Fei Yang

Fei Yang


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
20 November 2025 in Functional & Integrative Genomics
Verified
Journal Article
Open access
PDF file
HHHaoyang Hu
Haoyang Hu
MLMao Li
Mao Li
YWYan Wang
Yan Wang
YLYang Liu
Yang Liu
HZHong Zhao
Hong Zhao
FYFei Yang
Fei Yang
Mild traumatic brain injury (mTBI) is a prevalent condition accounting for over 70% of all traumatic brain injury (TBI) cases, and it is a major cause of posttraumatic cognitive impairment. Ferroptosis, a form of regulated cell death characterized by iron-dependent lipid peroxidation, has been implicated in the pathophysiology of mTBI. However, its precise role in mTBI - induced cognitive dysfunction and potential therapeutic strategies remain unclear. This study aimed to investigate the neuroprotective effects of mesenchymal stem cell - derived exosomes (MSC - Exos) against ferroptosis and cognitive dysfunction following mTBI. We established an mTBI rat model and administered MSC - Exos at different doses. Behavioral assessments, histological and molecular biological analyses, and bioinformatics approaches were used. The results showed that mTBI rats exhibited cognitive impairments, increased lipid peroxidation, and reduced GPX4 expression. MSC - Exos treatment improved cognitive function in a dose - dependent manner, attenuated lipid peroxidation, and restored GPX4 expression. Transcriptomic and bioinformatic analyses revealed that MSC - Exos activated the PI3K/AKT/mTOR signaling pathway, which upregulated GPX4 expression and inhibited ferroptosis. In conclusion, MSC - Exos alleviate cognitive deficits after mTBI by inhibiting ferroptosis via PI3K/AKT/mTOR - mediated upregulation of GPX4, providing a novel therapeutic strategy for mTBI.

22 October 2025 in CNS Neuroscience & Therapeutics
Verified
Journal Article
Open access
YLYang Liu
Yang Liu
ZZZihan Zhang
Zihan Zhang
ZJZihan Jing
Zihan Jing
BSBinbin Sun
Binbin Sun
HZHong Zhao
Hong Zhao
FYFei Yang
Fei Yang
Background: Critical illness polyneuropathy (CIP) is a common cause of weakness in critically ill patients, and early diagnosis and treatment are essential. Conventional comprehensive electrophysiological testing requires 60–90 min and can be invasive, limiting its utility in the intensive care setting. This study aimed to develop a rapid, efficient, and minimally invasive diagnostic model for CIP. Methods: Patients who met the inclusion criteria were recruited from the intensive care units (ICU) of the First Medical Center and Fifth Medical Center of the PLA General Hospital and Nanfang Hospital. To identify the optimal diagnostic model, we compared several machine learning approaches, including K‐nearest neighbor, support vector machine with radial basis function (SVM‐RBF), SVM with Gaussian kernel, random forest, and extreme gradient boosting (XGB), using all electrophysiological features. These were also compared with nerve conduction studies of the peroneal and sural nerves. To construct a rapid diagnostic model, different feature combinations were assessed across the machine learning methods. Model performance was evaluated using cross‐validated areas under the receiver operating characteristic curve (AUCs). Results: Of 14,768 admissions screened, 134 patients with CIP and 135 matched controls were included. In total, 41 electrophysiological features were analyzed. Feature ranking revealed that the distal compound muscle action potential (CMAP) of the peroneal nerve contributed most to diagnostic accuracy. After comparison, the SVM‐RBF method was selected to establish the final diagnostic model. The optimal model, based on seven electrophysiological features of the peroneal and ulnar nerves (proximal latency, distal latency, and CMAP amplitude), achieved an AUC of 0.93, which was comparable to the all‐feature XGB model (AUC = 0.95). In the independent validation set, the rapid diagnostic model maintained strong performance (AUC = 0.88), similar to the all‐feature model (AUC = 0.90). Conclusion: We developed a rapid diagnostic model for CIP using the SVM‐RBF method and seven electrophysiological features from the peroneal and ulnar nerves. This model enables efficient, minimally invasive, and timely diagnosis of CIP in critically ill patients. The source code and related scripts are available at [ https://github.com/PLAGH‐Neuro‐Yang/RDM‐CIP ].

9 June 2026 in Frontiers in Cardiovascular Medicine
Verified
Journal Article
Open access
PDF file
HHHaoyang Hu
Haoyang Hu
ZYZekai Yu
Zekai Yu
FQFeiwei Qin
Feiwei Qin
FYFei Yang
Fei Yang
Background: Cardiovascular diseases (CVDs) remain the leading cause of global mortality. While the 12-lead electrocardiogram (ECG) is the clinical gold standard, wearable devices typically provide only single-lead monitoring (Lead I), which lacks the spatial resolution required to detect complex pathologies such as myocardial infarction or ST-segment changes. Objective: This study proposes a deep learning framework capable of reconstructing a full 12-lead ECG from a single-lead input while preserving diagnostic integrity. Methods: We developed TONet (Transformer-based Optimization Network), a multi-task framework evaluated on the PTB-XL dataset ( n = 21,837). The architecture integrates a Convolutional Neural Network (CNN) encoder to extract local morphological features and a Transformer encoder to capture global temporal dependencies across the cardiac cycle. To ensure clinical validity, the model was trained using a dual-objective loss function, simultaneously optimizing for signal reconstruction (MSE) and multi-label diagnostic classification (BCE) across five diagnostic super-classes. Results: TONet achieved an overall Pearson Correlation Coefficient (PCC) reaching 0.673 on the independent test set. Clinical agreement assessment via Bland-Altman analysis revealed a negligible per-lead bias (|Bias| ≤ 0.0024 mV across all 11 reconstructed leads), indicating no systematic voltage drift. In terms of diagnostic utility, the reconstructed 12-lead signals achieved a macro-averaged AUROC of 0.821 (95% CI: 0.810–0.831), with per-class AUROCs of 0.874 (NORM), 0.770 (MI), 0.862 (STTC), 0.782 (CD), and 0.815 (HYP). To assess generalisability beyond a single cohort, we additionally performed an independent in-distribution validation on the Chapman-Shaoxing 12-lead ECG database ( n = 43,559 quality-controlled recordings), where TONet, retrained from scratch using the identical protocol, achieved an overall PCC of 0.612 (95% CI: 0.552–0.666) and substantially outperformed the linear-regression baseline (PCC = 0.498), confirming that the reconstruction benefit transfers to a second, hardware- and population-distinct cohort. Conclusion: TONet demonstrates that a hybrid CNN–Transformer network with a multi-task reconstruction-plus-classification objective can recover morphologically faithful 12-lead surrogates from a single Lead I drawn from clinical PTB-XL recordings. We position this work as a regularized deep-learning framework for multi-lead reconstruction rather than a deployable clinical device: the reconstructed precordial leads are insufficient for fine-grained ST-elevation myocardial infarction (STEMI) screening, and external validation on true wearable recordings remains essential.

14 October 2026 in Frontiers in Public Health
Verified
Journal Article
Open access
PDF file
HHHaoyang Hu
Haoyang Hu
ZYZekai Yu
Zekai Yu
FYFei Yang
Fei Yang
Background: Sexual and gender minority (SGM) populations increasingly rely on digital platforms for health information, yet the quality, reliability, and stigmatizing content of SGM health videos remain unclear. This study evaluated SGM health-related short-form videos on the Chinese platforms Douyin and Bilibili. Methods: We conducted a cross-sectional content analysis of prominently ranked videos retrieved using 15 research-team-developed search terms during the prespecified data-collection period (January 1–May 1, 2026). For each search term, the first 10 results returned under each platform's default ranking were recorded, yielding 150 candidate records per platform before deduplication. Eligible videos were assessed for uploader type, content theme, engagement metrics, Global Quality Score (GQS), modified DISCERN reliability, medical accuracy, misinformation, and an exploratory stigma-to-affirmation communication score. Non-parametric tests, Cramér's V with 95% confidence intervals, and unadjusted Spearman correlations were used. Results: Of 300 candidate records, 194 videos were included (118 Douyin, 76 Bilibili). Individual users and healthcare professionals were the most common uploaders. HIV/STI prevention and mental health were the leading themes, whereas healthcare access, privacy protection, HPV vaccination, and cancer screening were underrepresented. Overall quality was moderate [median GQS = 3.0; interquartile range (IQR), 2.0–4.0], but reliability was limited (median modified DISCERN = 2.0; IQR, 1.0–3.0). High-quality and highly reliable videos accounted for 38.1% and 19.1%, respectively. Misinformation appeared in 27.8% of videos, and 17.0% contained stigmatizing, pathologizing, fear-based, or discriminatory elements. Raw cumulative engagement metrics were strongly intercorrelated but showed weak unadjusted associations with quality, reliability, accuracy, or the exploratory communication score. Conclusion: Among prominently ranked SGM health videos identified using the predefined search strategy, overall quality was moderate, reliability was limited, thematic coverage was uneven, and misinformation and stigmatizing communication remained present. Raw engagement metrics showed weak associations with quality-related scores. These findings describe a time-bounded algorithmic snapshot and should not be generalized to all content on either platform.

30 October 2026 in Frontiers in Aging Neuroscience
Verified
Journal Article
Open access
PDF file
YWYan Wang
Yan Wang
HHHaoyang Hu
Haoyang Hu
JPJiaxuan Peng
Jiaxuan Peng
ZYZekai Yu
Zekai Yu
FYFei Yang
Fei Yang
Background: Naturalistic keyboard use enables repeated, low-burden assessment of fine-motor behavior. Whether day-to-day variability in key-hold duration provides information beyond average key-hold duration in Parkinson’s disease (PD) remains unclear. We tested whether within-person variability in daily key-hold duration was associated with a one-time self-reported PD Impact category. Methods: We reconstructed the public Tappy dataset from raw keystroke files. The primary phenotype was the within-person coefficient of variation (CV) of daily mean key-hold duration. Ordinal-trend and categorical Impact parameterizations were treated as parallel primary analyses. Secondary and exploratory analyses evaluated alternative variability measures, timing-filter thresholds, observation density, robust daily summaries, approximate disease duration, and unequal precision of daily means. Participant-day location-scale models used both a conventional finite-keystroke sampling-variance approximation and a session/block-aware alternative; participant-cluster bootstrap inference used 2,000 replicates. Results: Quality control retained 8,999,669 keystrokes and 5,490 valid participant-days. The complete-case analysis included 73 participants with PD (33 Mild, 30 Medium, and 10 Severe). Each one-category increase in self-reported PD Impact was associated with a 32.4% higher day-to-day CV (ratio, 1.32; 95% CI, 1.13–1.55; P < 0.001). In the categorical parameterization, ratios were 1.38 for Medium versus Mild and 1.70 for Severe versus Mild (overall robust Wald P = 0.001). Estimates were essentially unchanged with 1,000- or 1,500-ms timing limits and a hold-duration-only filter. Daily-median CV remained associated with Impact (ratio, 1.30; 95% CI, 1.08–1.56; P = 0.005). Among the same 52 participants eligible at 500 keystrokes/day, the ratio was 1.25 using the 100-keystroke definition and 1.20 using the 500-keystroke definition. In 72 participants contributing 3,193 participant-days, the session/block-aware location-scale model estimated a 1.62-fold higher latent residual SD per Impact category (2,000-replicate participant-cluster bootstrap 95% CI, 1.20–1.98). Conclusion: Greater self-reported PD Impact was associated with higher within-person variability in naturalistic key-hold duration. This cross-sectional association identifies day-to-day variability as a candidate feature for passive fine-motor assessment. Prospective validation against repeated clinical outcomes and independent naturalistic cohorts is needed.

13 August 2026 in Frontiers in Cellular and Infection Microbiology
Verified
Journal Article
Open access
PDF file
HHHaoyang Hu
Haoyang Hu
LHLei Hou
Lei Hou
YWYan Wang
Yan Wang
ZYZekai Yu
Zekai Yu
JZJiatang Zhang
Jiatang Zhang
FYFei Yang
Fei Yang
Background: Tuberculous meningitis (TBM) is a severe central nervous system infection in which dysregulated host inflammation contributes to neurological injury. The cellular organization of the cerebrospinal fluid (CSF) immune microenvironment in TBM, particularly in comparison with other inflammatory meningitis conditions, remains incompletely defined. This study characterized CSF immune-cell heterogeneity in TBM and purulent meningitis (PM) using single-cell transcriptomic profiling. Methods: Pre-treatment CSF immune cells from five patients with TBM and five patients with PM were profiled using 10x Genomics single-cell RNA sequencing. After quality control, ambient RNA assessment, and doublet filtering, 96,594 cells were retained. Major immune-cell populations and myeloid and granulocyte subclusters were annotated using canonical lineage markers and cluster-specific marker genes, with additional evaluation of doublet-related quality-control metrics and ambient RNA estimates. Immune-cell composition, CD4 + T-cell transcriptional states, macrophage migration inhibitory factor (MIF)/CD74 expression, CellChat-inferred communication, and in silico perturbation profiles were evaluated using donor-level comparisons where appropriate. Results: Seven major CSF immune-cell populations were identified, including CD4 + T cells, CD8 + T cells, myeloid cells, granulocytes, B cells, plasma cells, and natural killer T (NKT)-like cells. Immune-cell composition varied markedly across donors. Several PM donors exhibited myeloid- or granulocyte-rich profiles, whereas several TBM donors showed greater CD4 + T-cell representation; group-level differences were exploratory and did not consistently reach statistical significance. Reannotation of the innate immune-cell compartment identified broad macrophage-like, monocyte-like, neutrophil-like, and CD1C + dendritic cell–like populations. Two originally defined myeloid subclusters displayed coordinated T-cell-associated transcriptional programs and were interpreted as potentially misclassified or technically mixed cells rather than bona fide myeloid states. CD4 + T cells occupied a continuous transcriptional landscape comprising central memory (CM)-like, cytotoxicity/senescence-associated, and activated effector memory (EM)-like states with overlapping transcriptional programs. MIF was broadly detectable across immune-cell compartments, whereas CD74 was most prominent in antigen-presentation-associated populations. CellChat identified MIF-related communication in both disease groups, supporting a shared multicompartment inflammatory communication program rather than a TBM-specific pathway. Exploratory in silico perturbation prioritized MIF- and CD74-associated transcriptional programs for future validation. Conclusions: This study provides a donor-aware single-cell transcriptomic atlas of the inflammatory CSF immune microenvironment in TBM and PM. The findings highlight substantial inter-individual heterogeneity, broad innate immune-cell patterns, overlapping CD4 + T-cell transcriptional states, and multicompartment MIF/CD74-related communication. These transcriptome-inferred findings are hypothesis-generating and require validation in larger cohorts using protein-level and functional approaches.