This repository contains code for a project that utilizes automation and statistical modeling to design and conduct informative microscopy experiments. The project involved building a custom fluorescent microscope and developing software for its control. Automation enabled the collection of large datasets for analysis. A statistical framework was developed to account for measurement distortions and improve experiment information. As a side project, the use of convolutional neural networks for cell targeting was also developed.
This project aimed to develop smart microscopy tools for designing and conducting informative microscopy experiments. It combined statistical theories with microscope automation to propose experiments that would yield the most valuable data. This approach has the potential to significantly improve the efficiency and effectiveness of microscopy research.
The project began with the construction of a custom-built fluorescent microscope. This involved aligning optical components, writing C++ driver code, and developing Python automation code to control the microscope's operation. Building a custom microscope provided greater control over the experimental setup and enabled the integration of automation for large-scale data collection.
Microscope automation played a crucial role in enabling the collection of large datasets. This approach helped to maximize the amount of information gathered despite potential measurement distortions that might occur during the experiment. Automation allowed for faster and more consistent data collection, reducing human ineffeiciencies and enabling the exploration of a wider range of experimental conditions.
The collected data was used to investigate how measured protein levels changed under different experimental conditions. A simple linear operation was employed to analyze the difference between a low quality experiment and a high quality experiment. This analysis provided an understanding of the relationship between experimental conditions and the information in the experiment. Further statistical modeling can be employed to uncover more complex relationships within the data.
Then, the development of a mathematical framework that utilized Fisher information to account for experimental measurement distortions. This framework was then applied to a large dataset collected using the automated microscope. The dataset specifically focused on measuring "probability distortions."
As a separate project undertaken alongside the main microscopy research, the team explored the use of convolutional neural networks and galvo driver code. This side project aimed to develop a system for destroying cells with undesirable phenotypes. Convolutional neural networks are a powerful tool for image recognition, and in this project, they were used to identify and target specific cell types.
