About

Qi Zhang is an Associate Professor in the Department of Chemical Engineering and Materials Science at the University of Minnesota Twin Cities. His research is concerned with various aspects of decision making in complex process systems. He develops computational methods for the efficient solution of real-world optimization problems and the data-driven discovery of unknown decision processes. His areas of application include sustainable energy and process systems, advanced manufacturing, supply chain engineering, human-machine systems, and bioengineering. Dr. Zhang holds a B.S. in Mechanical Engineering from RWTH Aachen University, an M.S. in Chemical Engineering from Imperial College London, and a Ph.D. in Chemical Engineering from Carnegie Mellon University. His work has earned numerous honors and awards, including the NSF CAREER Award, the Hutchison Medal, and the McKnight Land-Grant Professorship, among others.

Research topics

• Artificial Intelligence
• Computer Science
• Natural Language Processing
• Chemistry
• Business
• Ecology
• Engineering
• Natural resource economics
• Atomic physics
• Biology
• Organic chemistry
• Computational chemistry
• Physics
• Combinatorial chemistry
• Biological system
• Economics
• Chemical physics
• Environmental economics
• Electrical engineering
• Environmental science

Selected publications

• A Mechanistic Model of rAAV Production in Synthetic Cell Lines

  Biotechnology and Bioengineering · 2026-04-16

  articleOpen accessSenior author

  The recombinant adeno-associated virus (rAAV) is a widely used vector for gene therapy. Its manufacturing faces significant challenges in producing the large quantities of vectors needed for clinical applications and reducing empty particles. We have previously constructed synthetic cell lines that harbor all genes to enable scalable rAAV production. Through a series of design-construct-test cycles, productivity has increased to a level comparable to the traditional manufacturing method of triple plasmid transfection. In this work, we construct a mechanistic model of rAAV production in these synthetic cell lines to facilitate this design-construct-test cycle. The model was fit to the experimentally measured time profile of viral genome, transcripts, and proteins of viral components, capsid data, and packaged rAAV. The model recapitulates the trends in viral component dynamics with different inducer concentration time profiles, provides mechanistic insights into the effect of inducer profile on rAAV production, and predicts the effect of host cell genetic modifications. Finally, the model was used to optimize the inducer profile to increase vector genome production and full particle content in two separate cases. In both cases, the model-prescribed inducer concentration profile was experimentally performed and had the same trend of shift in VG and full particle content.

  PublisherDOI

• Design of microgrids with ammonia-based energy storage via Bayesian optimization

  Energy Conversion and Management · 2026-04-25

  articleSenior authorCorresponding
  PublisherDOI

• Coordinated Post-Disaster Restoration for Coupled Distribution-Transportation System

  IEEE Access · 2025-01-01 · 2 citations

  articleOpen access

  Effective coordination of diverse resources for rapid post-disaster recovery is crucial for ensuring power grid security. Urban power grids encompass varied load categories, and the recovery process is closely interconnected with transportation networks. To comprehensively address these challenges, this paper proposes a collaborative post-disaster recovery approach for urban coupled distribution-microgrid-transportation systems considering tiered load demands. First, differentiated electricity load demand characteristics are analyzed leveraging Long Short-Term Memory-AdaBoost (LSTM-AdaBoost) model, enabling precise identification of post-disaster priority user demand. Then the post-disaster model for coupled system is established incorporating the dispatch of repair crews within the transportation network and the support role of microgrids. The optimization problem is addressed by Benders decomposition technique. Numerical results demonstrate the effectiveness of the proposed post-disaster collaborative recovery and solution approach in accelerating power supply restoration.

  PublisherDOI

• Matrix Completion from Quantized Samples Via Generalized Sparse Bayesian Learning

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Design of Temperature-Controlled Linear Actuator Based on SMA

  Smart innovation, systems and technologies · 2025-01-01

  book-chapter
  PublisherDOI

• DNA Reactive Agents

  Elsevier eBooks · 2025-02-13

  book-chapter1st authorCorresponding
  PublisherDOI

• Recurrent Neural Networks for Forecasting Time-Varying Process Behavior in Mammalian Cell Culture

  Industrial & Engineering Chemistry Research · 2025-04-23 · 5 citations

  articleSenior authorCorresponding

  Cell culture processes are the workhorse for the production of therapeutic protein biologics. With advances in process data acquisition and monitoring, there has been an increasing interest in developing cell culture process models for control, optimization, and scale-up. However, the kinetic behavior of cell culture processes is highly complex. As culture time progresses, cell metabolism may shift, and at times, similar culture conditions may give rise to very different time-varying process behavior. Hence, modeling complex metabolic shifts in biomanufacturing processes remains a major challenge. In this work, we systematically evaluated the application of recurrent neural networks (RNNs) for forecasting the time profiles of key process parameters, including glucose and lactate concentrations, viable cell density, and viability, using a comprehensive set of fed-batch biomanufacturing data. We compared the RNNs’ performance with that of traditional machine learning models and feedforward neural networks, included total base addition in the model input to embed secondary process information, and extended the RNN model to an encoder–decoder model that leverages the history of seed train profiles to enhance the prediction of process behavior at the production scale. Overall, the computational results highlight the potential of RNN-based models for predicting key process parameters in cell culture and demonstrate the impact of process history on cell culture performance in biologics biomanufacturing.

  PublisherDOI

• Assessing the Impact of Chain of Custody Models on the Long-Term Planning of Low-Carbon Ammonia Supply Chains

  Industrial & Engineering Chemistry Research · 2025-01-09 · 5 citations

  articleSenior authorCorresponding

  With growing concerns about climate change, efforts to invest in low-carbon alternatives to conventional fossil-fuel-based products are on the rise globally. As more low-carbon products are developed, current systematic methods for optimizing supply chain expansion and operations need to be redesigned due to differences in consumer preferences between sustainable and conventional products. In this work, we investigate how the choice of traceability method (i.e., the chain of custody model), which governs a consumer’s claim to a sustainable product, impacts decision-making in the expansion of low-carbon ammonia supply chains. We propose a mixed-integer nonlinear programming formulation that applies the mass balance and book-and-claim models, the two most relevant chain of custody models in the chemical industry, to an ammonia supply chain. We then present a comprehensive case study spanning 25 years that includes nine U.S. states and highlights differences in solutions between the two models, such as the extent of renewable technology expansions, ammonia distribution decisions, demand met, and the breakdown of different costs under various ammonia price scenarios. Moreover, we make key observations in the solutions provided by a rolling-horizon simulation to cases with uncertain ammonia demand. This work highlights the critical role of chain of custody models in evolving low-carbon ammonia supply chains, laying the groundwork for future research into their application and impact on other sustainable supply chains.

  PublisherDOI

• Peptide Sequence and Cross-Link Structure Influence Translesion Synthesis Polymerase Bypass of 5-Formylcytosine-Mediated DNA–Peptide Cross-Links

  Chemical Research in Toxicology · 2025-07-28

  article1st author

  DNA–peptide cross-links (DpCs) are generated via the proteolytic cleavage of DNA–protein cross-links (DPCs), ubiquitous DNA lesions that block DNA replication and transcription. Translesion synthesis (TLS) DNA polymerases can facilitate replication bypass of DpC adducts in either an error-free or error-prone manner. We have previously demonstrated that local DNA sequence context significantly influences hPol η-mediated replication bypass of 5-formylcytosine (5fC)-mediated DpC lesions. However, the effects of peptide sequence on the efficiency and fidelity of the TLS bypass of 5fC-mediated DpC lesions remained unknown. In the present study, model DpCs containing three different peptides (NH2-GGGKGLGK*GGA-COOH, NH2-RPK*PQQFFGLM-COOH, and NH2-RPKPQQFK*GLM-COOH, K* = oxy-lysine) were subjected to primer extension experiments in the presence of TLS polymerases. We found that in vitro replication of DpC-containing templates by hPol η was more efficient than that catalyzed by hPol l or hPol κ. HPLC-ESI-MS and HPLC-ESI-MS/MS analyses of hPol η primer extension products indicated that the replication bypass of DpC containing NH2-RPK*PQQFFGLM-COOH was more error-prone than replication of the other two DpCs, leading to targeted C → T transitions, small deletions, and untargeted mutations downstream from the lesion. Steady-state kinetics investigation of hPol η-catalyzed nucleotide incorporation opposite the DpC lesions containing three different peptides revealed that, in all cases, error-free replication was far more efficient than incorporation of incorrect nucleotides. For mutagenic bypass, the catalytic efficiency of hPol η-mediated dAMP misincorporation opposite DpC with peptide NH2-RPK*PQQFFGLM-COOH was higher than adenine misincorporation across from the other two DpCs and unmodified dC. These steady-state kinetic findings were further explained by molecular modeling and molecular dynamics simulations, revealing that the three different DpC lesions impose varying perturbations to the geometry of the C–G and C–A pairs at the hPol η active site. Collectively, our results reveal that the peptide sequence and conjugation chemistry of DpC lesions can influence the fidelity of lesion bypass by TLS polymerases.

  PublisherDOI

• Transcriptomic functional characterization of recombinant adeno‐associated virus producing cell line adapted to suspension‐growth

  Biotechnology Progress · 2025-05-21 · 2 citations

  articleOpen access

  Recombinant adeno-associated virus (rAAV) is a widely used delivery vehicle in gene therapy. A scalable production technology is essential for its wide clinical applications. We have taken a synthetic biology approach to generate HEK293-based cell lines which harbor integrated genetic elements encoding essential AAV and adenoviral helper components and can be induced to produce rAAV. Through cycles of cell line enhancement, a high rAAV productivity could be achieved. The cell lines, like their parental HEK293, grew adherently. For scalable production, cell cultivation in suspension is highly desirable. A producer cell line GX6B was adapted to suspension growth in serum-free medium (named GX6Bs). However, it had substantially reduced virus titer. Returning GX6Bs cells to adherent culture conditions using adherent medium and cultured stationarily brought the productivity back to close to the level of adherent GX6B. A survey of the transcriptome revealed that induction and rAAV production elicited a wide range of cellular changes in various functional classes, including host immune defense response and nucleosome organization. The response was more subdued in suspension-growing GX6Bs. Upon reverting to adherent growth, the cellular transcriptome change regained its vigor to be more similar to that seen in GX6B. The GX6Bs maintained in suspension serum-free conditions were then reverted to the adherent culture medium but under an agitated culture environment to keep suspension growth for rAAV production. The productivity returned to within 25%-50% of GX6B. This work demonstrated the feasibility of the suspension culture of synthetic cell lines for the expansion and production of rAAV.

  PublisherOA PDFDOI

Recent grants

• CAREER: Optimization-Based Computational Discovery of Decision-Making Processes

  NSF · $576k · 2021–2027

• GOALI: Coordination of Multi-Stakeholder Process Networks in a Highly Electrified Chemical Industry

  NSF · $362k · 2022–2026

• Adaptive Robust Optimization with Endogenous Uncertainty and Active Learning in Smart Manufacturing

  NSF · $307k · 2021–2025

Frequent coauthors

• Paul J. Dauenhauer

  University of Minnesota System

  20 shared

• Omar Abdelrahman

  University of Minnesota

  20 shared

• M. Alexander Ardagh

  20 shared

• Phillip Christopher

  12 shared

• Pródromos Daoutidis

  University of Minnesota

  11 shared

• Anatoliy Kuznetsov

  9 shared

• Dionisios G. Vlachos

  9 shared

• Manish Shetty

  Mitchell Institute

  9 shared

Education

• Ph.D., Chemical Engineering

  University of Minnesota

  2009

• M.S., Chemical Engineering

  University of Minnesota

  2005

• B.S., Chemical Engineering

  Tsinghua University

  2003

Awards & honors

• AIChE CAST Outstanding Young Researcher Award, 2024
• Guillermo E. Borja Career Development Award, 2024
• Junior Sargent Medal, 2024
• Hutchison Medal, 2024
• McKnight Land-Grant Professorship, 2023