Starting Data Analysis and Obtaining Fluorometry Data

By

  






Starting Data Analysis and Obtaining Fluorometry Data

This past week I decided to finally sift through the crazy DLS data and see if I could shed some light onto what in the world is happening. I've included links to the google CoLab sessions I used to code everything up and I highly recommend taking a look at those if you're interested. However, I've also included some of the primary charts/graphs right here on this post as well. Let's get started!

Starting DLS Data Analysis

My first step in data analysis was to assemble all of the raw excel sheets I exported from the DLS instrument into one coherent data-analysis-ready file. I deleted some redundant data and/or data that didn't really make sense or add any value to my current circumstances (like time data and viscosity for example). I mainly focused on intensity, amplitude, PDI (polydispersity index), and a whole host of other things that go into calculating the hydrodynamic radius. I stored the whole file as a .csv file.

I loaded the .csv file into the google colab and then got to work using pandas (a python library for data work-up). I pretty much just had to fill in empty cells with something that wasn't empty, but also not 0 (which would change the statistics). I ultimately filled those cells with the mean of each column, so it was something the code could work with (floating point numbers) while not actually changing the shape of the data.

My first step was to perform what's called Principle Component Analysis (PCA) on my data. PCA is super useful when your data has a ton of features (different columns) like mine does. Each feature (PDI, radius, amplitude, etc.) can be thought of as a dimension so when I have 35 features per data sample, things can get complicated fast. PCA is a way to take higher dimensional data and pack it into a 2D or 3D space based on what aspects of the data result in the most variance (ultimately I still don't understand the math behind this). 

Additionally, PCA can also provide some information (but not comprehensively) on how the data clusters together. This is really what I want because I want to see how buffer with mercury ions is different than, say, buffer with both mercury and DNA in solution. Below is the first PCA I did on the first dataset I collected:
















Figure 1: PCA of DLS dataset including four different categories


I know it looks a little goofy here and if you want to see the whole graph and actually be able to rotate it in 3D please take a look at the colab link!

The yellow clusters correspond to buffer with DNA and mercury, so those are hopefully the nanoparticles I'm looking for. Purple corresponds to buffer with just mercury, orange corresponds to just buffer with DNA, and blue corresponds to just the background buffer.

As you can see, it's not perfect, but there is definitely a separation, and it's not too bad either. This is exactly what I want. I want to see DNA and mercury behave differently than just mercury or just DNA.


When I did this PCA, I just used HEPES as the buffer. When I tested it with PIPES, I got different variance and clustering and it actually wasn't as separated. It looked like with the PIPES, a lot of the signals seemed suspiciously close to that of just buffer and mercury.














Figure 2: PCA of DLS dataset including both HEPES and PIPES data


You can see that the yellow data points are not as separated from the other colors as they were in Figure 1. A lot of them occupy the same territory as the buffer and mercury data points. Again, please see the full figure in this colab page.


Additionally, in the colab notebook linked above, I included a UMAP graph and an HDBSCAN graph of the PCA data. UMAP  is pretty much a way of discovering more complex patterns in the data than PCA can. PCA only works for linear relationships and even-so it's a good overall indicator of what's going on. However, knowing how my DLS data behaves, it is definitely not linear all the time. UMAP is good at finding nonlinear patterns and it preserves both local and global data structure by using density-based clustering. You can see all of that form the google colab notebooks.

The next step is to do even more data analysis and then hopefully move on to developing a categorical classifier machine learning model to characterize each of the four types of signals! That way, when I feed it new data straight from the DLS instrument, I can have a good idea of if a signal if mostly DNA/mercury nanoparticles, or more likely just buffer/mercury interactions.


Moving on to Fluorometry

This week I also looked at analyzing the DNA/mercury solution via a spectrofluorometer. This instrument (in the way I used it) basically operates by exciting a given fluorophore (a fluorescent molecule) at its excitation wavelength and then scanning across a given range of wavelengths to analyze where the fluorophore emits light. A fluorophore always emits light at a higher wavelength than it's excited at.

So what exactly did I do? I used a fluorophore called Thiazole Orange (TO). TO works by intercalating between the base pairs of DNA. When it does so, it emits light around 535nm. However, when it is not able to bind to DNA and is just floating around in solution, it doesn't really give off much light. This is because when it is excited at 510nm, it usually gives off that energy via a rotational bond within the molecule; the energy is given off as rotational energy. However, when it is bound between stacked DNA bases, this bond is not free to rotate and therefore the energy is "forced" to escape as light, which the machine picks up on.

So I made samples with HEPES buffer, mercury, and my single stranded DNA and tested several controls in the instrument (just buffer with TO, buffer with mercury and TO, buffer with single stranded DNA and TO). Then, I tested buffer with DNA, mercury and TO to test whether the mercury was inducing the single stranded DNA to form particles where they are close enough for TO to intercalate. Long story short, the data was wonderful and pretty much confirmed this!




Figure 3: TO intensity vs. wavelength in sample with single stranded DNA


From this graph we can see that even the single stranded DNA is resulting in some emittance. This could just be because some of the single stranded DNA is interacting with itself. this is somewhat expected. This graph was kind of a close up on 520-560nm wavelengths. However, when we look at the next graph for DNA and mercury from a range of 450-700nm we get this:



Figure 4: TO intensity vs. wavelength in sample with single stranded DNA and mercury.

From this graph we can definitely see an immense peak right around 530-535nm, which is about the literature value of TO emittance.

Therefore, mercury is definitely complexing with the DNA and bringing them so close together that the Thiazole Orange is able to intercalate between them!

So, that's where I'm at now. Soon I may start preparing for Scanning Electron Microscopy analysis!

Thank you for reading and see you next time!




Where thoughts orbit stars and dreams power suns...












Comments

Post a Comment

Popular posts from this blog

By
Image
  Project Perseus Introduction Dear reader, Thank you for visiting the Project Perseus blog. I hope that the topics I write about bring you as much joy as they do for me. The world is a wonderful place...if you know where to look. My goal is to help readers reach that place of finding what's wonderful, beautiful, and captivating in this world by giving you a unique perspective and allowing you to take an innocent look at this universe's beauty. But before I begin, I should address some important questions you undoubtedly have: Who are you? My name is Nathaniel Comstock Moll. I am a current student in the fields of biochemistry, molecular biology, and physics. Outside of the technical aspect of my studies I enjoy writing novels, composing original pieces for the piano, doing some amateur backyard astronomy (sometimes volunteering at my school's local observatory), and now apparently writing blog posts. Essentially, I love to learn and to stay curious; two aspects that have c...
Read more

Project Perseus AI ~ Post #1 ~ Machine Learning Origins

By
Image
  Project Perseus AI ~ Post #1 ~ Machine Learning Origins This is the first post in hopefully a series of many that will document my educational journey through the topics of AI and machine learning. Every post will discuss a topic that I've reviewed and/or models that I've programmed to help me better understand these machine learning concepts.  What is the goal with these posts? Again, the goal is to primarily give a record of my self-taught journey in machine learning and to one day create my own AI model which I will name Perseus. I'm quite far away from that point at present, but if I'm diligent enough, I believe I can get there.  I started this journey because I've recognized that the 21st century will most likely be defined as the advent of organic and artificial intelligence cooperation. I believe that a healthy and beneficial cooperation is possible, but that it will require deep understanding on the humans' part and (hopefully) from the AI as well. My ...
Read more

Learning and Experimenting with Dynamic Light Scattering

By
Image
  Learning and Experimenting with Dynamic Light Scattering This past week at my Ole Miss REU research program has been...interesting to say the least. Not in a bad way, just in a I'm-going-insane-because-all-my-data-is-wacky-no-matter-what-I-do kind of way. The purpose of these experiments that I've been trying to do involves gaining data about the size and distribution of DNA/mercury complexes in solution. I outlined in my first post of this series how Dr. Wadkins has demonstrated that mercury complexes with DNA and induces it to form secondary structures. It's also postulated that these structures can be thought of as fully-fledged nanoparticles.  Dynamic Light Scattering (DLS) is a common way to obtain data about nanoparticles. Therefore, getting usable data from the DLS instrument from my DNA/mercury samples was my first order of business in order to develop an idea of how increased mercury concentrations correlates with particle size in solution. I thought it would be ...
Read more

Project Perseus AI ~ Post #3 ~ Evaluating Models and Learning TensorFlow

By
Image
  What did I Learn this Week? This week in my educational journey in machine learning, I learned how to track model metrics, such as accuracy and precision, I learned how another kind of model works, called binary classification , and finally I learned some Tensorflow basics. Tensorflow is a really neat library created for the python programming language that lets you easily set up and train models using its Keras API . I started going through some videos that taught me the basics of how to import data, compile the model, fit the model, and evaluate the model's effectiveness.  So, I've boiled down everything that I've learned in the past couple weeks into a post that briefly describes the ML concepts I went through and the code I used to practice those concepts. Let's get started! Basics of a Binary Classification Model A binary classification model is one that seeks to categorize data points as one option out of two possible options. For example, it could be used to ...
Read more

Project Perseus AI ~ Post #2 ~ Digging Deeper Into how Machines Learn

By
Image
  Project Perseus AI ~ Post #2 ~ Digging Deeper Into how Machines Learn This week I built off of what I learned from last week! I didn't take giant leaps in stretching my horizons of the subject, but rather, I really fleshed out some more basic principles (something I could probably do forever), but at some point I would have to move on. However, what exactly did I learn about this week? What I Learned this Week I primarily learned about what an activation function is, and what the different types of activations there are. I also learned about logistic regression , which is a logical "next step" to linear regression which I talked about in Post #1. In a way, last week I learned how machines learn, this week I learned about how they think which their learning is a result from. So I just went a little bit deeper into the mechanics of things. Okay, let's start off with the meat of the material: Logistic Regression Many times, in machine learning contexts, you want the sy...
Read more