About

Mingyang Lu is an Associate Professor in the Department of Bioengineering at Northeastern University. He received his BS degree in Physics from Fudan University in 2003 and earned his PhD in Biochemistry and Molecular Biology with a Biophysics Track from Baylor College of Medicine in 2010. Prior to joining Northeastern, he established his independent research lab as an Assistant Professor at The Jackson Laboratory in 2016. His research interests focus on methodology development in computational systems biology, integrating mathematical modeling and bioinformatics to study gene regulatory networks, single cell genomics, epithelial-mesenchymal transition, coarse-graining, reverse engineering, machine learning, stochasticity, and heterogeneity in gene expression. Dr. Lu's work aims to model cellular state transitions by analyzing genomics data through computational approaches. He has received recognition for his research, including a five-year Maximizing Investigators' Research Award (MIRA) from the National Institute of General Medical Sciences.

Research topics

• Computer Science
• Computational biology
• Biology
• Genetics
• Cell biology
• Algorithm
• Bioinformatics

Selected publications

• NGRPO: Negative-enhanced Group Relative Policy Optimization

  ArXiv.org · 2025-09-23

  preprintOpen access

  RLVR has enhanced the reasoning capabilities of Large Language Models (LLMs) across various tasks. However, GRPO, a representative RLVR algorithm, suffers from a critical limitation: when all responses within a group are either entirely correct or entirely incorrect, the model fails to learn from these homogeneous responses. This is particularly problematic for homogeneously incorrect groups, where GRPO's advantage function yields a value of zero, leading to null gradients and the loss of valuable learning signals. To overcome this issue, we propose NGRPO (Negative-enhanced Group Relative Policy Optimization), an algorithm designed to convert homogeneous errors into robust learning signals. First, NGRPO introduces Advantage Calibration. This mechanism hypothesizes the existence of a virtual maximum-reward sample during advantage calculation, thereby altering the mean and variance of rewards within a group and ensuring that the advantages for homogeneously incorrect samples are no longer zero. Second, NGRPO employs Asymmetric Clipping, which relaxes the update magnitude for positive samples while imposing stricter constraints on that of negative samples. This serves to stabilize the exploration pressure introduced by the advantage calibration. Our experiments on Qwen2.5-Math-7B demonstrate that NGRPO significantly outperforms baselines such as PPO, GRPO, DAPO, and PSR-NSR on mathematical benchmarks including MATH500, AMC23, and AIME2025. These results validate NGRPO's ability to learn from homogeneous errors, leading to stable and substantial improvements in mathematical reasoning. Our code is available at https://github.com/nangongrui-ngr/NGRPO.

  PublisherOA PDFDOI

• Neglected vertical linkage: A study on the form of the canal network in the Huainan Salt Area during the Ming and Qing dynasties using space syntax measurements

  Frontiers of Architectural Research · 2025-01-07 · 3 citations

  articleOpen access1st authorCorresponding

  The Huainan Salt Area was the most essential salt production and trade base in ancient China. The government had strict control over the essential tax source region and left behind numerous official documents. However, when comparing with the modern surveying maps, we found that the boundaries between salt districts were unclear, which was obviously contradictory to the principle of independence between salt districts in official documents. Therefore, this study used the 1920s’ historical map of the Huainan Area to extract vector data such as the canal and road networks and the location of salt settlements, and then used space syntax to describe the transportation network morphology and settlement distribution structure to respond to the contradictions with higher precision. It was found that at a 2000 m × 2000 m raster resolution, the measurements of the canal network were well correlated ( p = 0.645, r = 30,000) with the salt settlement distribution. Moreover, the canal network was closer to a network-like than a tree-like structure, and vertical linkage emerged. The “government-supervised and merchant-managed” policy implemented in the Qing dynasty played an essential role, and it caused a big difference in the understanding of the salt industry system between folks and officials.

  PublisherDOI

• Comparative Assessment of Two Numerical Methods for Eddy Current Nondestructive Evaluation: Insights from Benchmark Studies

  Progress In Electromagnetics Research M · 2025-01-01

  articleOpen access
  PublisherDOI

• Buprofezin causes early developmental toxicity of zebrafish (Danio rerio) embryos: morphological, physiological and biochemical responses

  Aquatic Toxicology · 2025-04-20 · 1 citations

  articleOpen access

  Buprofezin (BPFN), a pesticide used to control crop pests and diseases, causes potential harm to aquatic animals and the environment by leaching into aquatic ecosystems. However, there are limited studies on the toxicity of BPFN to aquatic organisms. Using zebrafish embryos, we integrated flow cytometry, qRT-PCR, RNA-seq and other techniques to assess BPFN's developmental toxicity. Additionally, IBRv2 index and Mantel test correlation were applied to comprehensively evaluate the developmental toxicity of BPFN. The results showed that BPFN induced cytotoxicity, including increased reactive oxygen species levels, mitochondrial membrane potential depolarization, and apoptosis, which further resulted in developmental toxicity of zebrafish embryos such as delayed hatching, reduced survival rate, and severe morphological deformities. BPFN also affected the number and distribution of immune cells, resulting in immunotoxicity, and disrupted the endogenous antioxidant system by altering the activities of catalase, superoxide dismutase, and glutathione S-transferase and contents of malondialdehyde and glutathione. Gene expression analysis revealed that BPFN induced changes in the expression of genes associated with oxidative stress, apoptosis, inflammation, swim bladder development, and eye development. In the comprehensive evaluation, BPFN showed the strongest developmental toxic effect in the 20 μM BPFN-treated group at 48 hpf, and there was the significant correlation between embryonic development, oxidative stress, apoptosis, and inflammatory response. The rescue experiment confirmed that astaxanthin can alleviate the embryonic developmental toxicity caused by BPFN to a certain extent. In summary, BPFN induced early developmental toxicity in zebrafish embryos, which might be associated with mitochondria-mediated apoptosis pathway induced by oxidative stress.

  PublisherDOI

• Building dynamical models of multi-step state transitions from single cell gene expression trajectories

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-12-11

  preprintOpen accessSenior authorCorresponding

  Multi-step cell state transitions often occur in biological processes, such as cell differentiation and disease progression, yet the regulatory mechanisms governing these transitions remain unclear. Here, we introduce NetDes, a computational method that integrates top-down and bottom-up systems biology to infer core transcription factor regulatory networks and build ODE-based dynamical models from single-cell gene expression trajectories. We demonstrate that NetDes predicts regulatory interactions and reproduces gene expression dynamics through benchmarking using in-silico time trajectories with decoys, tests on gene circuit simulations of embryonic phenotypic switching, and application to time-series scRNA-seq data from human iPSC differentiation. Compared to existing approaches, NetDes has the advantage of capturing sequential state transitions within a single dynamical model. Network simulations and coarse-graining further elucidate the regulatory roles of genes and their combinations in driving these transitions. Our approach provides a generalizable framework for mechanistic modeling of gene regulation in complex cell state transitions.

  PublisherDOI

• Functional characterization of EuSIP1 reveals its central role in trans-polyisoprene biosynthesis in Eucommia ulmoides

  Industrial Crops and Products · 2025-11-24

  articleOpen access

  Eucommia ulmoides is a unique plant capable of synthesizing trans-polyisoprene (TPI), a form of natural rubber with outstanding mechanical and chemical properties. In addition to its industrial importance, the bark of E. ulmoides is a high-value traditional Chinese medicinal material used for promoting bone health, lowering blood pressure and supporting kidney function. Despite its dual utility, the molecular regulation of TPI biosynthesis in this species remains poorly understood. In this study, we performed a genome-wide analysis and identified five soluble inorganic pyrophosphatase (SIP) genes in E. ulmoides ( EuSIP1–EuSIP5 ). Among them, EuSIP1 showed strong sequence homology with SIPs previously associated with rubber production in Hevea brasiliensis . Functional characterization was conducted using Agrobacterium-mediated transformation to generate overexpression (OE) and RNA interference (RNAi) lines. Overexpression of EuSIP1 significantly reduced intracellular pyrophosphate (PPi) levels (by 40.3 %–73.1 %) and upregulated key rubber biosynthesis genes ( EuFPS1, EuSRPP1, EuIPI, EuGGPPS ), resulting in higher TPI accumulation (0.30 %–0.81 %) and increased molecular weight. In contrast, RNAi lines accumulated more PPi (1.65–2.25 × WT), exhibited suppressed biosynthetic gene expression, and showed a marked reduction in TPI content (0.05 %–0.18 %). Protein–protein docking and yeast two-hybrid assays revealed weak but reproducible interactions between EuSIP1 and rubber biosynthesis-related proteins EuFPS and EuSRPP. Subcellular localization confirmed EuSIP1 presence in both the cytoplasm and nucleus, supporting its role in intracellular PPi homeostasis. Together, these results suggest that EuSIP1 plays a dual role in regulating TPI biosynthesis, promoting TPI synthesis and elongation through the interaction of PPi inhibition by hydrolysis and rubber biosynthesis related proteins. This study advances our understanding of the molecular regulation of rubber production in E. ulmoides , a species of both economic and medicinal importance, and identifies EuSIP1 as a promising target for metabolic engineering to improve rubber yield and quality. • Genome-wide analysis identified five EuSIP genes in Eucommia ulmoides , EuSIP1 is homologous to rubber-associated SIPs. • EuSIP1 boosts rubber yield and MW by lowering PPi and interacting with EuFPS/EuSRPP, revealing SIP's role in TPI biosynthesis. • EuSIP1 interacts with core TPI enzymes (EuFPS/EuSRPP), suggesting dual roles in PPi scavenging & polymer elongation facilitation. • First evidence of SIP-mediated regulation in TPI biosynthesis.

  PublisherDOI

• Cell‐free transcriptomic profiles and mechanism insights in female androgenetic alopecia

  Clinical and Translational Medicine · 2025-11-01

  articleOpen access

  Dear Editor, Our study presents a novel predictive machine learning model that demonstrates the potential of plasma cell-free RNA (cfRNA) for diagnosing and prognosing female androgenetic alopecia (FAGA). We identified cell-free DNAJB9 as significantly associated with FAGA through bioinformatic analysis and machine learning followed by RT-qPCR validation (Figure 1A). FAGA manifests heterogeneously,1 often as diffuse thinning of the crown and frontal scalp.2 It's pathogenesis critically involves androgen-hair follicle interactions and WNT and JAK-STAT signalling.3 The cfRNA in bodily fluids have shown diagnostic/prognostic potential for various diseases.4 Machine learning is increasingly used to analyse complex cfRNA data.5 However, the potential association between cfRNA and FAGA remains unclear. For subsequent analyses comparing disease severity, we focused on the ‘upper’ group as patients in the top 25% of the FAGA-Index (scores > 5.53) and the ‘lower’ group as those in the bottom 25% (scores < 1.92). Blood test results (Figure S1 and Table S2) showed no significant differences in various haematological and biochemical indicators between the FAGA and control groups. However, testosterone exhibited a significantly lower level in the ‘upper’ group (Figure S5), supporting that the FAGA-Index effectively enhances the stratification of patients by severity and may facilitate identification of other potential biomarkers in FAGA progression. Greater variation in principal component analysis (PCA) of cfRNA expression profiles between upper and lower FAGA subgroups, compared to that between FAGA and control groups, also suggested increased heterogeneity or molecular diversity within FAGA subtypes (Figure 1C and D). The RNA biotypes were categorised based on Ensembl classifications with minor adjustments (Figure 1E). Analysis of differentially expressed genes (DEGs) showed that CYTB, RNY1, and TMSB4X were notably upregulated in FAGA patients, whilst EEF1A1 was significantly downregulated (Figure 2A; Table S5). Furthermore, genes including ND2, ATP6, ND6, and PARLP1 exhibited significant expression changes across varying disease severities (Table S7), suggesting their potential association with FAGA progression (Figure 2B). Functional enrichment analysis of these DEGs implicated pathways related to sensory perception, nuclear division, chromosome segregation, and mitosis in FAGA (Figure 2C and D; Tables S6 and S8). Pathway activity analysis reinforced the potential importance of JAK-STAT and WNTs pathways in FAGA (Figure 3A–C; Tables S17–S19). A comparison of transcription factor (TF) activities between FAGA patients and controls (Figure 3D; Table S9) revealed significantly increased activity of NCOA3 and MAX. However, no significant differences in TF activities were observed between upper- and lower-FAGA subgroups (Table S10). Crucially, we observed a significant negative correlation between cell-free DNAJB9 expression and the FAGA-Index, suggesting that low DNAJB9 expression may be associated with increased FAGA severity (Table S11). Protein–protein interaction network (PPI) analysis further revealed that in FAGA versus controls, upregulated genes drive endocrine compensation and mitochondrial stress responses, while downregulated genes suppress growth signalling (MET/mTORC1) and RNA metabolism (Figure S2). Comparing upper- versus lower-FAGA, there is increased mitochondrial/endocrine activity alongside impaired translation and calcium homeostasis (Figure S3). These results indicate worsening pathway dysregulation with FAGA progression (Tables S12–S15). Subsequently, we developed a predictive model for FAGA using machine learning, employing the GeneLLM downstream classification framework,6 a state-of-the-art method for cfRNA classification. After initial feature extraction, deep feature mining was conducted to uncover latent patterns within gene expression profiles indicative of FAGA. The cfRNA RPKM matrix was partitioned into training (40%), validation (40%), and testing (40%) sets (Figure 4A). The large, completely held-out test set offers a rigorous internal validation of the model's performance on unseen data from the same population. To ensure robustness, hyperparameter optimisation was conducted using 10-fold cross-validation. The model achieved an Area Under the Curve (AUC) of .707 for distinguishing FAGA patients from controls (Figure 4B) and .714 for separating high versus low FAGA-Index scores (Figure 4C). It identified several genes associated with FAGA, including VGLL3, CYP1A1, antisense to PDE7B, and notably DNAJB9 (Figure 4D). Features correlated with FAGA severity, such as the long non-coding RNA ARL14EP-DT and pseudogene TJAP1P1, were also highlighted (Figure 4E). We performed correlation analysis between the expression levels of the six candidate biomarkers and various blood parameters (Figure S6; Table S20). Nevertheless, the correlation coefficient indicated only weak to moderate associations, implying that more validation is needed between biochemical and molecular diagnosis. Further RT-qPCR validation in both internal and external cohorts confirmed that cell-free VGLL3, antisense to PDE7B, and DNAJB9 were significantly downregulated in the FAGA patients, while lncRNA ARL14EP-DT and pseudogene TJAP1P1 were significantly downregulated in the upper FAGA subgroup (Figure S4). Notably, DNAJB9 is a DNAJ/HSP40 heat shock protein essential for cellular stress responses.7 HSP40 family proteins are implicated in regulating androgen receptor (AR) activity, often maintaining AR in an inactive state.8, 9 Reduced DNAJB9 expression may potentially disrupt AR signalling in hair follicles, especially under stress. In summary, as the first investigation integrating cfRNA bioinformatic analysis with machine learning, this study establishes a crucial proof-of-concept for the utility of cfRNA in FAGA diagnosis and prognosis, addressing a critical gap in the field and providing a solid foundation for future work. Our exploratory model, whilst moderate in its predictive power, proved effective in highlighting cell-free DNAJB9 as a FAGA biomarker and a candidate for therapeutic intervention, meriting further investigation. Project supervision was overseen by H.J. Y.L. S.D., Y.J., H.J., and Y.L. were responsible for conceptualisation. L.J., M.L., and Y.L. managed clinical recruitment design. Y.J. designed and conducted the wet-lab experiments. Data analysis and interpretation were performed by S.D. and Z.A. The primary manuscript writing and revisions were undertaken by L.J., M.L., S.D., and Y.J. Coordination of human samples and data collection was managed by Y.Z., L.J., H.J., M.L., Y.L., Z.L., C.Z., and R.L. All authors contributed to discussions on the results and provided feedback on the manuscript. This research was funded by Ministry of Education of People's Republic of China (Grant 231103242232720 to Yufei Li), the East Hospital Affiliated to Tongji University Introduced Talent Research Startup Fund (Grant DFRC2019008 to Hua Jiang), the Clinical Research Plan (Grant numbers: SEHHH-2021(KJ)-0669-KJB-485 and 2023(KJ)-0143-KYB-111 to Yufei Li), the Shenzhen Science and Technology Program (Grant 20240724152335001 to Yongcheng Jin), and the Featured Clinical Discipline Project of Shanghai Pudong Fund (Grant PWYts2021-07 to Hua Jiang). No competing interests to disclose. All data preprocessing and downstream analyses were conducted using standard bioinformatics tools on a Linux CentOS 8-based High-Performance Computing (HPC) system and R version 4.1.3, as provided by OxTium Technology Co. Ltd. Details of the tools, including their names, versions, and specific usage, are outlined in the Methods section. Unless otherwise specified, default parameters were employed for all tools. All individuals in this study were thoroughly informed about the objectives, procedure, and possible risks, and written informed consent was obtained before participation. Furthermore, the study protocol received ethical review and approval from the Ethics Review Committee of Shanghai East Hospital (EC.D(BG).016.02.1). All the cfRNA sequencing data have been submitted to the NCBI SRA database, accessed under the BioProject accession PRJNA1146172. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

  PublisherOA PDFDOI

• A computational approach for perturbation-induced EMT transitions

  npj Systems Biology and Applications · 2025-11-13 · 4 citations

  articleOpen access

  The Epithelial-mesenchymal transition (EMT) is a cellular state transition fundamental to development, wound healing, and cancer metastasis. The gene regulatory mechanisms underlying EMT have been extensively documented, revealing gene regulatory networks (GRNs) involving groups of mutually inhibiting transcription factors and microRNAs. Despite significant progress from both experimental and computational approaches, the details of how the EMT GRN initiates EMT in response to various external inputs is still not well understood. Here, we apply both Boolean and ordinary differential equation (ODE)-based methods to simulate a well-studied 26-node, 100-edge EMT GRN, examining its response to a wide range of single- and double-node perturbations. We evaluate the characteristics of effective EMT-inducing signals, particularly examining the amplifying role of transcriptional noise in determining the likelihood and mean transit time of an EMT. Together, these models establish a complementary framework for understanding and predicting drivers of EMT in the context of a GRN. We anticipate that this framework for a systematic study of in-silico GRN perturbations will be useful in developing increasingly accurate dynamical GRN models for various biological processes.

  PublisherOA PDFDOI

• Effects of individual and group factors on the social relationships of Budgerigars

  Avian Research · 2025-08-06 · 1 citations

  articleOpen access

  Social animals often form dynamic relationships with group members, which have been associated with increased social learning, survival and reproductive success. Social relationships can be shaped by both group and individual factors; however, few studies have addressed their combined impact. In this study, we aimed to examine whether group factors and individual factors jointly affect social relations. We selected Budgerigars ( Melopsittacus undulatus ) as the focal species to investigate the influence of sex, personality, and body length as well as sex ratio and group size on social relationships. The results showed that the birds in 3-individual groups had higher aggression network weighted degree values than those in 5-individual groups. Individuals within opposite-sex groups showed higher levels of aggressive and affiliative interactions than those in same-sex groups. Additionally, females attained higher social ranks despite exhibiting significantly lower aggression behaviors than males. Individuals with longer body lengths exhibited higher aggression network weighted degree values. Our results suggest that group factors primarily influence the social networks, while individual factors play important roles in shaping the social relationships.

  PublisherDOI

• Prognostic Significance of Preoperative Inflammatory and Liver Fibrosis Markers in Hepatocellular Carcinoma

  Journal of Inflammation Research · 2025-12-01

  articleOpen accessCorresponding

  Purpose: Prognostic heterogeneity remains a challenge in hepatocellular carcinoma (HCC) following curative resection. Inflammatory markers such as the platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), prognostic nutritional index (PNI), and systemic inflammation response index (SIRI), as well as liver fibrosis markers including gamma-glutamyltransferase-to-lymphocyte ratio (GLR), gamma-glutamyltransferase-to-platelet ratio (GPR), and AST-to-ALT ratio (AAR), have shown prognostic potential. This study aimed to evaluate their predictive value for survival after resection. Methods: We retrospectively analyzed 187 patients with HCC who underwent curative resection between 2008 and 2022. Preoperative inflammatory and fibrosis markers, along with clinical variables, were assessed using univariate and multivariate Cox regression for overall survival (OS) and disease-free survival (DFS). Kaplan-Meier analysis and nomogram models were used to estimate 1-, 3-, and 5-year survival. Results: Multivariate analysis identified elevated SII, AAR, low PNI, tumor thrombus, and postoperative complications as independent predictors of poor OS, while only tumor thrombus and SIRI independently predicted DFS. SII, AAR, and PNI were also significantly associated with aggressive tumor characteristics. Patients with all three adverse markers (high SII, high AAR, low PNI) showed significantly poorer OS and DFS. A prognostic score incorporating these markers and nomogram models were constructed to predict OS and DFS after resection. Conclusion: SII, AAR, and PNI are independent prognostic indicators for HCC following resection. The proposed prognostic score and nomograms may assist in individualized survival assessment and postoperative decision-making.

  PublisherOA PDFDOI

Recent grants

• New Computational Systems Biology Methods for Modeling Gene Regulatory Circuits

  NIH · $3.3M · 2018–2029

Frequent coauthors

• Eshel Ben‐Jacob

  122 shared

• José N. Onuchic

  Rice University

  120 shared

• Herbert Levine

  Northeastern University

  81 shared

• Dongya Jia

  63 shared

• Benjamin Clauss

  Tufts University

  53 shared

• Bin Huang

  Guangzhou University

  52 shared

• Sheng Li

  Jackson Laboratory

  47 shared

• Jianpeng Ma

  45 shared

Labs

• Computational Systems Biology LabPI

Education

• Ph.D., Biochemistry and Molecular Biology

  Baylor College of Medicine

  2010

• B.S., Physics

  Fudan University

  2003

Awards & honors

• Maximizing Investigators' Research Award (MIRA) from the Nat…