| Steps + | Description + |
|---|---|
| Sample Preparation | +
|
+
| Gel Preparation | +
|
+
| Gel Electrophoresis | +
|
+
| Staining and Visualization | +
|
+
| Analysis | +
|
+
| Applications + | Description + |
|---|---|
| Western Blotting [3] | +
|
+
| Enzyme Zymography [4] | +
|
+
| Extraction for mass spectrometry [5] | +
|
+
| Electrophoretic mobility shift assay [6] | +
|
+
| Steps + | Description + |
|---|---|
| Sample Preparation | +
|
+
| Gel Preparation | +
|
+
| Gel Electrophoresis | +
|
+
| Staining and Visualization | +
|
+
| Analysis | +
|
+
| Applications + | Description + |
|---|---|
| Western Blotting [3] | +
|
+
| Enzyme Zymography [4] | +
|
+
| Extraction for mass spectrometry [5] | +
|
+
| Electrophoretic mobility shift assay [6] | +
|
+
-
-* * *
-
-## Application to Bioinformatics
-
-**Clustering** is an important tool that is often used in bioinformatics because of its ability to group data into different groups that we can then analyze. Clustering algorithms start with a collection of *n* objects that we want want to divide into *k* clusters so that objects within a cluster are more "similar" to each other than objects in other clusters.
-
-A common application is to use clustering on gene expression data. By analyzing how different genes are grouped, we can predict functions of unknown by using known ones. We can also discover shared regulatory regions in DNA sequences and discover subtypes of a given disease.
-
-For example, [researchers](https://www.ncbi.nlm.nih.gov/pubmed/11707567) were able to classify different subtypes of lung cancer by analyzing the differences in gene expression between them.
-
-
-
-Bhattacharjee et al. (2001) Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses Proc. Natl. Acad. Sci. USA, Vol. 98, 13790-13795
-
-
-* * *
-
-
-## Distance Functions
-
-Before we move into a discussion of various clustering algorithms, let's go into the basics of how we determine how to group different data points into clusters.
-
-In clustering, we usually try to group things that are "similar" together into clusters. We use **distance functions** as a way to quantitatively measure how "close" data points are to each other. Lower distances mean that the points are closer together, while larger values indicate less similarity.
-
-There are several different ways to calculate the distance between two data points. The most common is the **Euclidean distance function**, which uses the distance formula to calculate the distance between two points.
-
-Another alternative is to use **Pearson's correlation** (aka Pearson’s r) to measure how correlated the two profiles are, with values ranging between -1 and 1.
-
-* * *
-
-## Clustering Methods
-
-One we determine how to measure how "close" data points are to each other, we need to determine how we measure the distances between different clusters.
-
-One method is using the **Unweighted Pair Group Method with Arithmetic Mean (UPGMA)**, which calculates the average distance from each point in the cluster to all other points in another cluster. Once that is calculated, the two clusters with the lowest average distance are then joined together, which will create a new cluster.
-
-Another clustering method is **Complete Linkage**. This method instead measures how similar 2 clusters are using the biggest dissimilarity between a member of one cluster and a member of another cluster. This method will tend to produce very tight clusters.
-
-Another method is to do **Single Linkage**. This method measures how similar 2 clusters are using the minimum dissimilarity between members of the 2 clusters. This method of clustering tends to produce clusters in long chains and can identify outliers readily.
-
-
-
-
-
-* * *
-
-## Hierarchical Clustering
-
-Now let's look at one algorithm for clustering - the hierarchical clustering algorithm.
-
-Using the hierarchical clustering algorithm will result in a dendrogram that groups every single data point into one large cluster, with similar data points grouped closer together.
-
-The basic algorithm for hierarchical clustering is as follows:
-1. Calculate the similarity between all possible combinations of two profiles using a distance function.
-2. Place each profile in a separate cluster.
-3. Group the two most similar clusters together to form a new cluster.
-4. Recalculate the similarity between the new cluster and all the other remaining clusters using a user-defined clustering method.
-5. Repeat steps 3 and 4 until all of the profiles end up in one large cluster.
-
-We can then "cut" our dendrogram at a level that gives us the number of clusters that we want.
-
-Here is an animation that shows how hierarchical clustering behaves.
-
-
-
-* * *
-
-## K-Means Clustering
-
-The other algorithm we will cover is the k-means algorithm. Unlike hierarchical clustering, we know beforehand the number of clusters that we want (k).
-
-The basic algorithm for k-means clustering is as follows:
-
-1. Choose a k-value for the number of clusters we want to end up with.
-2. Select k number of starting points that we want to initialize to start our clusters.
-3. For each data point, find the closest mean vector and assign the object to the corresponding cluster.
-4. For each cluster, update its mean vector according to the current assignments.
-
-
-
-We keep repeating the last two steps until a stopping criteria is met. Unlike the hierarchical clustering algorithm, the k-means clustering algorithm isn't always guaranteed to terminate. It can stop during **convergence**, when the algorithm no longer reassigns points, or it can run indefinitely until it stops at a user-defined number of iterations.
-
-This contrasts with hierarchical clustering which has a more finite and predictable termination step (when everything is inside of one cluster). Additionally, the k-means algorithm may produce different outcomes based on how we initialize our initial k points.
-
-Here is an animation that shows how k-means clustering behaves.
-
-
-
-
-* * *
-
-## Summary
-
-In summary, machine learning is a powerful tool that we can use to analyze data. In bioinformatics, we commonly use clustering algorithms to analyze our data, with both k-means and hierarchical clustering algorithms as viable options. In the future, it is likely that machine learning will be used to drive furhter innovations in the fields of personalized medicine and genomics.
-
-* * *
-
-## Sources
-1. Bhattacharjee et al. (2001) Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses Proc. Natl. Acad. Sci. USA, Vol. 98, 13790-13795
-2. Sheng Zhong "Intro to machine learning", BENG 183
-3. Derek Jow, Joey Sun, Victoria Tom "Introduction to Machine Learning", BENG 183
-4. https://commons.wikimedia.org/wiki/File:K-means_convergence.gif
-5. https://dashee87.github.io/data%20science/general/Clustering-with-Scikit-with-GIFs/
-
diff --git a/finalPaper/IntroToMachineLearning_2/classvsclust.png b/finalPaper/IntroToMachineLearning_2/classvsclust.png
deleted file mode 100644
index 041c7c7..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/classvsclust.png and /dev/null differ
diff --git a/finalPaper/IntroToMachineLearning_2/clustering.png b/finalPaper/IntroToMachineLearning_2/clustering.png
deleted file mode 100644
index 07db665..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/clustering.png and /dev/null differ
diff --git a/finalPaper/IntroToMachineLearning_2/examples.PNG b/finalPaper/IntroToMachineLearning_2/examples.PNG
deleted file mode 100644
index f4647f3..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/examples.PNG and /dev/null differ
diff --git a/finalPaper/IntroToMachineLearning_2/hClust.gif b/finalPaper/IntroToMachineLearning_2/hClust.gif
deleted file mode 100644
index a756ab3..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/hClust.gif and /dev/null differ
diff --git a/finalPaper/IntroToMachineLearning_2/kImg.png b/finalPaper/IntroToMachineLearning_2/kImg.png
deleted file mode 100644
index fed6200..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/kImg.png and /dev/null differ
diff --git a/finalPaper/IntroToMachineLearning_2/methods.png b/finalPaper/IntroToMachineLearning_2/methods.png
deleted file mode 100644
index 37eb140..0000000
Binary files a/finalPaper/IntroToMachineLearning_2/methods.png and /dev/null differ
diff --git a/finalPaper/MachineLearngin_DimensionReduction/Chen_Li.md b/finalPaper/MachineLearngin_DimensionReduction/Chen_Li.md
deleted file mode 100644
index bec2301..0000000
--- a/finalPaper/MachineLearngin_DimensionReduction/Chen_Li.md
+++ /dev/null
@@ -1,61 +0,0 @@
-# Machine Learning - Dimension Reduction
-By Chen Li
-## Introduction
-After we obtain an expression matrix RNA-seq, in which every row is a gene, and every column is a sample, how can we get a general sense of the distribution of the data and extract important features (genes) from the data matrix? Note that the expression matrix is usually in a high dimensional space because every gene is a dimension, so directly visualizing the data is impossible for human beings. Dimension reduction techniques such as PCA and t-SNE can map the original data in high dimension to lower dimension while retaining some spatial information (similarity and difference between samples). In this way, we can easily visualize the data in 2D space. In addition, PCA can return the *importances* of the features (genes), which are the ability of these features in grouping and separating the data points. By extracting the most important genes from the raw data matrix, we are able to explore the data such as clustering more easily.
-
-1. [PCA](#pca)
-2. [t-SNE](#t-sne)
-3. [PCA vs t-SNE](#pca-vs-t-sne)
-
-### PCA
-PCA (Principle Component Analysis) is widely used upstream of calculations that handle high dimensional data badly. PCA is a linear transformation method. It preserves the correlation between point x and y after transformation. PCA can reduce 4 or higher dimension graph to 2D or 3D. Let’s take a expression matrix for 6 mouse samples as an example. We will only use two genes for illustration. Youtube Reference: [StatQuest](https://www.youtube.com/watch?v=FgakZw6K1QQ)
-
-
-
-Now let's go through the steps in PCA calculation:
-1. The data set is first moved so that its center is at origin. Then PCA calculates the top components with the highest variations in the data. What it does is to fit a line to the data set. For an arbitrary line in the plane through the origin, the (sum of squared) distances to the projections of all the points are calculated and maximized. This has the same effect as minimizing all the distances between the points and the line. (In real world, calculation is done by linear algebra.)
-
-
-
-
-2. This line is called PC1, or **Principal Component** 1, and it captures the largest variation in the data. From the resulting line, we also know what its compositions based on the slope: it contains 4 parts Gene1 plus 1 part Gene2.
-
-3. PC2 is simply the line perpendicular to PC1. If the data is in 3D, PC2 will be residing in a plane perpendicular to PC1, and the previous steps will be repeated. All dotted lines are perpendicular to each other.
-
-
-
-
-4. In the final step, we rotate the coordinates so that PC1 becomes x-axis and PC2 becomes y-axis. The top two PCs are able to explain 94% of all variations in the data.
-
-
-
-**Note:**
-When performing PCA transformation, only the top several PCs are used, and other PCs (dimensions) are discarded. Information is *lost* in this process.
-
-### t-SNE
-By projecting data points to a plane of high variability, PCA tries to place dissimilar data points far apart and only preserves the global structure of data points (which means it may not be powerful enough to distinguish subgroups). So we will need t-SNE to see more detailed neighboring structures.
-
-
-
-
-[Image Source](https://www.kaggle.com/puyokw/clustering-in-2-dimension-using-tsne/code)
-
-1. So how does t-SNE work? The name stands for t-distributed stochastic neighbor embedding. The underlining mathematics of t-SNE is very advance and will not be covered here. But basiclly, it applies neighborhood preserving mapping so that distances between neighboring points are truthfully preserved after transformation.
-
-2. But how do we determine neighbors? *Perplexity* represents roughly the number of potential neighbors considered for a cluster, so we can determine neighbors of each point and thus clusters by trying different perplexity parameters until a reasonable and clear clustering is visualized by t-SNE. This is usually determined arbitrarily. Youtube Reference: [Applied AI Course](https://www.youtube.com/watch?v=FQmCzpKWD48&list=PLupD_xFct8mHqCkuaXmeXhe0ajNDu0mhZ&index=1)
-
-
-
-3. Meanwhile, since there is always some randomality in t-SNE’s embedding, we need to run multiple *iterations* to improve the 2D embedding to best represent the original structure. Such number of iterations is another hyperparameter to choose, and generally, the more iterations t-SNE runs, the more credible the resulting embedding will be.
-
-
-
-**Note:**
-* The size of each cluster in a t-SNE plot means nothing because it tends to expand dense clusters and shrink sparse ones.
-* Since t-SNE only represents distances within a potential cluster, distances between clusters do not provide any information.
-
-### PCA vs t-SNE
-Like many other machine learning techniques, PCA and t-SNE have their own advantages and disadvantages. When to use them depends on what you want to achieve. Please note that they are only two of the vast number of dimension reduction methods, and there are also ones that serve very similar purposes such as *ICA* and *UMAP*. [Reference](https://www.datacamp.com/community/tutorials/introduction-t-sne)
-1. t-SNE is much more computationally expensive than PCA.
-2. PCA is deterministic while t-SNE is not. Hyperparameters for t-SNE are somewhat arbitrary.
-3. Information is lost during PCA calculation while t-SNE attempts to capture information from all dimensions.
diff --git a/finalPaper/PrecisionMedicine/Mengyi_Liu.md b/finalPaper/PrecisionMedicine/Mengyi_Liu.md
deleted file mode 100644
index 661d089..0000000
--- a/finalPaper/PrecisionMedicine/Mengyi_Liu.md
+++ /dev/null
@@ -1,126 +0,0 @@
-# Precision Medicine - Where do we go from here?
-### by Mengyi Liu (Miko)
-
-### Sections:
-##### 1. Introduction to Precision Medicine
-##### 2. Personalized Medicine through the Lens of Asthma
-##### 3. Current Hot Topics on Precision Medicine
-
-## 1. Introduction to Precision Medicine
-Definition of Precision Medicine (by Dr. Su-In Lee): Tailoring of medical treatment to the individual characteristics of each patient, especially by using genetic or molecular profiling.
-
-In other words, precision medicine should account for personal variations in genomic sequence and environmental exposure.
-
-> Here is a story of how scientists used precision medicine to treat breast cancer.
-
-In the old paradigm, cancer types are usually distinguished by the origin of tumors. For example, lung cancer, liver cancer, breast cancer.
-
-However, tumors with the same origin may be different in appearance and behavior, aggressiveness, and vulnerability. Thus we need to treat each kind of tumor differently.
-
-The early treatment of breast cancer in the 1970s:
-- All patients underwent removal of ovaries: The assumption was, no estrogen, no growth of tumor.
-- It only helped around 70% of patients with ER+ tumor.
-- But how tumor cells are built matters: For example, markers on tumor cell surface and growth circuits can lead to different tumor progression.
-
-Then a discovery in 1970 found that:
-- 70% of breast cancer are ER (estrogen-receptor) positive.
-- These patients can be treated with anti-estrogen agent to block cancer growth.
-> A much better treatment, but what about the remaining 30% of the patients?
-
-1984 discovery:
-- 20% of breast cancers have abnormal HER-2 gene expression.
-- A new drug, Herceptin, was developed to inhibit the function of this protein.
-
-Mid-1990s findings:
-- 5% of breast cancers have inherited defect in gene BRCA1 or BRCA2.
-- No preventive treatment, but can screen for such inborn defect and watch for tumor formation closely.
-
-Fortunately, the treatments using anti-estrogen agent or Herceptin can be effective for around 85% of the breast cancer patients. This improvement in cancer therapy is largely contributed to the effort to split the patients into different subgroups, then treat each subgroup on its own based on its distinct characteristics. That is the core of precision medicine. We will see how scientists implement this "splitting" in the next section as well.
-
-Note that the 5% of breast cancer with inherited defect in gene BRCA1 or BRCA2, are categoried as Triple Negative Breast Cancer (TNBC). This type of breast cancer does not express the genes for estrogen receptor (ER), progesterone receptor (PR) and HER2. And it is still in need of an effective treatment.
-
-In order to develop personalized treatment and advance the field of precision medicine, current genomics and machine learning researchers need to address:
-
-- identifying genetic or molecular markers for clinical phenotypes
-- discovering disease subtypes from genetic and/or molecular data
-- building prediction models for clinical events based on electronic medical record (EMR) data
-
-> Let's look at an example where scientists used the theory of precision medicine to study asthma.
-
-## 2. Personalized Medicine through the Lens of Asthma
-
-#### 1) What is Asthma?
-A condition where a person's lungs become inflamed, and produce extra mucus, making it hard to breath.
-
-Biology behind asthma: An excessive allergen-induced type 2 inflammation, orchestrated by memory CD4+ T cells that produce type 2 cytokines (Th2 cells).
-
-In a similar disease, rhinitis, its pathway also involves Th2 cells.
-
-But unfortunately, there is currently no cure for asthma. Newer therapies are only partially successful in certain subtypes.
-
-
-
-Thus we need a better understanding of the disease at a molecular level.
-
-#### 2) A precision medicine study on asthma
-
-The figure below shows how scientists split the asthma patients into 4 subgroups.
-
-They took into account 8 parameters obtained from patients that were correlated with their disease and the severity, and did a partition-around-medoids clustering.
-
-
-
-In the study, *Transcriptional profiling of Th2 cells identifies pathogenic features associated with asthma*, the researchers performed RNA-Seq on Th2 cells from a total of 80 samples from 77 patients, including 3 biological replicates.
-
-Then they did RNA-Seq analysis to identify genes differentially expressed between allergic asthma, rhinitis and healthy control groups, by performing negative binomial tests for pairwise comparisons employing the Bioconductor package DESeq2.
-
-They found the following:
-
-- They identified a total of 15 distinct gene modules.
-- DESeq analysis found 500 genes differentially expressed between asthmatic subjects and healthy subjects (Genes for apoptosis, zinc transporters, MAPK, NF-κB, TNF).
-- The expression of most of these genes was similar between rhinitis and asthma.
-- Genes that differentiate asthmatic from healthy subjects show an intermediate phenotype in allergic rhinitis subjects.
-
-These genetic markers identified through this study could offer insights for personalized medicine for asthma patients.
-
-## 3. Current Hot Topics on Precision Medicine
-
-> The study of precision medicine has gained an increased interest in the scientific community in the past decade. Some interesting work are presented as follows.
-
-#### 1) Mapping of patients data
-
-Atul Butte, a Distinguished Professor and the Director of the Computational Health Sciences Institute at UCSF, constructed a map illustrating how each individual patient developed different subtypes of diseases and how these diseases lead to mortality.
-
-Here is the map showing the paths for each patient:
-
-Each square and circle represent a certain subtype of disease, and each line represents a transition.
-
-
-
-This map offers physicians and scientists possible patterns of disease progression, which can help guide the treatments for future patients. Clearly, this is one way to *build prediction models for clinical events based on electronic medical record (EMR) data*. The amount of detailed information encompassed by this map is powerful and grants possibility for precision care for patients.
-
-> Let's look at another application of precision medicine.
-
-#### 2) Opening the black box of machine learning models
-
-Su-In Lee, an Associate Professor at Paul G. Allen School of Computer Science & Engineering at UW, found weaknesses in the conventional way to identify molecular markers. She tries to interpret the complex machine learning models that are used to analyze patient data, and strives to offer individualized explanations for a particular prediction, and for a particular patient.
-
-
-
-
-These are both very interesting and cutting edge research efforts in the field of precision medicine. So what's next in precision medicine? Well, let's keep in mind the three things that scientists may want to address:
-
-- identifying genetic or molecular markers for clinical phenotypes
-- discovering disease subtypes from genetic and/or molecular data
-- building prediction models for clinical events based on electronic medical record (EMR) data
-
-Whether it be cancer vaccines or personalized drugs for a particular subtype of cancer, the biomedical research community will witness a growth in precision cancer care. We all have the potential to make what's next.
-
-
-
-## References:
-1. Seumois, Grégory et al. “Transcriptional Profiling of Th2 Cells Identifies Pathogenic Features Associated with Asthma” Journal of immunology (Baltimore, Md. : 1950) vol. 197,2 (2016): 655-64.
-2. Su-In Lee Lab: https://suinlee.cs.washington.edu/research
-3. Su-In Lee: "Interpretable Machine Learning for Precision Medicine" https://youtu.be/La7KTIe2DeU
-4. Precisely practicing medicine with a trillion points of data. | Atul Butte | TEDxSanFrancisco https://www.youtube.com/watch?v=fbZZ_1Jbm6w
-5. Zhong, Sheng. BENG 183 lecture on precision medicine. https://docs.google.com/presentation/d/1VD0KbnLThYzqJ9eN6EWcOhd5i726QfYc8KnCt7bUhPM/edit#slide=id.g404a8c7ebe_0_517
diff --git a/finalPaper/PrecisionMedicine_2/Reysha_Patel.md b/finalPaper/PrecisionMedicine_2/Reysha_Patel.md
deleted file mode 100644
index 518dec9..0000000
--- a/finalPaper/PrecisionMedicine_2/Reysha_Patel.md
+++ /dev/null
@@ -1,118 +0,0 @@
-1. [Introduction - What is Precision Medicine?](#1)
-2. [Overview of Precision Medicine/Pipeline](#2)| Attribute / Parameter - | Whole Genome Sequencing (WGS) - | Whole Exome Sequencing (WES) - | Targeted Sequencing - |
|---|---|---|---|
| Information Level | -Everything | -Genes | -User Defined | -
| Cost Per Sample | -$5000 | -$2000 | -$200 | -
| Low Frequency Mutation Detection | -Not Possible | -Not Likely | -Yes | -
| DNA Input Amount | -1 ug | -100 - 200 ng | -10 ng | -
| # of Samples in Parallel | -1 | -2 | -96 | -
| Method - | Targets - | Resolution - | Notes - |
|---|---|---|---|
| 3C [3] | -one-vs-one | -~1–10 kb |
-
|
-
| 4C [4] | -one-vs-all | -~2 kb | -
|
-
| 5C [5] | -many-vs-many | -~1 kb | -
|
-
| Hi-C [6] | -all-vs-all | -0.1–1 Mb | -
|
-
| ChIA-PET [7] | -Interaction of whole genome mediated by protein | -Depends on read depth and the size of the genome region bound by the protein of interest | -
|
-