Zhihui Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhihui Xu has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
- Cong Pian
School of Basic Medicine and Clinical Pharmacy, China Pharmaceutical University, Nanjing 211198, Jiangsu, China
15 shared publications - Jie Zhao
National Engineering Laboratory for Internet Medical Systems and Applications The First Affiliated Hospital of Zhengzhou University Zhengzhou Henan China
14 shared publications - Jinwen Ji
The State Key Laboratory of Neurology and Oncology Drug Development, Jiangsu Simcere Diagnostics Co., Ltd., Nanjing, China
14 shared publications - Lifeng Li
National Engineering Laboratory for Internet Medical Systems and Applications, The First Affiliated Hospital of Zhengzhou University, Zhengzhou University, Zhengzhou, China
14 shared publications - Meijia Yang
National Engineering Laboratory for Internet Medical Systems and Applications The First Affiliated Hospital of Zhengzhou University Zhengzhou Henan China
14 shared publications - Rui Yan
Institute of Clinical Epidemiology, Key Laboratory of Public Health Safety, Ministry of Education, Key Lab of Health Technology Assessment of Ministry of Health, School of Public Health, Fudan University, Shanghai, China
14 shared publications
Latest Publications
- Open Access
- 1 September 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (5)
ToxMamba: a novel method for toxic peptide prediction based on the fusion of pretrained embeddings and multi-scale state space models
Peptide therapeutics offer great promise in drug discovery, yet the accurate identification of peptide toxicity remains essential for both safety assessment and biodefense. We developed ToxMamba, a toxic peptide prediction framework that integrates Implicit Structure Model (ISM)-derived pretrained embeddings, state space models, multi-scale feature learning, and ensemble inference. Built upon the Mamba architecture, ToxMamba enables efficient sequence modeling and captures contextual patterns across multiple receptive fields from parallel branches. On the independent test set, ToxMamba consistently surpasses existing methods, improving the F1 score and Matthews Correlation Coefficient by 3.9–18.3 and 7.5–46.9 percentage points, respectively. Ablation analyses further showed that Mamba-based models generally achieved stronger overall performance than Transformer-, Convolutional Neural Network-Long Short-Term Memory-Attention-, and Kolmogorov–Arnold Network-based variants. Feature visualization showed that ISM embeddings yielded clearer separation between toxic and non-toxic peptides in the latent space. Intermediate-layer feature analysis highlighted residue-level regions associated with toxicity-related sequence motifs, supporting the biological interpretability of ToxMamba. Overall, this study provides an efficient and interpretable computational framework for high-throughput toxic peptide screening.
- 1 September 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (5)
Open AccessToxMamba: a novel method for toxic peptide prediction based on the fusion of pretrained embeddings and multi-scale state space models
Peptide therapeutics offer great promise in drug discovery, yet the accurate identification of peptide toxicity remains essential for both safety assessment and biodefense. We developed ToxMamba, a toxic peptide prediction framework that integrates Implicit Structure Model (ISM)-derived pretrained embeddings, state space models, multi-scale feature learning, and ensemble inference. Built upon the Mamba architecture, ToxMamba enables efficient sequence modeling and captures contextual patterns across multiple receptive fields from parallel branches. On the independent test set, ToxMamba consistently surpasses existing methods, improving the F1 score and Matthews Correlation Coefficient by 3.9–18.3 and 7.5–46.9 percentage points, respectively. Ablation analyses further showed that Mamba-based models generally achieved stronger overall performance than Transformer-, Convolutional Neural Network-Long Short-Term Memory-Attention-, and Kolmogorov–Arnold Network-based variants. Feature visualization showed that ISM embeddings yielded clearer separation between toxic and non-toxic peptides in the latent space. Intermediate-layer feature analysis highlighted residue-level regions associated with toxicity-related sequence motifs, supporting the biological interpretability of ToxMamba. Overall, this study provides an efficient and interpretable computational framework for high-throughput toxic peptide screening.
- 16 June 2026
- Published by Cold Spring Harbor Laboratory in Genome Research
MAPLE enables ultrasensitive detection of low-frequency cfDNA methylation haplotypes using short capture probes with cost-efficient performance
One major challenge in disease detection using methylation biomarkers is accurately identifying low-frequency, disease-specific methylation haplotypes. Conventional methods detect these haplotypes using unbiased long capture probes, requiring high sequencing depth, with most reads being uninformative. Here, we present Methylation Anchor Probe for Low-signal Enrichment (MAPLE), a novel technology that uses rationally designed, ultrasensitive short capture probes to selectively enrich low-frequency methylation haplotypes, significantly reducing sequencing costs while enhancing detection sensitivity. We experimentally validate MAPLE on reference samples and develop the Enrichment Factor Model (EFM) to quantify the enrichment efficiency of individual probes. To our knowledge, this is the first model to enable correction of observed disease-specific haplotype fractions to their original levels, essential for estimating the limit of detection (LOD) in assay performance evaluation. Finally, two panels are designed using MAPLE and the conventional method respectively to target 13,264 colorectal cancer (CRC)-related methylation haplotypes and are validated through a blood plasma cohort of 162 samples including CRC patients and controls. In 24 stage I and II cancer samples, MAPLE maps 22.7% of unique reads to 5045 targeted haplotypes, compared to 0.8% of reads mapping to 190 haplotypes with the conventional method. MAPLE also enriches tumor-specific fragmentomic characteristics, including fragment size and DNA cleavage profiles. Aided by a LightGBM classifier, the MAPLE panel accurately identifies patients with 95% specificity and 80% sensitivity. Even with just 0.05% of total sequencing reads, MAPLE exhibits only a modest performance decline, in contrast to the conventional method’s significant decline, highlighting MAPLE’s superior efficiency and robustness.
- 16 June 2026
- Published by Cold Spring Harbor Laboratory in Genome Research
MAPLE enables ultrasensitive detection of low-frequency cfDNA methylation haplotypes using short capture probes with cost-efficient performance
One major challenge in disease detection using methylation biomarkers is accurately identifying low-frequency, disease-specific methylation haplotypes. Conventional methods detect these haplotypes using unbiased long capture probes, requiring high sequencing depth, with most reads being uninformative. Here, we present Methylation Anchor Probe for Low-signal Enrichment (MAPLE), a novel technology that uses rationally designed, ultrasensitive short capture probes to selectively enrich low-frequency methylation haplotypes, significantly reducing sequencing costs while enhancing detection sensitivity. We experimentally validate MAPLE on reference samples and develop the Enrichment Factor Model (EFM) to quantify the enrichment efficiency of individual probes. To our knowledge, this is the first model to enable correction of observed disease-specific haplotype fractions to their original levels, essential for estimating the limit of detection (LOD) in assay performance evaluation. Finally, two panels are designed using MAPLE and the conventional method respectively to target 13,264 colorectal cancer (CRC)-related methylation haplotypes and are validated through a blood plasma cohort of 162 samples including CRC patients and controls. In 24 stage I and II cancer samples, MAPLE maps 22.7% of unique reads to 5045 targeted haplotypes, compared to 0.8% of reads mapping to 190 haplotypes with the conventional method. MAPLE also enriches tumor-specific fragmentomic characteristics, including fragment size and DNA cleavage profiles. Aided by a LightGBM classifier, the MAPLE panel accurately identifies patients with 95% specificity and 80% sensitivity. Even with just 0.05% of total sequencing reads, MAPLE exhibits only a modest performance decline, in contrast to the conventional method’s significant decline, highlighting MAPLE’s superior efficiency and robustness.
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Data from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Background:Lung cancer is the most lethal malignancy worldwide and urgently requires effective early detection strategies. Current noninvasive approaches based on plasma cell-free DNA (cfDNA) fragmentomics often suffer from limited sensitivity in early-stage patients because of low tumor DNA fractions.Methods:We developed a novel computational feature, first-order transition probability (FOTP), to capture nucleotide sequential dependencies within cfDNA fragments. Using low-pass whole-genome sequencing data from 1,036 participants, we systematically analyzed cfDNA fragment ends to identify discriminative regions for cancer detection and trained a support vector machine (SVM) model leveraging the FOTP features.Results:Analysis revealed that the first 10 base pairs (bp) at the 5′-end of cfDNA fragments contained the most discriminative information. The SVM model achieved an area under the ROC curve of 0.942, with 73.9% sensitivity for stage I and 81.8% for stage II lung cancer at 95% specificity, significantly outperforming existing fragmentomic features. Nucleotide frequency stability and entropy patterns beyond the initial 10 bp supported the biological basis of the approach, reflecting nuclease cleavage biases and chromatin features. The method generalized robustly across independent cohorts and multicancer validation sets, showing potential for tissue-of-origin prediction.Conclusions:FOTP is a biologically interpretable and highly efficient feature for early cancer detection. It captures key nucleotide dependencies at cfDNA fragment ends, enhancing sensitivity for early-stage lung cancer and other cancers.Impact:This approach offers a scalable and generalizable strategy for early cancer screening.
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessData from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Background:Lung cancer is the most lethal malignancy worldwide and urgently requires effective early detection strategies. Current noninvasive approaches based on plasma cell-free DNA (cfDNA) fragmentomics often suffer from limited sensitivity in early-stage patients because of low tumor DNA fractions.Methods:We developed a novel computational feature, first-order transition probability (FOTP), to capture nucleotide sequential dependencies within cfDNA fragments. Using low-pass whole-genome sequencing data from 1,036 participants, we systematically analyzed cfDNA fragment ends to identify discriminative regions for cancer detection and trained a support vector machine (SVM) model leveraging the FOTP features.Results:Analysis revealed that the first 10 base pairs (bp) at the 5′-end of cfDNA fragments contained the most discriminative information. The SVM model achieved an area under the ROC curve of 0.942, with 73.9% sensitivity for stage I and 81.8% for stage II lung cancer at 95% specificity, significantly outperforming existing fragmentomic features. Nucleotide frequency stability and entropy patterns beyond the initial 10 bp supported the biological basis of the approach, reflecting nuclease cleavage biases and chromatin features. The method generalized robustly across independent cohorts and multicancer validation sets, showing potential for tissue-of-origin prediction.Conclusions:FOTP is a biologically interpretable and highly efficient feature for early cancer detection. It captures key nucleotide dependencies at cfDNA fragment ends, enhancing sensitivity for early-stage lung cancer and other cancers.Impact:This approach offers a scalable and generalizable strategy for early cancer screening.
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Figure S6 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Sensitivity of FOTP-based model across different clinical subgroups in the independent cohort at 95% specificity
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessFigure S6 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Sensitivity of FOTP-based model across different clinical subgroups in the independent cohort at 95% specificity
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Table S4 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Number of Principal components explaining 99% of total variance
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessTable S4 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Number of Principal components explaining 99% of total variance
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Table S3 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Model sensitivity based on FOTP features derived from sequences of varying lengths at 95% specificity
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessTable S3 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Model sensitivity based on FOTP features derived from sequences of varying lengths at 95% specificity
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Figure S8 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Tissue-of-origin results from the external multi-cancer dataset
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessFigure S8 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Tissue-of-origin results from the external multi-cancer dataset
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Figure S5 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Classification score distributions across non-cancer individuals and cancer patients in the independent external test cohort
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessFigure S5 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Classification score distributions across non-cancer individuals and cancer patients in the independent external test cohort
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Figure S7 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Nucleotide frequencies within the first 20 bp from fragment ends for healthy individuals and HCC patients
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessFigure S7 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Nucleotide frequencies within the first 20 bp from fragment ends for healthy individuals and HCC patients
- Open Access
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Figure S2 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Nucleotide frequencies within the first 20 bp from fragment ends for a healthy individual and an LC patient
- 2 June 2026
- Published by American Association for Cancer Research (AACR)
Open AccessFigure S2 from Early Lung Cancer Detection Using Nucleotide Transition Probabilities in Plasma Cell-Free DNA
Nucleotide frequencies within the first 20 bp from fragment ends for a healthy individual and an LC patient