Xiruo Ding

All Publications

• An adult proteomic signature associated with very low birth weight and chronic disease risk. Med (New York, N.Y.) Huang, C. Y., Seong, D., Reiss, J., Shu, C. H., Sadeghi, M. A., Chung, P., Yang, L. K., Xue, L., Fang, Y., Xu, R., Ding, X., Sheybani, S., Han, L., Mawuli, B. C., Ghanem, M., Kiamari, M., Chang, A. L., James, T., Mataraso, S. J., Espinosa, C., Berson, E., Tsao, P. N., Angst, M. S., Gaudiliere, B., Prince, L., Shaw, G., Stevenson, D., Aghaeepour, N. 2026: 101297

  Abstract

  Adults born at very low birth weight (VLBW, <1.5 kg) have elevated risks of chronic disease, but it is unclear whether adult plasma proteomics contains a detectable signature associated with recalled VLBW.We analyzed 2,923 plasma proteins in 26,061 UK Biobank participants (median sampling age 56.4 years) with recalled birth weight, including 368 with VLBW and 25,693 with normal birth weight. We derived a machine learning proteomic prematurity index (PPI) distinguishing these groups and evaluated associations with 425 adult diagnoses, discrimination relative to birth weight and electronic health record (EHR) information, incident diseases after proteomic sampling, and exploratory biological interpretation.The PPI was associated with recalled VLBW (odds ratio [OR], 7.32; 95% confidence interval [95% CI], 4.01-13.42) and remained independently associated after adjustment for EHR-derived VLBW probability (OR, 5.82; 95% CI, 3.99-8.47). Recalled VLBW was associated with adult disease burden, and PPI showed higher discrimination than birth weight for 370 of 425 diseases. Incident-only analyses preserved the cross-disease hazard ratio patterns, while fewer events and adult factor adjustment attenuated some associations. Exploratory biological annotations paralleled the disease patterns. Among participants with normal birth weight, higher PPI was associated with disease burden and subsequent disease risk, indicating that PPI is not VLBW specific.Adult plasma proteomics identifies a signature associated with recalled VLBW that provides information about adult disease risk beyond birth weight and EHR-derived measures. These findings are hypothesis generating and do not establish a direct molecular consequence of prematurity. External validation is required before clinical application.The March of Dimes, the Alfred E. Mann Foundation, the Universe of Mind, and the Bill & Melinda Gates Foundation (INV-037517 and INV-076306).

  View details for DOI 10.1016/j.medj.2026.101297

  View details for PubMedID 42748912

• Tailoring task arithmetic to address bias in models trained on multi-institutional datasets. Journal of biomedical informatics Ding, X., Sheng, Z., Hur, B., Tauscher, J., Ben-Zeev, D., Yetisgen, M., Pakhomov, S., Cohen, T. 2025; 168: 104858

  Abstract

  OBJECTIVE: Multi-institutional datasets are widely used for machine learning from clinical data, to increase dataset size and improve generalization. However, deep learning models in particular may learn to recognize the source of a data element, leading to biased predictions. For example, deep learning models for image recognition trained on chest radiographs with COVID-19 positive and negative examples drawn from different data sources can respond to indicators of provenance (e.g., radiological annotations outside the lung area per institution-specific practices) rather than pathology, generalizing poorly beyond their training data. Bias of this sort, called confounding by provenance, is of concern in natural language processing (NLP) because provenance indicators (e.g., institution-specific section headers, or region-specific dialects) are pervasive in language data. Prior work on addressing such bias has focused on statistical methods, without providing a solution for deep learning models for NLP.METHODS: Recent work in representation learning has shown that representing the weights of a trained deep network as task vectors allows for their arithmetic composition to govern model capabilities towards desired behaviors. In this work, we evaluate the extent to which reducing a model's ability to distinguish between contributing sites with such task arithmetic can mitigate confounding by provenance. To do so, we propose two model-agnostic methods, Task Arithmetic for Provenance Effect Reduction (TAPER) and Dominance-Aligned Polarized Provenance Effect Reduction (DAPPER), extending the task vectors approach to a novel problem domain.RESULTS: Evaluation on three datasets shows improved robustness to confounding by provenance for both RoBERTa and Llama-2 models with the task vector approach, with improved performance at the extremes of distribution shift.CONCLUSION: This work emphasizes the importance of adjusting for confounding by provenance, especially in extreme cases of the shift. In use of deep learning models, DAPPER and TAPER show efficiency in mitigating such bias. They provide a novel mitigation strategy for confounding by provenance, with broad applicability to address other sources of bias in composite clinical data sets. Source code is available within the DeconDTN toolkit: https://github.com/LinguisticAnomalies/DeconDTN-toolkit.

  View details for DOI 10.1016/j.jbi.2025.104858

  View details for PubMedID 40494422

• A flexible symbolic regression method for constructing interpretable clinical prediction models NPJ DIGITAL MEDICINE La Cava, W. G. G., Lee, P. C. C., Ajmal, I., Ding, X., Solanki, P., Cohen, J. B. B., Moore, J. H. H., Herman, D. S. S. 2023; 6 (1): 107

  Abstract

  Machine learning (ML) models trained for triggering clinical decision support (CDS) are typically either accurate or interpretable but not both. Scaling CDS to the panoply of clinical use cases while mitigating risks to patients will require many ML models be intuitively interpretable for clinicians. To this end, we adapted a symbolic regression method, coined the feature engineering automation tool (FEAT), to train concise and accurate models from high-dimensional electronic health record (EHR) data. We first present an in-depth application of FEAT to classify hypertension, hypertension with unexplained hypokalemia, and apparent treatment-resistant hypertension (aTRH) using EHR data for 1200 subjects receiving longitudinal care in a large healthcare system. FEAT models trained to predict phenotypes adjudicated by chart review had equivalent or higher discriminative performance (p < 0.001) and were at least three times smaller (p < 1 × 10-6) than other potentially interpretable models. For aTRH, FEAT generated a six-feature, highly discriminative (positive predictive value = 0.70, sensitivity = 0.62), and clinically intuitive model. To assess the generalizability of the approach, we tested FEAT on 25 benchmark clinical phenotyping tasks using the MIMIC-III critical care database. Under comparable dimensionality constraints, FEAT's models exhibited higher area under the receiver-operating curve scores than penalized linear models across tasks (p < 6 × 10-6). In summary, FEAT can train EHR prediction models that are both intuitively interpretable and accurate, which should facilitate safe and effective scaling of ML-triggered CDS to the panoply of potential clinical use cases and healthcare practices.

  View details for DOI 10.1038/s41746-023-00833-8

  View details for Web of Science ID 001000875800001

  View details for PubMedID 37277550

  View details for PubMedCentralID PMC10241925

• From benchmark to bedside: transfer learning from social media to patient-provider text messages for suicide risk prediction JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION Burkhardt, H. A., Ding, X., Kerbrat, A., Comtois, K., Cohen, T. 2023; 30 (6): 1068-1078

  Abstract

  Compared to natural language processing research investigating suicide risk prediction with social media (SM) data, research utilizing data from clinical settings are scarce. However, the utility of models trained on SM data in text from clinical settings remains unclear. In addition, commonly used performance metrics do not directly translate to operational value in a real-world deployment. The objectives of this study were to evaluate the utility of SM-derived training data for suicide risk prediction in a clinical setting and to develop a metric of the clinical utility of automated triage of patient messages for suicide risk.Using clinical data, we developed a Bidirectional Encoder Representations from Transformers-based suicide risk detection model to identify messages indicating potential suicide risk. We used both annotated and unlabeled suicide-related SM posts for multi-stage transfer learning, leveraging customized contemporary learning rate schedules. We also developed a novel metric estimating predictive models' potential to reduce follow-up delays with patients in distress and used it to assess model utility.Multi-stage transfer learning from SM data outperformed baseline approaches by traditional classification performance metrics, improving performance from 0.734 to a best F1 score of 0.797. Using this approach for automated triage could reduce response times by 15 minutes per urgent message.Despite differences in data characteristics and distribution, publicly available SM data benefit clinical suicide risk prediction when used in conjunction with contemporary transfer learning techniques. Estimates of time saved due to automated triage indicate the potential for the practical impact of such models when deployed as part of established suicide prevention interventions.This work demonstrates a pathway for leveraging publicly available SM data toward improving risk assessment, paving the way for better clinical care and improved clinical outcomes.

  View details for DOI 10.1093/jamia/ocad062

  View details for Web of Science ID 000970328900001

  View details for PubMedID 37043748

  View details for PubMedCentralID PMC10198538

• Automated Detection of Cognitive Distortions in Text Exchanges Between Clinicians and People With Serious Mental Illness PSYCHIATRIC SERVICES Tauscher, J. S., Lybarger, K., Ding, X., Chander, A., Hudenko, W. J., Cohen, T., Ben-Zeev, D. 2023; 74 (4): 407-410

  Abstract

  The authors tested whether natural language processing (NLP) methods can detect and classify cognitive distortions in text messages between clinicians and people with serious mental illness as effectively as clinically trained human raters.Text messages (N=7,354) were collected from 39 clients in a randomized controlled trial of a 12-week texting intervention. Clinical annotators labeled messages for common cognitive distortions: mental filtering, jumping to conclusions, catastrophizing, "should" statements, and overgeneralizing. Multiple NLP classification methods were applied to the same messages, and performance was compared.A tuned model that used bidirectional encoder representations from transformers (F1=0.62) achieved performance comparable to that of clinical raters in classifying texts with any distortion (F1=0.63) and superior to that of other models.NLP methods can be used to effectively detect and classify cognitive distortions in text exchanges, and they have the potential to inform scalable automated tools for clinical support during message-based care for people with serious mental illness.

  View details for DOI 10.1176/appi.ps.202100692

  View details for Web of Science ID 000972484400011

  View details for PubMedID 36164769

• Improving Classification of Infrequent Cognitive Distortions: Domain-Specific Model vs. Data Augmentation Ding, X., Lybarger, K., Tauscher, J. S., Cohen, T., ASSOC COMPUTAT LINGUIST ASSOC COMPUTATIONAL LINGUISTICS-ACL. 2022: 68-75

  View details for Web of Science ID 000860760300009

• Augmenting aer2vec: Enriching distributed representations of adverse event report data with orthographic and lexical information JOURNAL OF BIOMEDICAL INFORMATICS Ding, X., Mower, J., Subramanian, D., Cohen, T. 2021; 119: 103833

  Abstract

  Adverse Drug Events (ADEs) are prevalent, costly, and sometimes preventable. Post-marketing drug surveillance aims to monitor ADEs that occur after a drug is released to market. Reports of such ADEs are aggregated by reporting systems, such as the Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS). In this paper, we consider the topic of how best to represent data derived from reports in FAERS for the purpose of detecting post-marketing surveillance signals, in order to inform regulatory decision making. In our previous work, we developed aer2vec, a method for deriving distributed representations (concept embeddings) of drugs and side effects from ADE reports, establishing the utility of distributional information for pharmacovigilance signal detection. In this paper, we advance this line of research further by evaluating the utility of encoding orthographic and lexical information. We do so by adapting two Natural Language Processing methods, subword embedding and vector retrofitting, which were developed to encode such information into word embeddings. Models were compared for their ability to distinguish between positive and negative examples in a set of manually curated drug/ADE relationships, with both aer2vec enhancements offering advantages in performances over baseline models, and best performance obtained when retrofitting and subword embeddings were applied in concert. In addition, this work demonstrates that models leveraging distributed representations do not require extensive manual preprocessing to perform well on this pharmacovigilance signal detection task, and may even benefit from information that would otherwise be lost during the normalization and standardization process.

  View details for DOI 10.1016/j.jbi.2021.103833

  View details for Web of Science ID 000674512900008

  View details for PubMedID 34111555

  View details for PubMedCentralID PMC8260467

• A maximum likelihood approach to electronic health record phenotyping using positive and unlabeled patients JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION Zhang, L., Ding, X., Ma, Y., Muthu, N., Ajmal, I., Moore, J. H., Herman, D. S., Chen, J. 2020; 27 (1): 119-126

  Abstract

  Phenotyping patients using electronic health record (EHR) data conventionally requires labeled cases and controls. Assigning labels requires manual medical chart review and therefore is labor intensive. For some phenotypes, identifying gold-standard controls is prohibitive. We developed an accurate EHR phenotyping approach that does not require labeled controls.Our framework relies on a random subset of cases, which can be specified using an anchor variable that has excellent positive predictive value and sensitivity independent of predictors. We proposed a maximum likelihood approach that efficiently leverages data from the specified cases and unlabeled patients to develop logistic regression phenotyping models, and compare model performance with existing algorithms.Our method outperformed the existing algorithms on predictive accuracy in Monte Carlo simulation studies, application to identify hypertension patients with hypokalemia requiring oral supplementation using a simulated anchor, and application to identify primary aldosteronism patients using real-world cases and anchor variables. Our method additionally generated consistent estimates of 2 important parameters, phenotype prevalence and the proportion of true cases that are labeled.Upon identification of an anchor variable that is scalable and transferable to different practices, our approach should facilitate development of scalable, transferable, and practice-specific phenotyping models.Our proposed approach enables accurate semiautomated EHR phenotyping with minimal manual labeling and therefore should greatly facilitate EHR clinical decision support and research.

  View details for DOI 10.1093/jamia/ocz170

  View details for Web of Science ID 000548300200015

  View details for PubMedID 31722396

  View details for PubMedCentralID PMC6913222

• Designing risk prediction models for ambulatory no-shows across different specialties and clinics JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION Ding, X., Gellad, Z. F., Mather, C., Barth, P., Poon, E. G., Newman, M., Goldstein, B. A. 2018; 25 (8): 924-930

  Abstract

  As available data increases, so does the opportunity to develop risk scores on more refined patient populations. In this paper we assessed the ability to derive a risk score for a patient no-showing to a clinic visit.Using data from 2 264 235 outpatient appointments we assessed the performance of models built across 14 different specialties and 55 clinics. We used regularized logistic regression models to fit and assess models built on the health system, specialty, and clinic levels. We evaluated fits based on their discrimination and calibration.Overall, the results suggest that a relatively robust risk score for patient no-shows could be derived with an average C-statistic of 0.83 across clinic level models and strong calibration. Moreover, the clinic specific models, even with lower training set sizes, often performed better than the more general models. Examination of the individual models showed that risk factors had different degrees of predictability across the different specialties. Implementation of optimal modeling strategies would lead to capturing an additional 4819 no-shows per-year.Overall, this work highlights both the opportunity for and the importance of leveraging the available electronic health record data to develop more refined risk models.

  View details for DOI 10.1093/jamia/ocy002

  View details for Web of Science ID 000440955200002

  View details for PubMedID 29444283

  View details for PubMedCentralID PMC6077778

• An analogous wood barrel theory to explain the occurrence of hormesis: A case study of sulfonamides and erythromycin on <i>Escherichia coli</i> growth PLOS ONE Wang, D., Lin, Z., Wang, T., Ding, X., Liu, Y. 2017; 12 (7)

  View details for DOI 10.1371/journal.pone.0181321

  View details for Web of Science ID 000443975500044

• The Comparison of the Combined Toxicity between Gram-negative and Gram-positive Bacteria: a Case Study of Antibiotics and Quorum-sensing Inhibitors MOLECULAR INFORMATICS Wang, D., Lin, Z., Ding, X., Hu, J., Liu, Y. 2016; 35 (2): 54-61

  Abstract

  Quorum-sensing inhibitors (QSIs) are being used increasingly in diverse fields, and are likely to end up in the environment, where they may encounter the antibiotics and consequently cause joint effects on biological systems. However, the potential joint effects of QSIs and antibiotics have received little attention. In this study, the joint effects of antibiotics, represented by sulfonamides (SAs) and penicillin, as well as three potential QSIs, were investigated using both Gram-negative (Escherichia coli, E. coli) and Gram-positive bacteria (Bacillus subtilis, B. subtilis). It was found that E. coli tend to be more sensitive to the individual drugs than B. subtilis, whereas the joint effects on the two bacteria showed no difference regarding the same combination of antibiotics and QSIs. In general, SAs presented additive effects with γ-Valerolactone and 2-Pyrrolidinone, but antagonistic effects with L-(+)-Prolinol; penicillin exhibited antagonistic effects with all three QSIs. Moreover, it was found that the rate of resistance in E. coli against the individual antibiotics was reduced through the addition of the QSIs, which suggests a promising use of the QSIs in the bacterial infection treatment. This study also offers a valuable reference for the risk assessment of the antibiotics and QSIs in the real environment.

  View details for DOI 10.1002/minf.201500061

  View details for Web of Science ID 000370173700002

  View details for PubMedID 27491790