About

Minshuo Chen is an Assistant Professor of Industrial Engineering and Management Sciences at Northwestern University. His research focuses on developing principled methodologies and theoretical foundations of learning and decision-making. He is particularly interested in Generative AI, including diffusion models for data modeling and beyond, as well as the foundations of learning such as approximation, optimization, and efficiency. His work also encompasses deep reinforcement learning, with applications in diagnosis and control in complex systems. Chen has contributed to the understanding and advancement of diffusion transformers, statistical rates of conditional diffusion transformers, theory of consistency diffusion models, and efficient reinforcement learning with impaired observability, among other topics.

Research topics

• Chemistry
• Organic chemistry
• Photochemistry
• Atomic physics
• Physics
• Chemical physics
• Chemical engineering
• Molecular physics
• Nuclear chemistry
• Materials science
• Optics
• Physical chemistry
• Crystallography

Selected publications

• Asymmetric Dialysis: Truly Unified and Simultaneous Single‐Pass Concentration and Buffer Exchange

  Biotechnology and Bioengineering · 2026-05-12

  articleOpen access

  ABSTRACT Biopharmaceutical manufacturing has been using ultrafiltration (UF) and diafiltration (DF) for buffer exchange, desalting, and formulation of biologics. The legacy UF/DF is commonly a two‐step batch process that is challenging to integrate into end‐to‐end continuous biomanufacturing. Here, we introduce asymmetric dialysis, a novel one‐step continuous process that combines UF and DF. It works by utilizing asymmetric flow between the inlet and outlet of the retentate and complementary flow of the dialysate solution, achieving product concentration, buffer exchange, and salt removal using a commercially available hollow fiber device. Asymmetric dialysis can achieve product concentrations of 105 (3.8×), 200 (10×), and 64 g/L (9.4×) starting from feed concentrations of 30, 20, and 7 g/L, respectively, with modest pressures of 6–7 psi. The interplay between feed and exchange buffer flow rates was exploited to make the process sustainable by reducing buffer consumption by 75% (25 L/kg mAb) compared to conventional batch UF/DF (100 L/kg, mAb). We successfully processed 7 kg of mAb at 20 g/L feed using 5‐day asymmetric dialysis with a daily productivity of 0.8 kg/m 2 /day to product concentration of 200 g/L. These results demonstrate the potential of asymmetric dialysis, a simple, sustainable, and low‐cost bioprocessing technology for continuous bioprocessing.

  PublisherDOI

• Exploring Rapid-Response Flyby Reconnaissance Mission Architectures Enabling Asteroid Mass Estimation

  2026-03-07

  article

  Flyby missions are the fastest, cheapest, and sometimes only means of obtaining critical measurements for Planetary Defense (PD) scenarios. Past missions, primarily focused on asteroid science, have measured the orbit, binarity, shape, and rotation of asteroids. Mass, one of the most important asteroid characteristics for PD scenarios, has unfortunately proven much more difficult to measure via flyby, only being possible with certain combinations of large targets and slow flybys. An adaptation of a relative tracking scheme, a measurement technique previously used to map the gravitational fields of the Earth and Moon, has the potential to substantially increase flyby mass measurement sensitivity. A relative tracking scheme adapted for a small-body flyby is envisioned to consist of two spacecraft conducting simultaneous flybys of a target. If at least one of the spacecraft passes very close to the asteroid, its change in motion induced by the small body's gravity can be observed through relative measurements of the other spacecraft. This paper begins by establishing some high-level mission objectives for a rapid reconnaissance architecture that includes a mass estimate. High-level trades pertaining to the flight system architecture, relative tracking payload selection and imaging payload configuration are then considered. In conclusion a concept of operations and flight system capable of measuring the most critical hazardous asteroid characteristics, including asteroid mass are presented.

  PublisherDOI

• Insights into mRNA lipid nanoparticle polydispersity and shape using quantitative solution biophysics

  Structural Dynamics · 2025-03-01 · 1 citations

  articleOpen access

  Lipid nanoparticles (LNPs) are the most advanced delivery system currently available for RNA therapeutics. Their development has accelerated rapidly since the success of Patisiran, the first siRNA-LNP therapeutic, and the SARS-CoV-2 mRNA vaccines that emerged during the COVID-19 pandemic. Designing LNPs with specific targeting, high potency, and minimal side effects is crucial for their successful clinical use. However, our understanding of how the composition and mixing methods influence the structure, biophysical properties, and biological activity of the resulting particles remains limited. While microfluidic technologies have significantly improved the speed and uniformity of LNP production, a major challenge that remains is that ~60-80% of mRNA-LNP formulations are unloaded (empty lipid particles). This study tackles this challenge by relating current standard characterization methods with more powerful emerging methods, including 1. multi-wavelength analytical ultracentrifugation (MWL-AUC), 2. In-line multi-angle light scattering (MALS) methods, and 3. synchrotron size-exclusion chromatography in-line with small-angle X-ray scattering (SEC-SAXS) coupled with singular-value decomposition methods (SVD). We will present the strengths and weaknesses of each approach and showcase the increased detail newer advanced methods provide by comparing LNP formulations made using two common small-scale production methods: microfluidic rapid mixing and bulk mixing. The characterization techniques employed here can enhance our understanding of LNP structure-function relationships and enable researchers to define their RNA LNP products more precisely, which can improve LNP quality and potentially accelerate pharmaceutical development.

  PublisherDOI

• Elucidating lipid nanoparticle properties and structure through biophysical analyses

  Nature Biotechnology · 2025-10-23 · 14 citations

  article
  PublisherDOI

• Correction to “Optical Detection of Interleukin-6 Using Liquid Janus Emulsions Using Hyperthermophilic Affinity Proteins”

  ACS Omega · 2024-10-04

  erratumOpen access1st author

  [This corrects the article DOI: 10.1021/acsomega.4c03959.].

  PublisherDOI

• Elucidating lipid nanoparticle properties and structure through biophysical analyses

  bioRxiv (Cold Spring Harbor Laboratory) · 2024-12-22 · 4 citations

  preprintOpen access

  Designing lipid nanoparticle (LNP) delivery systems with specific targeting, potency and minimal side effects is crucial for their clinical use. However, traditional characterization methods, such as dynamic light scattering, cannot accurately quantify physicochemical properties of LNPs and how they are influenced by the lipid composition and mixing method. Here we structurally characterize polydisperse LNP formulations by applying emerging solution-based biophysical methods that have higher resolution and provide biophysical data beyond size and polydispersity. These techniques include sedimentation velocity analytical ultracentrifugation, field-flow fractionation followed by multi-angle light scattering, and size-exclusion chromatography in-line with synchrotron small-angle X-ray scattering. We show that LNPs have intrinsic polydispersity in size, RNA loading, and shape, which depends on both the formulation technique and lipid composition. Lastly, we predict LNP transfection in vitro and in vivo by examining the relationship between mRNA translation and physicochemical characteristics. Solution-based biophysical methods will be essential for determining LNP structure-function relationships, facilitating the creation of new design rules for LNPs.

  PublisherDOI

• Correction to: Magnetic Resonance Spectroscopy (MRS) in Alzheimer’s Disease

  Methods in molecular biology · 2024-01-01

  erratumOpen access
  PublisherOA PDFDOI

• Pearls & Oy-sters: KLHL11 IgG Paraneoplastic–Associated Hearing Loss and Rhombencephalitis in a Woman With Metastatic Müllerian Tumor

  Neurology · 2024-03-14 · 7 citations

  articleOpen accessCorresponding

  Kelch-like protein-11 (KLHL11) immunoglobulin G (IgG) is a recently reported paraneoplastic autoantibody associated with rhombencephalitis, which commonly presents with ataxia, diplopia, vertigo, hearing loss, tinnitus, and gaze palsies. The association of this high-risk paraneoplastic autoantibody with testicular germ cell tumors is widely accepted, but it has not been associated with Müllerian tumors. In this study, we report a woman without a known germ cell tumor presenting with signs and symptoms suggesting autoimmune encephalitis. She was found to have metastatic ovarian serous carcinoma with KLHL11 immunoreactivity on histopathology. This case demonstrates a rare cancer association of KLHL11 IgG-seropositive rhombencephalitis with Müllerian tumor and highlights that this autoantibody can also be detected in female patients. Thus, this case expands on the current knowledge of KLHL11-related autoimmune encephalitis including the paraneoplastic presentation, associated tumor types, and management of this syndrome in women.

  PublisherOA PDFDOI

• Analyzing Doxorubicin and Cardioprotective PLGA Nanoparticles on Murine Cardiomyocytes using Machine Learning Modeling to Mitigate Cardiotoxicity

  2024-12-03

  article

  Doxorubicin (DOX) is a chemotherapeutic drug used to treat cancerous cells by inducing immunogenic cell death (ICD) and interfering with DNA replication. The viability of cardiomyocytes decreases significantly when treated with DOX and results in a cardiomyopathy mortality rate upwards of 50%, but this is reduced when encapsulated in a Poly-lactic-co-glycolic acid (PLGA) nanoparticle. We collected and treated murine cardiomyocytes in-vitro grouping untreated, DOX, and DOX with PLGA cells, and characterized them utilizing a fluorescence-activated cell sorting technique, cell flow cytometry. Using machine learning, we found which characteristics make PLGA effective at performing drug-carrying capabilities. We created a random forest model for processing the murine cells to predict which of the three treatments it received. We perform feature engineering utilizing correlation matrices between the features, Variation Inflation Factor (VIF) scores, and their Permutation Importance (PI) scores. Our random forest model achieved a 92.63% accuracy in predicting treatments with a dataset composed of highly independent features. By examining the most important features for the model’s accuracy, cell death, DOX, and granularity amount, we discern that because PLGA restores the lysosomal compartment of cells, it allows healthy cells to recycle their cytoplasmic structures. We also discovered that PLGA administers less of the toxic drug when cells are not tumorous, and restores the cellular membranes of cells after they are treated with DOX.

  PublisherDOI

• Magnetic Resonance Spectroscopy (MRS) in Alzheimer’s Disease

  Methods in molecular biology · 2024-01-01 · 8 citations

  article
  PublisherDOI

Frequent coauthors

• Michael R. Wasielewski

  Northwestern University

  39 shared

• Mercouri G. Kanatzidis

  Northwestern University

  24 shared

• Constantinos C. Stoumpos

  24 shared

• Timothy M. Swager

  Massachusetts Institute of Technology

  19 shared

• Weijun Ke

  Wuhan University

  17 shared

• Eleftherios P. Diamandis

  Mount Sinai Hospital

  15 shared

• Hsinhan Tsai

  National Taiwan University

  15 shared

• Aditya D. Mohite

  Rice University

  15 shared

Education

• Ph.D., Chemistry

  Northwestern University

  2020