Skip to content

Senior thesis work in the Londergan Lab modeling oxidative stress with machine learning and Raman spectroscopy.

Notifications You must be signed in to change notification settings

mklavin/thesis-mkl2

Repository files navigation

thesis-mkl2

The goal of this project is to determine the concentrations of reduced (GSH) and oxidized (GSSG) glutathione based on their Raman signals. Glutathione is an antioxidant which neutralizes free radicals by cycling between its reduced and oxidized forms. Increased amounts of free radicals in the body is known as oxidative stress and is correlated with higher risk for a number of diseases. Determining the ratio of reduced to oxidized glutathione can therefore describe an organisms state of oxidative stress for applications in disease prevention.

key ideas:

  • The integration of a peak in Raman spectra is directly proportionate to the concentration of its corresponding bonds.
  • The GSSG peak appears in the "580" region and the GSH peak appears in the "610" region. 580 & 610 denote the two areas Raman spectra are taken.
  • These regions are separated into two datasets and models are trained on both separately.

py files:

data_collection

functions for quickly organizing raw data into Pandas DataFrames.

modeling

ML model training and testing.

preprocessing

functions for preprocessing data, like normalizing, standardizing etc.

stratedgy_seach

uses some functions from preprocessing for a search algorithm which finds the best permutation of preprocessing methods and the best baseline removal function out of ~30 pybaslines functions. Baselines are evaluated based on the covariance between peak points and labels which would ideally be 1.

visualize

plotting functions.

folders:

data

contains Raman data- daniels raw Raman data, PCA processed data, Mimi's raw Raman data, and preprocessed data.

models

will contain finished models

plots

contains images of plots used for presentations or written work.

About

Senior thesis work in the Londergan Lab modeling oxidative stress with machine learning and Raman spectroscopy.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages