Hi, I'm Mitchell Daneker

I'm a graduate student and researcher passionate about leveraging machine learning, data science, and bioengineering to tackle complex biological challenges. My research focuses on cutting-edge areas such as Genetically Encoded Voltage Indicators (GEVIs), protein engineering, and modeling and simulation in biopharmaceutical processes.

On this GitHub, you'll find projects, code, and publications showcasing my experience in machine-learning-guided protein design, machine learning for solving ODEs/PDEs, and advanced data analytics. I'm particularly interested in combining computational techniques with biological research to drive innovation in healthcare and biotechnology.

Feel free to explore my work, or contact me at mitchelldaneker@gmail.com if you're interested in collaborating!

Papers

1. Engaging Alkenes and Alkynes in Deaminative Alkyl–Alkyl and Alkyl–Vinyl Cross-Couplings of Alkylpyridinium Salts
   My first paper during undergrad at the University of Delaware with Dr. Mary P. Watson. I contributed to developing new cross-coupling reactions.

2. Systems Biology: Identifiability Analysis and Parameter Identification via Systems-Biology-Informed Neural Networks
   Explores how physics-informed neural networks (PINNs) can solve inverse problems for biological ODEs. The paper also addresses structural and practical identifiability of parameters.
   GitHub Code

3. Effective Data Sampling Strategies and Boundary Condition Constraints of Physics-Informed Neural Networks for Identifying Material Properties in Solid Mechanics
   Investigates how various data sampling techniques can enhance PINN performance in inverse problems related to solid mechanics.
   GitHub Code

4. Transfer Learning on Physics-Informed Neural Networks for Tracking Hemodynamics in the Evolving False Lumen of Dissected Aorta
   Uses PINN and MRI data to resolve the hemodynamic domain in aneurysms and introduces a transfer learning technique that reduces runtime across varied boundaries and resolutions.
   GitHub Code

5. Identifying Heterogeneous Micromechanical Properties of Biological Tissues via Physics-Informed Neural Networks
   Solves inverse problems to learn parameter fields in realistic 2D tissues using PINN, outperforming other machine learning techniques.
   GitHub Code

6. ADEPT: A Noninvasive Method for Determining Elastic Properties of Valve Tissue
   Introduces ADEPT, a method using 3D transesophageal (TEE) imaging to predict patient-specific material properties of the tricuspid valve for better surgical modeling.
   GitHub Code (Under Revision)

7. Physics-Informed and Black-Box Identification of Robotic Actuator with a Flexible Joint
   Applies PINN to infer friction models in robotic actuators, comparing results to traditional black-box approaches.

Current Work

1. SBINN Continuation:
   Collaborating with researchers at Johns Hopkins to solve inverse problems for large scale cell models, focusing on methods for low-data scenarios.

2. DeepONet Uncertainty Quantification (UQ):
   Developing UQ methods for Deep Operator Networks, enhancing their potential for active learning in biological systems.

3. DeepONet Surrogate Models:
   Creating surrogate models for spatiotemporal biological systems, simplifying complex models to temporal-only formats with UQ integration.

4. Genetically Encoded Voltage Indicators (GEVIs):
   Applying machine learning to optimize GEVIs for brain studies through active learning techniques, aiming to improve experimental conditions.