community beyond nationalism during a pandemic.

Imidazolium-Functionalized Diacetylene Amphiphiles: Strike a Lighter and Wear Polaroid Glasses to Decipher the Secret Code.

The development of smart inks that change color and transparency in response to external stimuli is very important for various fields, from modern art to safety and anticounterfeiting technology. A uniaxially oriented diacetylene thin film on a macroscopic area is obtained by coating, self-assembling and topochemical photopolymerizing of imidazolium-functionalized diacetylenes (M-DA and T-DA) and 4,6-decadiyne ink (70 wt%:20 wt%:10 wt%) exhibiting a lyotropic smectic A liquid-crystalline phase at room temperature. The color and transparency of letters and symbols written with the DA-based secret inks change reversibly from blue to red as well as from colorless transparent to black opaque depending on the temperature and polarization axis. A secret code written with thermoresponsive and polarization-dependent secret inks consisting of imidazolium-functionalized diacetylenes is successfully deciphered by wearing polaroid glasses and holding a burning torch.

Diacetylene-Functionalized Dendrons: Self-Assembled and Photopolymerized Three-Dimensional Networks for Advanced Self-Healing and Wringing Soft Materials.

The physical properties of supramolecular soft materials strongly depend on the molecular packing structures constructed by thermodynamically and kinetically controlled molecular self-assembly. To investigate the relationship between molecular function and self-assembled molecular packing structure, a series of diacetylene (DA)-based supramolecules was synthesized by chemically connecting flexible dendrons to DA with amide (aDA-D) or ester (eDA-D) functions. The three-dimensional (3D) organogel network of amide-functionalized aDA-D was prepared in both polar and nonpolar solvents due to the intermolecular hydrogen bonding. 3D networks of aDA-D can be further stabilized by topochemical photopolymerization. The self-healing behavior of aDA-D was observed in the sheet-like structure formed in n-dodecane by the hydrophobic interaction between the gelator and solvent. The wringing behavior of aDA-D was also demonstrated using the dynamic interaction of amide function with n-butanol solvent. Kinetically controlled and photostabilized 3D networks can be a key component from biomedical devices to soft robotic applications.

BioBERT: a pre-trained biomedical language representation model for biomedical text mining.

MOTIVATION: Biomedical text mining is becoming increasingly important as the number of biomedical documents rapidly grows. With the progress in natural language processing (NLP), extracting valuable information from biomedical literature has gained popularity among researchers, and deep learning has boosted the development of effective biomedical text mining models. However, directly applying the advancements in NLP to biomedical text mining often yields unsatisfactory results due to a word distribution shift from general domain corpora to biomedical corpora. In this article, we investigate how the recently introduced pre-trained language model BERT can be adapted for biomedical corpora. RESULTS: We introduce BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical Text Mining), which is a domain-specific language representation model pre-trained on large-scale biomedical corpora. With almost the same architecture across tasks, BioBERT largely outperforms BERT and previous state-of-the-art models in a variety of biomedical text mining tasks when pre-trained on biomedical corpora. While BERT obtains performance comparable to that of previous state-of-the-art models, BioBERT significantly outperforms them on the following three representative biomedical text mining tasks: biomedical named entity recognition (0.62% F1 score improvement), biomedical relation extraction (2.80% F1 score improvement) and biomedical question answering (12.24% MRR improvement). Our analysis results show that pre-training BERT on biomedical corpora helps it to understand complex biomedical texts. AVAILABILITY AND IMPLEMENTATION: We make the pre-trained weights of BioBERT freely available at https://github.com/naver/biobert-pretrained, and the source code for fine-tuning BioBERT available at https://github.com/dmis-lab/biobert.

CollaboNet: collaboration of deep neural networks for biomedical named entity recognition.

BACKGROUND: Finding biomedical named entities is one of the most essential tasks in biomedical text mining. Recently, deep learning-based approaches have been applied to biomedical named entity recognition (BioNER) and showed promising results. However, as deep learning approaches need an abundant amount of training data, a lack of data can hinder performance. BioNER datasets are scarce resources and each dataset covers only a small subset of entity types. Furthermore, many bio entities are polysemous, which is one of the major obstacles in named entity recognition. RESULTS: To address the lack of data and the entity type misclassification problem, we propose CollaboNet which utilizes a combination of multiple NER models. In CollaboNet, models trained on a different dataset are connected to each other so that a target model obtains information from other collaborator models to reduce false positives. Every model is an expert on their target entity type and takes turns serving as a target and a collaborator model during training time. The experimental results show that CollaboNet can be used to greatly reduce the number of false positives and misclassified entities including polysemous words. CollaboNet achieved state-of-the-art performance in terms of precision, recall and F1 score. CONCLUSIONS: We demonstrated the benefits of combining multiple models for BioNER. Our model has successfully reduced the number of misclassified entities and improved the performance by leveraging multiple datasets annotated for different entity types. Given the state-of-the-art performance of our model, we believe that CollaboNet can improve the accuracy of downstream biomedical text mining applications such as bio-entity relation extraction.

Coatable Compensator for Flexible Display: Single-Layered Negative Dispersion Retarder Fabricated by Coating, Self-Assembling, and Polymerizing Host-Guest Reactive Mesogens.

Antireflective films for flexible display devices should be made by a coating process rather than a stretching process. Additionally, the compensator, which is the core of the antireflective film, must exhibit a wavelength-independent phase retardation in the visible-light region to act as an ideal retarder. To satisfy all of these requirements, we intend to make a single-layered negative dispersion retarder (SNDR) with a single coating process followed by polymer stabilization. To achieve this goal, X-shaped reactive mesogen (X2RM) is newly synthesized as a guest RM and mixed into a host RM that exhibits a smectic A mesophase. Based on the thermal, spectroscopic, microscopic, and scattering analyses combined with computer simulation, the content of X2RM in the HCM026 molecule is optimized to be 40 wt %. The SNDR thin film is fabricated by coating the optimized H-G mixture on the rubbed alignment substrate and subsequent heat treatment. The trans-to-cis photoisomerization of imine bond can help X2RM to be located in the smectic interlayer of the HCM026. The molecular long axis of HCM026 is parallel to the rubbing direction of the alignment layer and the conjugated benzene rings of X2RM aligned perpendicular to the molecular long axis of smectic RM, which is the ideal molecular arrangement of negative dispersion retarder. Additionally, polarized UV polymerization improves the mechanical and chemical stability as well as the molecular orientation of SNDR.

Interesting phase behaviors and ion-conducting properties of dicationic N-alkylimidazolium tetrafluoroborate salts.

A series of dicationic imidazolium bis(tetrafluoroborate) salts were newly synthesized, and their phase transition behaviors were correlated with thermal, scattering, optical and conductivity results. The bis-imidazolium salts having side-chain lengths of C6-C10 showed plastic crystal mesophases, while a liquid crystal mesophase was formed in the bis-imidazolium salts with long side-chains (C11 and C12). Soft plastic and liquid crystalline phases were also confirmed by wide-angle X-ray diffraction. For the bis-imidazolium salts exhibiting a plastic crystal mesophase, the ionic conductivity suddenly increased at the melting temperature. However, the bis-imidazolium salts with long side-chains showed a slope increase during the liquid crystal-liquid crystal transition.

Programmed Hierarchical Hybrid Nanostructures from Fullerene-Dendrons and Pyrene-Dendrons.

The construction of fullerene (C60 ) hierarchical nanostructures with the help of amphiphilic molecules remains a challenging task in nanoscience and nanotechnology. Utilizing the host-guest complex concept, sub-10 nm layered superstructures are constructed from a monofunctionalized C60 dendron (C60 D, guest) and tweezer-like pyrene dendron (PD, host). Since C60 D and PD are asymmetric shape amphiphiles having liquid crystal (LC) dendrons, both C60 D and PD construct head-to-head bilayer superstructures by themselves. From fluorescence titration experiments, it is realized that the host-guest complex shows 1:1 stoichiometric binding with a binding constant (Ksv = 2.45 × 105 m-1 ). Based on the morphological observations and scattering analyses, it is found that buckle-like asymmetric building blocks (C60 D·PD) are self-assembled by the host-guest complex and construct multilayer hybrid nanostructures. The hierarchical hybrid nanostructures consist of the self-assembled C60 D·PD bilayer with a 2D C60 ·P nanoarray sandwiched between LC dendrons. This advanced strategy is expected to be a practicable and rational guideline for the fabrication of programmed hierarchical hybrid nanostructures.

Kinetically Controlled Polymorphic Superstructures of Pyrene-Based Asymmetric Liquid Crystal Dendron: Relationship Between Hierarchical Superstructures and Photophysical Properties.

To understand the relationship between kinetically controlled hierarchical superstructures and photophysical properties, pyrene-based asymmetric liquid crystal (LC) dendrons (abbreviated as PD) were newly synthesized by covalently attaching a pyrene moiety (P) at a biphenyl-based LC dendritic group (D). The phase transition behavior of PD has been systematically studied with a combined technique of thermal analysis, microscopy, spectroscopy, and scattering analysis. PD formed two different crystalline structures depending on the cooling rate: a stable crystalline phase (Ks , slow cooling) and a metastable crystalline phase (Kms , quenching). The kinetically controlled molecular packing structures of PD depend on the competition and cooperation of intermolecular physical interactions with nanophase separation. Upon slow cooling, the PD dimer formed by intermoelcular H-bonding constructed a layered hierarchical structure with the help of nanophase separation. Owing to the strong π-π stacking (J-aggregation) with weak H-bondings, the PD dimer in the layer was slightly tilted to give a monoclinic layered structure with a periodic layer d-spacing of 6.6 nm. In contrast, the metastable Kms phase formed by the quenching process showed a significant tilt of the PD dimer in the layer (d-spacing=4.4 nm) due to the weak π-π stacking (H-aggregation) and the strong H-bondings.

Comparative studies of bioactive organosulphur compounds and antioxidant activities in garlic (Allium sativum L.), elephant garlic (Allium ampeloprasum L.) and onion (Allium cepa L.).

We evaluated organosulphur compounds in Allium vegetables, including garlic, elephant garlic and onion, using high-performance liquid chromatography. Among organosulphur compounds, elephant garlic had considerable γ-glutamyl peptides, and garlic had the highest alliin content. Onion had low level of organosulphur compounds than did elephant garlic and garlic. In addition, antioxidant capacities were evaluated by oxygen radical absorbance capacity (ORAC) values and 2,2-diphenyl-1-picrylhydrazyl (DPPH) and 2,2'-azinobis(3-ethylbenzothiazoline-6-sulphonic acid) (ABTS) radical scavenging assay. The results showed that garlic had the highest antioxidant capacity, followed by elephant garlic and onion. Furthermore, a positive correlation was observed between antioxidant activities and organosulphur compounds (R > 0.77). Therefore, our results indicate that there was a close relationship between antioxidant capacity and organosulphur compounds in Allium vegetables.

Azobenzene Molecular Machine: Light-Induced Wringing Gel Fabricated from Asymmetric Macrogelator.

To develop light-triggered wringing gels, an asymmetric macrogelator (1AZ3BP) was newly synthesized by the chemically bridging a photoisomerizable azobenzene (1AZ) molecular machine and a biphenyl-based (3BP) dendron with a 1,4-phenylenediformamide connector. 1AZ3BP was self-assembled into a layered superstructure in the bulk state, but 1AZ3BP formed a three-dimensional (3D) network organogel in solution. Upon irradiating UV light onto the 3D network organogel, the solvent of the organogel was squeezed and the 3D network was converted to the layered morphology. It was realized that the metastable 3D network organogels were fabricated mainly due to the nanophase separation in solution. UV isomerization of 1AZ3BP provided sufficient molecular mobility to form strong hydrogen bonds for the construction of the stable layered superstructure. The light-triggered wringing gels can be smartly applied in remote-controlled generators, liquid storages, and sensors.

Construction of Polymer-Stabilized Automatic MultiDomain Vertical Molecular Alignment Layers with Pretilt Angles by Photopolymerizing Dendritic Monomers under Electric Fields.

The synthesized itaconic acid-based dendritic amphiphile (Ita3C12) monomers and the methacryl polyhedral oligomeric silsesquioxane (MAPOSS) cross-linkers were directly introduced for the construction of automatic vertical alignment (auto-VA) layers in the host nematic liquid crystal (NLC) medium. The auto-VA layer can be stabilized by irradiating UV light. For the automatic fabrication of a polymer-stabilized multidomain VA (PS auto-MDVA) layer with a pretilt angle, Ita3C12 and MAPOSS were photopolymerized under the electric field by irradiating UV light on the multidomain electrode cell. Mainly because of the pretilted NLC at zero voltage, the electro-optic properties of the PS auto-MDVA cell were dramatically improved. From the morphological observations combined with surface chemical analyses, it was found that various sizes of protrusions on the solid substrates were automatically constructed by the two-step mechanisms. We demonstrated the PS auto-MDVA cell with the enhancement of electro-optic properties as a single-step process and investigated how the protrusions were automatically developed during the polymer stabilization.

Multifunctional Optical Thin Films Fabricated by the Photopolymerization of Uniaxially Oriented Lyotropic Liquid Crystal Monomers for Electro-Optical Devices.

A multifunctional optical thin film (MOTF) is fabricated by coating the newly synthesized perylene-based reactive mesogen (PBRM) and stabilized by the subsequent photopolymerization. Based on the spectroscopic results combined with morphological observations, it is found that nematic liquid crystal (NLC) is aligned parallel to the molecular long axis of PBRM not only due to the long-range physical anchoring effect but also due to the short-range molecular physical interactions between alignment layer and NLC molecules. From the electro-optical properties of LC test cells fabricated with the PBRM MOTF, it is clearly demonstrated that the PBRM MOTF can work as the planar LC alignment layer as well as the in-cell coatable polarizer. The coatable PBRM MOTF from lyotropic chromonic reactive mesogens can pave a new way for the flexible optoelectronic devices.

Phytochemical Compositions of Immature Wheat Bran, and Its Antioxidant Capacity, Cell Growth Inhibition, and Apoptosis Induction through Tumor Suppressor Gene.

The purpose of this study was to investigate the phytochemical compositions and antioxidant capacity, cell growth inhibition, and apoptosis induction in extracts of immature wheat bran. Immature wheat bran (IWB) was obtained from immature wheat harvested 10 days earlier than mature wheat. The phytochemical compositions of bran extract samples were analyzed by ultra-high performance liquid chromatography. The total ferulic acid (3.09 mg/g) and p-coumaric acid (75 µg/g) in IWB were significantly higher than in mature wheat bran (MWB, ferulic acid: 1.79 mg/g; p-coumaric acid: 55 µg/g). The oxygen radical absorbance capacity (ORAC: 327 µM Trolox equivalents (TE)/g) and cellular antioxidant activity (CAA: 4.59 µM Quercetin equivalents (QE)/g) of the IWB were higher than those of the MWB (ORAC: 281 µM TE/g; CAA: 0.63 µM QE/g). When assessing cell proliferation, the IWB extracts resulted in the lowest EC50 values against HT-29 (18.9 mg/mL), Caco-2 (7.74 mg/mL), and HeLa cells (8.17 mg/mL) among bran extract samples. Additionally, the IWB extracts increased the gene expression of p53 and PTEN (tumor suppressor genes) in HT-29 cells, indicating inhibited cell growth and induced apoptosis through tumor suppressor genes.

Self-Assembled Hierarchical Superstructures from the Benzene-1,3,5-Tricarboxamide Supramolecules for the Fabrication of Remote-Controllable Actuating and Rewritable Films.

The well-defined hierarchical superstructures constructed by the self-assembly of programmed supramolecules can be organized for the fabrication of remote-controllable actuating and rewritable films. To realize this concept, we newly designed and synthesized a benzene-1,3,5-tricarboxamide (BTA) derivative (abbreviated as BTA-3AZO) containing photoresponsive azobenzene (AZO) mesogens on the periphery of the BTA core. BTA-3AZO was first self-assembled to nanocolumns mainly driven by the intermolecular hydrogen-bonds between BTA cores, and these self-assembled nanocolumns were further self-organized laterally to form the low-ordered hexagonal columnar liquid crystal (LC) phase below the isotropization temperature. Upon cooling, a lamello-columnar crystal phase emerged at room temperature via a highly ordered lamello-columnar LC phase. The three-dimensional (3D) organogel networks consisted of fibrous and lamellar superstructures were fabricated in the BTA-3AZO cyclohexane-methanol solutions. By tuning the wavelength of light, the shape and color of the 3D networked thin films were remote-controlled by the conformational changes of azobenzene moieties in the BTA-3AZO. The demonstrations of remote-controllable 3D actuating and rewritable films with the self-assembled hierarchical BTA-3AZO thin films can be stepping stones for the advanced flexible optoelectronic devices.