KYKevin Yuan

Kevin Yuan

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Kevin Yuan has a total of 66 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Machine learning and clinician predictions of antibiotic resistance in Enterobacterales bloodstream infections

    by Kevin Yuan, Augustine Luk, Jia Wei, A. Sarah Walker, Tingting Zhu, David W. Eyre


  • Open Access

    Predicting individual patient and hospital-level discharge using machine learning

    by Jia Wei, Jiandong Zhou, Zizheng Zhang, Kevin Yuan, Qingze Gu, Augustine Luk, Andrew J. Brent, David A. Clifton, A. Sarah Walker, David W. Eyre

    Background: Accurately predicting hospital discharge events could help improve patient flow and the efficiency of healthcare delivery. However, using machine learning and diverse electronic health record (EHR) data for this task remains incompletely explored. Methods: We used EHR data from February-2017 to January-2020 from Oxfordshire, UK to predict hospital discharges in the next 24 h. We fitted separate extreme gradient boosting models for elective and emergency admissions, trained on the first two years of data and tested on the final year of data. We examined individual-level and hospital-level model performance and evaluated the impact of training data size and recency, prediction time, and performance in subgroups. Results: Our models achieve AUROCs of 0.87 and 0.86, AUPRCs of 0.66 and 0.64, and F1 scores of 0.61 and 0.59 for elective and emergency admissions, respectively. These models outperform a logistic regression model using the same features and are substantially better than a baseline logistic regression model with more limited features. Notably, the relative performance increase from adding additional features is greater than the increase from using a sophisticated model. Aggregating individual probabilities, daily total discharge estimates are accurate with mean absolute errors of 8.9% (elective) and 4.9% (emergency). The most informative predictors include antibiotic prescriptions, medications, and hospital capacity factors. Performance remains robust across patient subgroups and different training strategies, but is lower in patients with longer admissions and those who died in hospital. Conclusions: Our findings highlight the potential of machine learning in optimising hospital patient flow and facilitating patient care and recovery.


  • Open Access

    Transformers and large language models are efficient feature extractors for electronic health record studies

    by Kevin Yuan, Chang Ho Yoon, Qingze Gu, Henry Munby, Sarah Walker, Tingting Zhu, David Eyre

    While unstructured free-text data is abundant in electronic health records, challenges in accurate and scalable information extraction often result in a preference for incorporating information from less specific clinical codes instead. We investigate the efficacy of modern natural language processing methods (NLP) and large language models (LLMs) in supervised and few-shot scenarios to extract features from 938,150 hospital antibiotic prescriptions from Oxfordshire, UK. A subset of 4000 most frequent indications for antibiotic use was labelled by clinical researchers into 11 categories, describing the infection source/clinical syndrome, for model training. On separate internal (n = 2000) and external test datasets (n = 2000), the fine-tuned domain-specific Bio + Clinical BERT model averaged an F1 score of 0.97 and 0.98 respectively across the classes and outperformed traditional regex (F1 = 0.71 and 0.74) and n-grams/XGBoost (F1 = 0.86 and 0.84). A few-shot OpenAI GPT4 model achieved F1 scores of 0.71 and 0.86 without using labelled training data and a fine-tuned GPT3.5 model F1 scores of 0.95 and 0.97. Finetuned BERT-based transformer models currently outperform LLMs for structured tasks, while few shot LLMs match the performance of traditional NLP without the need for labelling. Comparing infection sources extracted from ICD10 codes to those parsed from free-text indications, free-text indications revealed 31% more specific infection sources. With their high accuracy, modern transformer-based models have the potential to be used widely throughout medicine to structure free-text records, providing more granular information than clinical codes, facilitating better research and patient care.

  • Open Access

    Leveraging transformers and large language models with antimicrobial prescribing data to predict sources of infection for electronic health record studies

    by Kevin Yuan, Chang Ho Yoon, Qingze Gu, Henry Munby, A Sarah Walker, Tingting Zhu, David W Eyre

    Background Electronic health records frequently contain extensive unstructured free-text data, but extracting information accurately from these data and at scale is challenging. Using free-text from antibiotic prescribing data as an example, we investigate the performance of modern natural language processing methods (NLP) and large language models (LLMs) as tools for extracting features from medical records. Methods We used 938,150 hospital antibiotic prescriptions from Oxfordshire, UK. The 4000 most frequently used free-text indications justifying antibiotic use were labelled by clinical researchers into 11 categories describing the infection source/clinical syndrome being treated and used for model training. Traditional classification methods, fuzzy regex matching and n-grams with XGBoost, were compared against modern transformer models: we fine-tuned generic and domain-specific BERT models, fine-tuned GPT3.5, and investigated few-shot learning with GPT4. Models were evaluated on internal and external test datasets (2000 prescriptions each). Infection sources determined from ICD10 codes were also used for comparisons. Results In internal and external test datasets, the fine-tuned domain-specific Bio+Clinical BERT model averaged an F1 score of 0.97 and 0.98 respectively across the classes and outperformed the traditional regex (F1=0.71 and 0.74) and n-grams/XGBoost (F1=0.86 and 0.84). OpenAI’s GPT4 model achieved F1 scores of 0.71 and 0.86 without using labelled training data and a fine-tuned GPT3.5 model F1 scores of 0.95 and 0.97. Comparing infection sources extracted from ICD10 codes to those parsed from free-text indications, free-text indications revealed 31% more specific infection sources. Conclusion Modern transformer-based models can efficiently and accurately categorise semi-structured free-text in medical records, such as prescription free-text. Finetuned local transformer models outperform LLMs currently for structured tasks. Few shot LLMs match the performance of traditional NLP without the need for labelling. Transformer-based models have the potential to be used widely throughout medicine to analyse medical records more accurately, facilitating beter research and patient care.


  • Open Access

    Defining normal inflammatory marker and vital sign responses to suspected bloodstream infection in adults with positive and negative blood cultures

    by Qingze Gu, Jia Wei, Chang Ho Yoon, Kevin Yuan, Nicola Jones, Andrew J Brent, Martin J Llewelyn, Tim Ea Peto, Koen B Pouwels, David W Eyre, et al.

    Background Patients respond differently to bloodstream infection (BSI) and associated antibiotic treatment, for many reasons, including different causative pathogens, sources of infection, and patient characteristics. This heterogeneity can hamper use of different clinical parameters to track treatment response as the same absolute values, or even change from presentation, may have different implications, depending on the expected trajectory, which is often incompletely understood. Methods We included patients ≥16y from Oxford University Hospitals (01-January-2016 to 28-June-2021) with any blood culture taken, grouping cultures into suspected BSI episodes (14-day de-duplication). We used linear and latent class mixed models to estimate trajectories in C-reactive protein (CRP), white blood count, heart rate, respiratory rate and temperature and identify subgroups with heterogenous CRP responses. Centile charts for expected CRP responses were constructed via the lambda-mu-sigma method. Findings 88,348 suspected BSI episodes occurred in 60,647 adults; 6,910(7.8%) were culture-positive with a probable pathogen (1,914[2.2%] Gram-positive, 3,736[4.2%] Gram-negative, 1,260[1.4%] other pathogens/polymicrobial), 4,307(4.9%) contained potential contaminants, and 77,131(87.3%) were culture-negative. Overall, CRP levels generally peaked between day 1-2 after blood culture collection, with varying responses for different pathogens and infection sources in adjusted models (interaction pInterpretation Although infection sources and pathogens are associated with varying responses to BSI, there is distinct underlying heterogeneity in responses. The centile reference charts developed could facilitate more precise tracking of recovery, enable identification of patients not recovering as expected, and help personalise infection management.

  • Open Access

    TF-Prioritizer: a Java pipeline to prioritize condition-specific transcription factors

    by Markus Hoffmann, Nico Trummer, Leon Schwartz, Jakub Jankowski, Hye Kyung Lee, Lina-Liv Willruth, Olga Lazareva, Kevin Sheng-Po Yuan, Nina Baumgarten, Florian Schmidt, et al.

    BackgroundEukaryotic gene expression is controlled by cis-regulatory elements (CREs), including promoters and enhancers, which are bound by transcription factors (TFs). Differential expression of TFs and their binding affinity at putative CREs determine tissue- and developmental-specific transcriptional activity. Consolidating genomic datasets can offer further insights into the accessibility of CREs, TF activity, and, thus, gene regulation. However, the integration and analysis of multimodal datasets are hampered by considerable technical challenges. While methods for highlighting differential TF activity from combined chromatin state data (e.g., chromatin immunoprecipitation [ChIP], ATAC, or DNase sequencing) and RNA sequencing data exist, they do not offer convenient usability, have limited support for large-scale data processing, and provide only minimal functionality for visually interpreting results. ResultsWe developed TF-Prioritizer, an automated pipeline that prioritizes condition-specific TFs from multimodal data and generates an interactive web report. We demonstrated its potential by identifying known TFs along with their target genes, as well as previously unreported TFs active in lactating mouse mammary glands. Additionally, we studied a variety of ENCODE datasets for cell lines K562 and MCF-7, including 12 histone modification ChIP sequencing as well as ATAC and DNase sequencing datasets, where we observe and discuss assay-specific differences. ConclusionTF-Prioritizer accepts ATAC, DNase, or ChIP sequencing and RNA sequencing data as input and identifies TFs with differential activity, thus offering an understanding of genome-wide gene regulation, potential pathogenesis, and therapeutic targets in biomedical research.


  • Open Access

    TF-Prioritizer: a Java pipeline to prioritize condition-specific transcription factors

    by Markus Hoffmann, Nico Trummer, Leon Schwartz, Jakub Jankowski, Hye Kyung Lee, Lina-Liv Willruth, Olga Lazareva, Kevin Yuan, Nina Baumgarten, Florian Schmidt, et al.

    Eukaryotic gene expression is controlled by cis-regulatory elements (CREs), including promoters and enhancers, which are bound by transcription factors (TFs). Differential expression of TFs and their binding affinity at putative CREs determine tissue- and developmental-specific transcriptional activity. Consolidating genomic datasets can offer further insights into the accessibility of CREs, TF activity, and, thus, gene regulation. However, the integration and analysis of multimodal datasets are hampered by considerable technical challenges. While methods for highlighting differential TF activity from combined chromatin state data (e.g., chromatin immunoprecipitation [ChIP], ATAC, or DNase sequencing) and RNA sequencing data exist, they do not offer convenient usability, have limited support for large-scale data processing, and provide only minimal functionality for visually interpreting results. We developed TF-Prioritizer, an automated pipeline that prioritizes condition-specific TFs from multimodal data and generates an interactive web report. We demonstrated its potential by identifying known TFs along with their target genes, as well as previously unreported TFs active in lactating mouse mammary glands. Additionally, we studied a variety of ENCODE datasets for cell lines K562 and MCF-7, including 12 histone modification ChIP sequencing as well as ATAC and DNase sequencing datasets, where we observe and discuss assay-specific differences. TF-Prioritizer accepts ATAC, DNase, or ChIP sequencing and RNA sequencing data as input and identifies TFs with differential activity, thus offering an understanding of genome-wide gene regulation, potential pathogenesis, and therapeutic targets in biomedical research.


  • Open Access

    TF-Prioritizer: a java pipeline to prioritize condition-specific transcription factors

    by Markus Hoffmann, Nico Trummer, Leon Schwartz, Jakub Jankowski, Hye Kyung Lee, Lina-Liv Willruth, Olga Lazareva, Kevin Yuan, Nina Baumgarten, Florian Schmidt, et al.

    Background Eukaryotic gene expression is controlled by cis-regulatory elements (CREs) including promoters and enhancers which are bound by transcription factors (TFs). Differential expression of TFs and their putative binding sites on CREs cause tissue and developmental-specific transcriptional activity. Consolidating genomic data sets can offer further insights into the accessibility of CREs, TF activity, and thus gene regulation. However, the integration and analysis of multi-modal data sets are hampered by considerable technical challenges. While methods for highlighting differential TF activity from combined ChIP-seq and RNA-seq data exist, they do not offer good usability, have limited support for large-scale data processing, and provide only minimal functionality for visual result interpretation. Results We developed TF-Prioritizer, an automated java pipeline to prioritize condition-specific TFs derived from multi-modal data. TF-Prioritizer creates an interactive, feature-rich, and user-friendly web report of its results. To showcase the potential of TF-Prioritizer, we identified known active TFs (e.g., Stat5, Elf5, Nfib, Esr1), their target genes (e.g., milk proteins and cell-cycle genes), and newly classified lactating mammary gland TFs (e.g., Creb1, Arnt). Conclusion TF-Prioritizer accepts ChIP-seq and RNA-seq data, as input and suggests TFs with differential activity, thus offering an understanding of genome-wide gene regulation, potential pathogenesis, and therapeutic targets in biomedical research.

  • Open Access

    Targeting infectious diseases: The CoVex example

    by Sepideh Sadegh, Julian Matschinske, David B. Blumenthal, Gihanna Galindez, Tim Kacprowski, Markus List, Reza Nasirigerdeh, Mhaned Oubounyt, Andreas Pichlmair, Tim Daniel Rose, et al.

    In a pandemic, such as the one caused by the coronavirus in 2020, fast action is required to provide insights into the disease mechanisms and to find potential drug targets to slow the progress of the disease and lower mortality. We developed CoVex, an interactive online platform for the exploration of the SARS-CoV-2 host interactome and the identification of drug candidates. CoVex integrates the virus-human protein interactions, human protein-protein interactions, and drug-target interactions, and allows for visual exploration of the data. The exploration is guided by network-based systems medicine algorithms that enable the assembly of ranked lists of candidates for already approved drugs. Unlike novel drugs, approved drugs have the advantage that most of their side effects and contraindications are known. Additionally, they are already available and can be distributed quickly, which makes them promising candidates to stop the uncontrolled spread of the disease early. CoVex is an example of how this could be achieved in future pandemics and provides valuable lessons on what is required in such a situation to maximize the potential of computational approaches that can significantly speed up drug discovery.

  • BiCoN: network-constrained biclustering of patients and omics data

    by Olga Lazareva, Stefan Canzar, Kevin Yuan, Jan Baumbach, David B Blumenthal, Paolo Tieri, Tim Kacprowski, Markus List

    Motivation: Unsupervised learning approaches are frequently used to stratify patients into clinically relevant subgroups and to identify biomarkers such as disease-associated genes. However, clustering and biclustering techniques are oblivious to the functional relationship of genes and are thus not ideally suited to pinpoint molecular mechanisms along with patient subgroups. Results: We developed the network-constrained biclustering approach Biclustering Constrained by Networks (BiCoN) which (i) restricts biclusters to functionally related genes connected in molecular interaction networks and (ii) maximizes the difference in gene expression between two subgroups of patients. This allows BiCoN to simultaneously pinpoint molecular mechanisms responsible for the patient grouping. Network-constrained clustering of genes makes BiCoN more robust to noise and batch effects than typical clustering and biclustering methods. BiCoN can faithfully reproduce known disease subtypes as well as novel, clinically relevant patient subgroups, as we could demonstrate using breast and lung cancer datasets. In summary, BiCoN is a novel systems medicine tool that combines several heuristic optimization strategies for robust disease mechanism extraction. BiCoN is well-documented and freely available as a python package or a web interface. Availability and implementation: PyPI package: https://pypi.org/project/bicon. Web interface: https://exbio.wzw.tum.de/bicon. Supplementary information: Supplementary data are available at Bioinformatics online.