Preliminary Analysis of the Role of Trace Elements in Kidney Stone Pathogenesis for Enhanced Early Detection
Naina Kumar & Nishita Paruchuri
Lay Summary:
Our project aims to understand the role of trace elements-minerals that exist in small amounts- in different kidney stone formation using machine learning and statistical analysis. Our project is in collaboration with the Sunita Ho Lab at the University of San Fransisco which found significance of certain proteins in stone formation that we use in our analysis.
Abstract:
600,000 Americans suffer from kidney stone disease each year. Kidney stone disease is when hard deposits of minerals and salts form in the urinary tract. Although most kidney stones are composed of these minerals, their crystallization is influenced by the bioaccumulation of trace elements such as zinc, copper, nickel, aluminum, strontium, cadmium, and lead. Zinc has been shown to play an important role in nucleation (association of free ions into crystalline-like particles), the initial process of stone formation. However, its role in stone-forming pathways is still not widely understood. The purpose of the study is to understand the function of zinc by analyzing the correlation between the expression of protein markers (TRPV4, Piezo1, MFN2, HIFα) in stone formers vs. non-stone formers, involved in stone formation, and zinc accumulation using machine learning and deep learning models. In collaboration with the Sunita Ho Lab at the University of California San Francisco, this study developed a Multi-layer perceptron (MLP) and Random Forest to predict the average nuclei intensity of the proteins from the average zinc intensity. Furthermore, this study uses robust linear regression to analyze these associations. Results indicate a significant positive association between average zinc intensity and Piezo1 average nuclei intensity, whereas the other associations remain mixed across patients. Future work for this study includes examining zonal instead of whole-slide correlations, experimenting with various deep-learning models, including GNNs, and understanding the function of other trace elements in kidney stone formation.
Q&A:
Bios: Naina Kumar,Nishita Paruchuri
Program Track: Advanced Research
GitHub Username:
https://github.com/NainaMKumar -Naina Kumar
https://github.com/Nishita1234-cloud -Nishita Paruchuri
What was your favorite seminar? Why?
My favorite seminar was the one where Jason Wei from OpenAI attended. I enjoyed learning about deep learning concepts such as quantization and AutoML and getting his insights on working in a startup vs. a big company like Google as well as current blockers in building medical LLMs. -Naina Kumar
My favorite was Aruesha’s seminar. I found her project about colon cancer very insightful, and it was interesting to learn about how they utilized TRACE to coregister multiple modalities. It was very fun learning about other applications of TRACE, especially since we used TRACE in our project as well. -Nishita Paruchuri
If you were to summarize your summer internship experience in one sentence, what would it be?
Intense, high-growth learning opportunity -Naina Kumar
An amazing learning experience -Nishita Paruchuri