Identification of potential pan-coronavirus therapies using a computational drug repurposing platform.

In the past 20 years, there have been several infectious disease outbreaks in humans for which the causative agent has been a zoonotic coronavirus. Novel infectious disease outbreaks, as illustrated by the current coronavirus disease 2019 (COVID-19) pandemic, demand a rapid response in terms of identifying effective treatments for seriously ill patients. The repurposing of approved drugs from other therapeutic areas is one of the most practical routes through which to approach this. Here, we present a systematic network-based drug repurposing methodology, which interrogates virus-human, human protein-protein and drug-protein interactome data. We identified 196 approved drugs that are appropriate for repurposing against COVID-19 and 102 approved drugs against a related coronavirus, severe acute respiratory syndrome (SARS-CoV). We constructed a protein-protein interaction (PPI) network based on disease signatures from COVID-19 and SARS multi-omics datasets. Analysis of this PPI network uncovered key pathways. Of the 196 drugs predicted to target COVID-19 related pathways, 44 (hypergeometric p-value: 1.98e-04) are already in COVID-19 clinical trials, demonstrating the validity of our approach. Using an artificial neural network, we provide information on the mechanism of action and therapeutic value for each of the identified drugs, to facilitate their rapid repurposing into clinical trials.

Concept embedding to measure semantic relatedness for biomedical information ontologies.

There have been many attempts to identify relationships among concepts corresponding to terms from biomedical information ontologies such as the Unified Medical Language System (UMLS). In particular, vector representation of such concepts using information from UMLS definition texts is widely used to measure the relatedness between two biological concepts. However, conventional relatedness measures have a limited range of applicable word coverage, which limits the performance of these models. In this paper, we propose a concept-embedding model of a UMLS semantic relatedness measure to overcome the limitations of earlier models. We obtained context texts of biological concepts that are not defined in UMLS by utilizing Wikipedia as an external knowledgebase. Concept vector representations were then derived from the context texts of the biological concepts. The degree of relatedness between two concepts was defined as the cosine similarity between corresponding concept vectors. As a result, we validated that our method provides higher coverage and better performance than the conventional method.

Global Analysis of SUMO-Binding Proteins Identifies SUMOylation as a Key Regulator of the INO80 Chromatin Remodeling Complex.

SUMOylation is a critical regulator of a broad range of cellular processes, and is thought to do so in part by modulation of protein interaction. To comprehensively identify human proteins whose interaction is modulated by SUMOylation, we developed an in vitro binding assay using human proteome microarrays to identify targets of SUMO1 and SUMO2. We then integrated these results with protein SUMOylation and protein-protein interaction data to perform network motif analysis. We focused on a single network motif we termed a SUMOmodPPI (SUMO-modulated Protein-Protein Interaction) that included the INO80 chromatin remodeling complex subunits TFPT and INO80E. We validated the SUMO-binding activity of INO80E, and showed that TFPT is a SUMO substrate both in vitro and in vivo We then demonstrated a key role for SUMOylation in mediating the interaction between these two proteins, both in vitro and in vivo By demonstrating a key role for SUMOylation in regulating the INO80 chromatin remodeling complex, this work illustrates the power of bioinformatics analysis of large data sets in predicting novel biological phenomena.

A fast detection of fusion genes from paired-end RNA-seq data.

BACKGROUND: Fusion genes are known to be drivers of many common cancers, so they are potential markers for diagnosis, prognosis or therapy response. The advent of paired-end RNA sequencing enhances our ability to discover fusion genes. While there are available methods, routine analyses of large number of samples are still limited due to high computational demands. RESULTS: We develop FuSeq, a fast and accurate method to discover fusion genes based on quasi-mapping to quickly map the reads, extract initial candidates from split reads and fusion equivalence classes of mapped reads, and finally apply multiple filters and statistical tests to get the final candidates. We apply FuSeq to four validated datasets: breast cancer, melanoma and glioma datasets, and one spike-in dataset. The results reveal high sensitivity and specificity in all datasets, and compare well against other methods such as FusionMap, TRUP, TopHat-Fusion, SOAPfuse and JAFFA. In terms of computational time, FuSeq is two-fold faster than FusionMap and orders of magnitude faster than the other methods. CONCLUSIONS: With this advantage of less computational demands, FuSeq makes it practical to investigate fusion genes in large numbers of samples. FuSeq is implemented in C++ and R, and available at https://github.com/nghiavtr/FuSeq for non-commercial uses.

Association between fever pattern and clinical manifestations of adult-onset Still's disease: unbiased analysis using hierarchical clustering.

OBJECTIVES: To perform unbiased analysis of fever patterns and to investigate their association with clinical manifestations and outcome of patients with adult-onset Still's disease (AOSD). METHODS: AOSD patients who were treated as in-patients from 2004 through 2015 were grouped according to 24-hour body temperature (BT) by hierarchical clustering using a Euclidean distance metric with complete linkage. The clinical and laboratory characteristics of the groups were then examined. RESULTS: Hierarchical clustering partitioned 70 AOSD patients into three distinct groups. Group 1 (n=14) had the highest mean BT (38.1± 0.4°C) and the widest variation in BT (2.7±0.9°C). Group 2 (n=35) had a lower mean BT (37.4±0.3°C) and a smaller variation (2.1±0.7°C). Group 3 (n=21) had the lowest mean BT (36.7±0.3°C) and the smallest variation (1.5±0.6°C). Clinical features and extent of organ involvement did not differ significantly between groups. However, Group 1 had lower platelet counts and higher lactate dehydrogenase, ferritin levels, and prothrombin time than the other groups. In addition, Group 1 exhibited higher risk of having a macrophage activation syndrome (MAS) and tended to require more intense treatment with corticosteroids and immunosuppressant to achieve clinical remission as compared to other groups. CONCLUSIONS: Hierarchical clustering identified three distinct fever patterns in patients with AOSD. Higher BT was associated with wider variations in diurnal temperature, higher risk of developing MAS, more intense treatment, and longer time to clinical remission, suggesting that fever pattern is a prognostic factor for AOSD.

Lhx2 Is an Essential Factor for Retinal Gliogenesis and Notch Signaling.

Müller glia (MG) are the only glial cell type produced by the neuroepithelial progenitor cells that generate the vertebrate retina. MG are required to maintain retinal homeostasis and support the survival of retinal neurons. Furthermore, in certain vertebrate classes, MG function as adult stem cells, mediating retinal regeneration in response to injury. However, the mechanisms that regulate MG development are poorly understood because there is considerable overlap in gene expression between retinal progenitor cells and differentiated MG. We show that the LIM homeodomain transcription factor Lhx2 is required for the development of MG in the mouse retina. Temporally controlled knock-out studies reveal a requirement for Lhx2 during all stages of MG development, ranging from the proliferation of gliocompetent retinal progenitors, activation of Müller-specific gene expression, and terminal differentiation of MG morphological features. We show that Lhx2 regulates gliogenesis in part by regulating directly the expression of Notch pathway genes including Notch1, Dll1, and Dll3 and gliogenic transcription factors such as Hes1, Hes5, Sox8, and Rax. Conditional knock-out of Lhx2 resulted in a rapid downregulation of Notch pathway genes and loss of Notch signaling. We further demonstrate that Müller gliogenesis induced by misexpression of the potently gliogenic Notch pathway transcriptional effector Hes5 requires Lhx2 expression. These results indicate that Lhx2 not only directly regulates expression of Notch signaling pathway components, but also acts together with the gliogenic Notch pathway to drive MG specification and differentiation.

Context-specific functional module based drug efficacy prediction.

BACKGROUND: It is necessary to evaluate the efficacy of individual drugs on patients to realize personalized medicine. Testing drugs on patients in clinical trial is the only way to evaluate the efficacy of drugs. The approach is labour intensive and requires overwhelming costs and a number of experiments. Therefore, preclinical model system has been intensively investigated for predicting the efficacy of drugs. Current computational drug sensitivity prediction approaches use general biological network modules as their prediction features. Therefore, they miss indirect effectors or the effects from tissue-specific interactions. RESULTS: We developed cell line specific functional modules. Enriched scores of functional modules are utilized as cell line specific features to predict the efficacy of drugs. Cell line specific functional modules are clusters of genes, which have similar biological functions in cell line specific networks. We used linear regression for drug efficacy prediction. We assessed the prediction performance in leave-one-out cross-validation (LOOCV). Our method was compared with elastic net model, which is a popular model for drug efficacy prediction. In addition, we analysed drug sensitivity-associated functions of five drugs - lapatinib, erlotinib, raloxifene, tamoxifen and gefitinib- by our model. CONCLUSIONS: Our model can provide cell line specific drug efficacy prediction and also provide functions which are associated with drug sensitivity. Therefore, we could utilize drug sensitivity associated functions for drug repositioning or for suggesting secondary drugs for overcoming drug resistance.

Global analysis of phosphorylation networks in humans.

Phosphorylation-mediated signaling plays a crucial role in nearly every aspect of cellular physiology. A recent study based on protein microarray experiments identified a large number of kinase-substrate relationships (KSRs), and built a comprehensive and reliable phosphorylation network in humans. Analysis of this network, in conjunction with additional resources, revealed several key features. First, comparison of the human and yeast phosphorylation networks uncovered an evolutionarily conserved signaling backbone dominated by kinase-to-kinase relationships. Second, although most of the KSRs themselves are not conserved, the functions enriched in the substrates for a given kinase are often conserved. Third, the prevalence of kinase-transcription factor regulatory modules suggests that phosphorylation and transcriptional regulatory networks are inherently wired together to form integrated regulatory circuits. Overall, the phosphorylation networks described in this work promise to offer new insights into the properties of kinase signaling pathways, at both the global and the protein levels. This article is part of a Special Issue entitled: Computational Proteomics, Systems Biology & Clinical Implications. Guest Editor: Yudong Cai.

Prediction of promoters and enhancers using multiple DNA methylation-associated features.

BACKGROUND: Regulatory regions (e.g. promoters and enhancers) play an essential role in human development and disease. Many computational approaches have been developed to predict the regulatory regions using various genomic features such as sequence motifs and evolutionary conservation. However, these DNA sequence-based approaches do not reflect the tissue-specific nature of the regulatory regions. In this work, we propose to predict regulatory regions using multiple features derived from DNA methylation profile. RESULTS: We discovered several interesting features of the methylated CpG (mCpG) sites within regulatory regions. First, a hypomethylation status of CpGs within regulatory regions, compared to the genomic background methylation level, extended out >1000 bp from the center of the regulatory regions, demonstrating a high degree of correlation between the methylation statuses of neighboring mCpG sites. Second, when a regulatory region was inactive, as determined by histone mark differences between cell lines, methylation level of the mCpG site increased from a hypomethylated state to a hypermethylated state, the level of which was even higher than the genomic background. Third, a distinct set of sequence motifs was overrepresented surrounding mCpG sites within regulatory regions. Using 5 types of features derived from DNA methylation profiles, we were able to predict promoters and enhancers using machine-learning approach (support vector machine). The performances for prediction of promoters and enhancers are quite well, showing an area under the ROC curve (AUC) of 0.992 and 0.817, respectively, which is better than that simply based on methylation level, especially for prediction of enhancers. CONCLUSIONS: Our study suggests that DNA methylation features of mCpG sites can be used to predict regulatory regions.

Construction of human activity-based phosphorylation networks.

The landscape of human phosphorylation networks has not been systematically explored, representing vast, unchartered territories within cellular signaling networks. Although a large number of in vivo phosphorylated residues have been identified by mass spectrometry (MS)-based approaches, assigning the upstream kinases to these residues requires biochemical analysis of kinase-substrate relationships (KSRs). Here, we developed a new strategy, called CEASAR, based on functional protein microarrays and bioinformatics to experimentally identify substrates for 289 unique kinases, resulting in 3656 high-quality KSRs. We then generated consensus phosphorylation motifs for each of the kinases and integrated this information, along with information about in vivo phosphorylation sites determined by MS, to construct a high-resolution map of phosphorylation networks that connects 230 kinases to 2591 in vivo phosphorylation sites in 652 substrates. The value of this data set is demonstrated through the discovery of a new role for PKA downstream of Btk (Bruton's tyrosine kinase) during B-cell receptor signaling. Overall, these studies provide global insights into kinase-mediated signaling pathways and promise to advance our understanding of cellular signaling processes in humans.

Genetic and Metabolic Characteristics of Lean Nonalcoholic Fatty Liver Disease in a Korean Health Examinee Cohort.

Background/Aims: The pathophysiology of lean nonalcoholic fatty liver disease (NAFLD) is unclear but has been shown to be associated with more diverse pathogenic mechanisms than that of obese NAFLD. We investigated the characteristics of genetic or metabolic lean NAFLD in a health checkup cohort. Methods: This retrospective cross-sectional study analyzed single nucleotide polymorphism data for 6,939 health examinees. Lean individuals were categorized according to a body mass index cutoff of 23 kg/m2. Single nucleotide polymorphisms were analyzed using genotyping arrays. Results: The prevalence of lean NAFLD was 21.6% among all participants with NAFLD, and the proportion of lean NAFLD was 18.5% among lean participants. The prevalence of metabolic syndrome and diabetes among lean patients with NAFLD was 12.4% and 10.4%, respectively. Lean NAFLD appeared to be metabolic-associated in approximately 20.1% of patients. The homozygous minor allele (GG) of PNPLA3 (rs738409) and heterozygous minor alleles (CT, TT) of TM6SF2 (rs58542926) were associated with lean NAFLD. However, the prevalence of fatty liver was not associated with the genetic variants MBOAT7 (rs641738), HSD17B13 (rs72613567), MARC1 (rs2642438), or AGXT2 (rs2291702) in lean individuals. Lean NAFLD appeared to be associated with PNPLA3 or TM6SF2 genetic variation in approximately 32.1% of cases. Multivariate risk factor analysis showed that metabolic risk factors, genetic risk variants, and waist circumference were independent risk factors for lean NAFLD. Conclusions: In a considerable number of patients, lean NAFLD did not appear to be associated with known genetic or metabolic risk factors. Further studies are required to investigate additional risk factors and gain a more comprehensive understanding of lean NAFLD.

Rule-based multi-scale simulation for drug effect pathway analysis.

BACKGROUND: Biological systems are robust and complex to maintain stable phenotypes under various conditions. In these systems, drugs reported the limited efficacy and unexpected side-effects. To remedy this situation, many pharmaceutical laboratories have begun to research combination drugs and some of them have shown successful clinical results. Complementary action of multiple compounds could increase efficacy as well as reduce side-effects through pharmacological interactions. However, experimental approach requires vast cost of preclinical experiments and tests as the number of possible combinations of compound dosages increases exponentially. Computer model-based experiments have been emerging as one of the most promising solutions to cope with such complexity. Though there have been many efforts to model specific molecular pathways using qualitative and quantitative formalisms, they suffer from unexpected results caused by distant interactions beyond their localized models. RESULTS: In this work, we propose a rule-based multi-scale modelling platform. We have tested this platform with Type 2 diabetes (T2D) model, which involves the malfunction of numerous organs such as pancreas, circulation system, liver, and adipocyte. We have extracted T2D-related 190 rules by manual curation from literature, pathway databases and converting from different types of existing models. We have simulated twenty-two T2D drugs. The results of our simulation show drug effect pathways of T2D drugs and whether combination drugs have efficacy or not and how combination drugs work on the multi-scale model. CONCLUSIONS: We believe that our simulation would help to understand drug mechanism for the drug development and provide a new way to effectively apply existing drugs for new target. It also would give insight for identifying effective combination drugs.

Epigenetic modification of gene expression in cancer cells by terahertz demethylation.

Terahertz (THz) radiation can affect the degree of DNA methylation, the spectral characteristics of which exist in the terahertz region. DNA methylation is an epigenetic modification in which a methyl (CH3) group is attached to cytosine, a nucleobase in human DNA. Appropriately controlled DNA methylation leads to proper regulation of gene expression. However, abnormal gene expression that departs from controlled genetic transcription through aberrant DNA methylation may occur in cancer or other diseases. In this study, we demonstrate the modification of gene expression in cells by THz demethylation using resonant THz radiation. Using an enzyme-linked immunosorbent assay, we observed changes in the degree of global DNA methylation in the SK-MEL-3 melanoma cell line under irradiation with 1.6-THz radiation with limited spectral bandwidth. Resonant THz radiation demethylated living melanoma cells by 19%, with no significant occurrence of apurinic/apyrimidinic sites, and the demethylation ratio was linearly proportional to the power of THz radiation. THz demethylation downregulates FOS, JUN, and CXCL8 genes, which are involved in cancer and apoptosis pathways. Our results show that THz demethylation has the potential to be a gene expression modifier with promising applications in cancer treatment.

Physicians' Agreement on and Implementation of the 2019 European Alliance of Associations for Rheumatology Vaccination Guideline: An International Survey.

Objective: To evaluate the perspective of healthcare professionals towards the 2019 European Alliance of Associations for Rheumatology (EULAR) vaccination guideline in patients with autoimmune inflammatory rheumatic diseases (AIIRD). Methods: Healthcare professionals who care for patients with AIIRD were invited to participate in an online survey regarding their perspective on the 2019 update of the EULAR recommendations for vaccination in adult patients with AIIRD. Level of agreement and implementation of the 6 overarching principles and 9 recommendations were rated on a 5-point Likert scale (1~5). Results: Survey responses of 371 healthcare professionals from Asia (42.2%) and North America (41.6%), Europe (13.8%), and other countries were analyzed. Only 16.3% of participants rated their familiarity with the 2019 EULAR guideline as 5/5 ("very well"). There was a high agreement (≥4/5 rating) with the overarching principles, except for the principles applying to live-attenuated vaccines. There was a high level of agreement with the recommendations regarding influenza and pneumococcal vaccinations; implementation of these recommendations was also high. Participants also reported a high level of agreement with the remaining recommendations but did not routinely implement these recommendations. Conclusion: The 2019 update of EULAR recommendations for the vaccination of adult patients with AIIRD is generally thought to be important by healthcare professionals, although implementation of adequate vaccination is often lacking. Better education of healthcare providers may be important to optimize the vaccination coverage for patients with AIIRD.

Physician Perspectives on Vaccination in Patients With Autoimmune Inflammatory Rheumatic Diseases: An International Survey.

OBJECTIVE: To evaluate the perspective of physicians who care for patients with autoimmune inflammatory rheumatic disease (AIIRD) toward vaccination. METHODS: Physicians who care for patients with AIIRD were invited to participate in an online survey regarding their vaccination perspectives in adult patients with AIIRD. RESULTS: Survey responses of 370 physicians from Asia (41.1%), North America (41.6%), Europe (13.8%), and other countries (3.5%) were analyzed. Participants stated that rheumatologists (58.2%) should be primarily responsible for vaccination coverage, followed by general internists (19.3%) and family medicine practitioners (12.8%). Additionally, 96.7% of participants considered vaccination very important (≥ 4/5 rating) for patients with AIIRD. Despite these sentiments, only one-third (37%) reported vaccinating the majority (≥ 60%) of their patients. CONCLUSION: Physicians who care for patients with AIIRD agree that vaccines are effective and safe in patients with AIIRD. Unfortunately, they often do not ensure that their patients are adequately vaccinated. Further studies are needed to investigate how to improve vaccination coverage for this high-risk patient population.

Extru-seq: a method for predicting genome-wide Cas9 off-target sites with advantages of both cell-based and in vitro approaches.

We present a novel genome-wide off-target prediction method named Extru-seq and compare it with cell-based (GUIDE-seq), in vitro (Digenome-seq), and in silico methods using promiscuous guide RNAs with large numbers of valid off-target sites. Extru-seq demonstrates a high validation rate and retention of information about the intracellular environment, both beneficial characteristics of cell-based methods. Extru-seq also shows a low miss rate and could easily be performed in clinically relevant cell types with little optimization, which are major positive features of the in vitro methods. In summary, Extru-seq shows beneficial features of cell-based and in vitro methods.

Construction and analysis of an integrated regulatory network derived from high-throughput sequencing data.

We present a network framework for analyzing multi-level regulation in higher eukaryotes based on systematic integration of various high-throughput datasets. The network, namely the integrated regulatory network, consists of three major types of regulation: TF→gene, TF→miRNA and miRNA→gene. We identified the target genes and target miRNAs for a set of TFs based on the ChIP-Seq binding profiles, the predicted targets of miRNAs using annotated 3'UTR sequences and conservation information. Making use of the system-wide RNA-Seq profiles, we classified transcription factors into positive and negative regulators and assigned a sign for each regulatory interaction. Other types of edges such as protein-protein interactions and potential intra-regulations between miRNAs based on the embedding of miRNAs in their host genes were further incorporated. We examined the topological structures of the network, including its hierarchical organization and motif enrichment. We found that transcription factors downstream of the hierarchy distinguish themselves by expressing more uniformly at various tissues, have more interacting partners, and are more likely to be essential. We found an over-representation of notable network motifs, including a FFL in which a miRNA cost-effectively shuts down a transcription factor and its target. We used data of C. elegans from the modENCODE project as a primary model to illustrate our framework, but further verified the results using other two data sets. As more and more genome-wide ChIP-Seq and RNA-Seq data becomes available in the near future, our methods of data integration have various potential applications.

Computational Discovery of Cancer Immunotherapy Targets by Intercellular CRISPR Screens.

Cancer immunotherapy targets the interplay between immune and cancer cells. In particular, interactions between cytotoxic T lymphocytes (CTLs) and cancer cells, such as PD-1 (PDCD1) binding PD-L1 (CD274), are crucial for cancer cell clearance. However, immune checkpoint inhibitors targeting these interactions are effective only in a subset of patients, requiring the identification of novel immunotherapy targets. Genome-wide clustered regularly interspaced short palindromic repeats (CRISPR) screening in either cancer or immune cells has been employed to discover regulators of immune cell function. However, CRISPR screens in a single cell type complicate the identification of essential intercellular interactions. Further, pooled screening is associated with high noise levels. Herein, we propose intercellular CRISPR screens, a computational approach for the analysis of genome-wide CRISPR screens in every interacting cell type for the discovery of intercellular interactions as immunotherapeutic targets. We used two publicly available genome-wide CRISPR screening datasets obtained while triple-negative breast cancer (TNBC) cells and CTLs were interacting. We analyzed 4825 interactions between 1391 ligands and receptors on TNBC cells and CTLs to evaluate their effects on CTL function. Intercellular CRISPR screens discovered targets of approved drugs, a few of which were not identifiable in single datasets. To evaluate the method's performance, we used data for cytokines and costimulatory molecules as they constitute the majority of immunotherapeutic targets. Combining both CRISPR datasets improved the recall of discovering these genes relative to using single CRISPR datasets over two-fold. Our results indicate that intercellular CRISPR screens can suggest novel immunotherapy targets that are not obtained through individual CRISPR screens. The pipeline can be extended to other cancer and immune cell types to discover important intercellular interactions as potential immunotherapeutic targets.

dialogi: Utilising NLP With Chemical and Disease Similarities to Drive the Identification of Drug-Induced Liver Injury Literature.

Drug-Induced Liver Injury (DILI), despite its low occurrence rate, can cause severe side effects or even lead to death. Thus, it is one of the leading causes for terminating the development of new, and restricting the use of already-circulating, drugs. Moreover, its multifactorial nature, combined with a clinical presentation that often mimics other liver diseases, complicate the identification of DILI-related (or "positive") literature, which remains the main medium for sourcing results from the clinical practice and experimental studies. This work-contributing to the "Literature AI for DILI Challenge" of the Critical Assessment of Massive Data Analysis (CAMDA) 2021- presents an automated pipeline for distinguishing between DILI-positive and negative publications. We used Natural Language Processing (NLP) to filter out the uninformative parts of a text, and identify and extract mentions of chemicals and diseases. We combined that information with small-molecule and disease embeddings, which are capable of capturing chemical and disease similarities, to improve classification performance. The former were directly sourced from the Chemical Checker (CC). For the latter, we collected data that encode different aspects of disease similarity from the National Library of Medicine's (NLM) Medical Subject Headings (MeSH) thesaurus and the Comparative Toxicogenomics Database (CTD). Following a similar procedure as the one used in the CC, vector representations for diseases were learnt and evaluated. Two Neural Network (NN) classifiers were developed: a baseline model that accepts texts as input and an augmented, extended, model that also utilises chemical and disease embeddings. We trained, validated, and tested the classifiers through a Nested Cross-Validation (NCV) scheme with 10 outer and 5 inner folds. During this, the baseline and extended models performed virtually identically, with F1-scores of 95.04 ± 0.61% and 94.80 ± 0.41%, respectively. Upon validation on an external, withheld, dataset that is meant to assess classifier generalisability, the extended model achieved an F1-score of 91.14 ± 1.62%, outperforming its baseline counterpart which received a lower score of 88.30 ± 2.44%. We make further comparisons between the classifiers and discuss future improvements and directions, including utilising chemical and disease embeddings for visualisation and exploratory analysis of the DILI-positive literature.

DILI C : An AI-Based Classifier to Search for Drug-Induced Liver Injury Literature.

Drug-induced liver injury (DILI) is a class of adverse drug reactions (ADR) that causes problems in both clinical and research settings. It is the most frequent cause of acute liver failure in the majority of Western countries and is a major cause of attrition of novel drug candidates. Manual trawling of the literature is the main route of deriving information on DILI from research studies. This makes it an inefficient process prone to human error. Therefore, an automatized AI model capable of retrieving DILI-related articles from the huge ocean of literature could be invaluable for the drug discovery community. In this study, we built an artificial intelligence (AI) model combining the power of natural language processing (NLP) and machine learning (ML) to address this problem. This model uses NLP to filter out meaningless text (e.g., stop words) and uses customized functions to extract relevant keywords such as singleton, pair, and triplet. These keywords are processed by an apriori pattern mining algorithm to extract relevant patterns which are used to estimate initial weightings for a ML classifier. Along with pattern importance and frequency, an FDA-approved drug list mentioning DILI adds extra confidence in classification. The combined power of these methods builds a DILI classifier (DILI C ), with 94.91% cross-validation and 94.14% external validation accuracy. To make DILI C as accessible as possible, including to researchers without coding experience, an R Shiny app capable of classifying single or multiple entries for DILI is developed to enhance ease of user experience and made available at https://researchmind.co.uk/diliclassifier/. Additionally, a GitHub link (https://github.com/sanjaysinghrathi/DILI-Classifier) for app source code and ISMB extended video talk (https://www.youtube.com/watch?v=j305yIVi_f8) are available as supplementary materials.