What I Left Behind

By Sophia Wang @ 2025-12-09T02:52 (0)

I used to think that some limits were meant to be broken.

After painfully hitting a knowledge ceiling I have now learned to respect these innate limitations. I no longer care about my academic status in relation to higher education. I still hold great respect and gratitude towards MIT and other institutions but have accepted that it is better for me to depart from previous academic accomplishment goals.

It is hard to leave things behind until recognizing them in a safe, contained, and orderly manner that considers potential harm if kept secret versus shared. I am publishing this to hold myself accountable for some previous ignorance and not having enough patience, time, and understanding to fully appreciate the creation process and functions of the advance tools I was eventually asked to use.

Some details may be of questionable accuracy, i.e. the precise mechanisms of Alzheimer's diease. 

In Feb/March 2024 I was an undergraduate, I had just changed my major to physics. After several email exchanges and meetings with a grad student supervisor and materials science faculty, I submitted a proposal to the department for academic credit for an undergraduate research opportunity. This credit proposal was approved.

However the scope of the project involved tools and methods far beyond my competence back then. The project focused on investigating and defining a possible relationship between sequence and structure of protein aggregates involved in the advancement of Alzheimer's function, which comes with neurodegeneration. inhibiting microglia function, and other biological details I have yet to be able to fully and accurately explain. The task involved using computational models that were trained on cryoEM images of folded proteins to predict the structure of more complex proteins with different confidence levels for different regions. The task asked for comparison of the ability of different encoding models to convert different sequences to vector form for further proccessing. Each letter of a protein sequence corresponds to a physical amino acid with differing physical and chemical properties. A researcher can choose to represent each amino acid differently: these methods contain different types and amounts of information. The two types of information I was asked to compare were sequences (and their various representations) and the confidence level of protein structure prediction.

I may add in more detail later on the exact methods I used but they did not extend very far beyond binary and multiclass classification tasks based on thresholds of accuracy set arbitrarily. I did not finish this semester successfully.