Counting Transcripts - Extension.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Counting Transcripts - Extension\n",
    "\n",
    "> This is part four of the Counting Transcripts series. We will use the knowledge from the previous parts to evaluate differentially expressed transcripts. This notebook should be located in the same folder as the previous notebooks.\n",
    "\n",
    "## Setup\n",
    "\n",
    "> As usual, we will first need to setup our working directory."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "###IMPORTANT: if you ever restart this notebook, you MUST rerun this code cell\n",
    "import os\n",
    "\n",
    "#feel free to change this location, because we will be working in this folder for the entire tutorial\n",
    "os.environ['WORKDIR'] = './data' "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Extension 1: Evaluating differentially expressed transcripts \n",
    "\n",
    "From this one sample, we can see that some transcripts are expressed more or less than other transcripts, but this data does not give any insight about the effect of changes in transcript abundance and cellular behavior. This is why we need to compare the transcript expression of different samples. For example, imagine we would like to address the following question: which transcripts are responsible for making this a cancerous cell? We would look at the transcript expression between a normal cell and a cancerous cell, and analyze transcripts that are differentially expressed between the two samples. In this extension, we will begin looking at differentially expressed transcripts by visualizing the levels of transcription between different samples.\n",
    "\n",
    "We will continue to use samples from the adaptive prediction paper ([PMID 28854640](https://www.ncbi.nlm.nih.gov/pubmed/28854640)) to analyze the effects of caffeine on yeast at the transcriptome level. \n",
    "\n",
    "Before we download the data from the other samples, let us clean our working directories. For the following folders `$WORKDIR/BAM/`, `$WORKDIR/cleanFASTQ/`, `$WORKDIR/cufflinks/`, `$WORKDIR/FASTQ/`, and `$WORKDIR/kallisto/`, create new folders in each folder specifying the experiment (1a, 1c, 2a, 2c, 3a, 3c). In the pipeline tutorials, we were working with data from 1a, so move any existing files into your new 1a folder. However, in the kallisto folder, leave the transcripts.idx file under /data/kallisto because we can reuse the kallisto index and it should be shared between the samples.\n",
    "\n",
    "For example, your `$WORKDIR/kallisto/` folder should look something like this:\n",
    "<html>\n",
    "<div style=\"width: 50%;margin: 0 auto; background-color: rgb(242, 242, 242);\">\n",
    "    <code><i>kallisto</i>/\n",
    "    transcripts.idx\n",
    "    1a/\n",
    "        abundance.h5\n",
    "        abundance.tsv\n",
    "        run_info.json\n",
    "    1c/\n",
    "    2a/\n",
    "    2c/\n",
    "    3a/\n",
    "    3c/</code>\n",
    "</div>\n",
    "</html>\n",
    "\n",
    "We will now download the sequencing data from replicate 1 after caffeine, replicate 2 before and after caffeine, and replicate 3 before and after caffeine. Research scientists use replicates to check that outcomes are not due to chance. This is different from a trial, because a **trial** refers to a certain action performed many times repeatedly using the exact same object, while a **replicate** would perform a certain action on many of the same object. \n",
    "\n",
    "An example of a trial is flipping the same coin three times, while replicates would be flipping three different coins.\n",
    "\n",
    "\n",
    "| Replicate # | Before Caffeine (a) | After Caffeine (c) |\n",
    "| --- | --- | --- |\n",
    "| 1 | [SRX2785036](https://www.ncbi.nlm.nih.gov/sra/SRX2785036[accn]) | [SRX2785039](https://www.ncbi.nlm.nih.gov/sra/SRX2785039[accn]) |\n",
    "| 2 | [SRX2785037](https://www.ncbi.nlm.nih.gov/sra/SRX2785037[accn]) | [SRX2785040](https://www.ncbi.nlm.nih.gov/sra/SRX2785040[accn]) |\n",
    "| 3 | [SRX2785038](https://www.ncbi.nlm.nih.gov/sra/SRX2785038[accn]) | [SRX2785041](https://www.ncbi.nlm.nih.gov/sra/SRX2785041[accn]) |\n",
    "\n",
    "SRA toolkit provides a very easy to use executable called `prefetch`, which we can call to help us download the sequencing data files. All we have to do is provide the SRA accession, a unique identifier provided by the [NCBI](https://www.ncbi.nlm.nih.gov/) for each sample."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!prefetch SRX2785037\n",
    "!prefetch SRX2785038\n",
    "!prefetch SRX2785039\n",
    "!prefetch SRX2785040\n",
    "!prefetch SRX2785041"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These files usually download automatically to a new location, the `~/ncbi/public/sra/` folder, so lets move all the files to our /data/seqData folder."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!mv -v  ~/ncbi/public/sra/* $WORKDIR/seqData"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using the same procedure as above, let's generate metrics to measure the level of gene expression in each experiment. Feel free to reference above if you need help, but **remember to change file names and locations**.\n",
    "\n",
    "1. convert to FASTQ\n",
    "2. clean the reads\n",
    "3. map the reads to the reference transcriptome sequence (we already generated the transcriptomeIndex, so no need to do it again!)\n",
    "4. quantify the aligned reads\n",
    "\n",
    "If you get stuck, here is the list of commands to run:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!fastq-dump --split-files -O $WORKDIR/FASTQ/2a $WORKDIR/seqDATA/SRR5511058.sra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!fastq-dump --split-files -O $WORKDIR/FASTQ/3a $WORKDIR/seqDATA/SRR5511059.sra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!fastq-dump --split-files -O $WORKDIR/FASTQ/1c $WORKDIR/seqDATA/SRR5511060.sra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!fastq-dump --split-files -O $WORKDIR/FASTQ/2c $WORKDIR/seqDATA/SRR5511061.sra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!fastq-dump --split-files -O $WORKDIR/FASTQ/3c $WORKDIR/seqDATA/SRR5511062.sra"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!java -jar ~/software/Trimmomatic-0.38/trimmomatic-0.38.jar PE -phred33 \\\n",
    "    $WORKDIR/FASTQ/2a/SRR5511058_1.fastq $WORKDIR/FASTQ/2a/SRR5511058_2.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/2a/SRR5511058.forward_paired.fastq $WORKDIR/cleanFASTQ/2a/SRR5511058.forward_unpaired.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/2a/SRR5511058.reverse_paired.fastq $WORKDIR/cleanFASTQ/2a/SRR5511058.reverse_unpaired.fastq \\\n",
    "    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!java -jar ~/software/Trimmomatic-0.38/trimmomatic-0.38.jar PE -phred33 \\\n",
    "    $WORKDIR/FASTQ/3a/SRR5511059_1.fastq $WORKDIR/FASTQ/3a/SRR5511059_2.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/3a/SRR5511059.forward_paired.fastq $WORKDIR/cleanFASTQ/3a/SRR5511059.forward_unpaired.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/3a/SRR5511059.reverse_paired.fastq $WORKDIR/cleanFASTQ/3a/SRR5511059.reverse_unpaired.fastq \\\n",
    "    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!java -jar ~/software/Trimmomatic-0.38/trimmomatic-0.38.jar PE -phred33 \\\n",
    "    $WORKDIR/FASTQ/1c/SRR5511060_1.fastq $WORKDIR/FASTQ/1c/SRR5511060_2.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/1c/SRR5511060.forward_paired.fastq $WORKDIR/cleanFASTQ/1c/SRR5511060.forward_unpaired.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/1c/SRR5511060.reverse_paired.fastq $WORKDIR/cleanFASTQ/1c/SRR5511060.reverse_unpaired.fastq \\\n",
    "    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!java -jar ~/software/Trimmomatic-0.38/trimmomatic-0.38.jar PE -phred33 \\\n",
    "    $WORKDIR/FASTQ/2c/SRR5511061_1.fastq $WORKDIR/FASTQ/2c/SRR5511061_2.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/2c/SRR5511061.forward_paired.fastq $WORKDIR/cleanFASTQ/2c/SRR5511061.forward_unpaired.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/2c/SRR5511061.reverse_paired.fastq $WORKDIR/cleanFASTQ/2c/SRR5511061.reverse_unpaired.fastq \\\n",
    "    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!java -jar ~/software/Trimmomatic-0.38/trimmomatic-0.38.jar PE -phred33 \\\n",
    "    $WORKDIR/FASTQ/3c/SRR5511062_1.fastq $WORKDIR/FASTQ/3c/SRR5511062_2.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/3c/SRR5511062.forward_paired.fastq $WORKDIR/cleanFASTQ/3c/SRR5511062.forward_unpaired.fastq \\\n",
    "    $WORKDIR/cleanFASTQ/3c/SRR5511062.reverse_paired.fastq $WORKDIR/cleanFASTQ/3c/SRR5511062.reverse_unpaired.fastq \\\n",
    "    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!kallisto quant -i $WORKDIR/kallisto/transcripts.idx -o $WORKDIR/kallisto/2a \\\n",
    "    $WORKDIR/cleanFASTQ/2a/SRR5511058.forward_paired.fastq $WORKDIR/cleanFASTQ/2a/SRR5511058.reverse_paired.fastq\n",
    "\n",
    "!kallisto quant -i $WORKDIR/kallisto/transcripts.idx -o $WORKDIR/kallisto/3a \\\n",
    "    $WORKDIR/cleanFASTQ/3a/SRR5511059.forward_paired.fastq $WORKDIR/cleanFASTQ/3a/SRR5511059.reverse_paired.fastq\n",
    "\n",
    "!kallisto quant -i $WORKDIR/kallisto/transcripts.idx -o $WORKDIR/kallisto/1c \\\n",
    "    $WORKDIR/cleanFASTQ/1c/SRR5511060.forward_paired.fastq $WORKDIR/cleanFASTQ/1c/SRR5511060.reverse_paired.fastq\n",
    "\n",
    "!kallisto quant -i $WORKDIR/kallisto/transcripts.idx -o $WORKDIR/kallisto/2c \\\n",
    "    $WORKDIR/cleanFASTQ/2c/SRR5511061.forward_paired.fastq $WORKDIR/cleanFASTQ/2c/SRR5511061.reverse_paired.fastq\n",
    "\n",
    "!kallisto quant -i $WORKDIR/kallisto/transcripts.idx -o $WORKDIR/kallisto/3c \\\n",
    "    $WORKDIR/cleanFASTQ/3c/SRR5511062.forward_paired.fastq $WORKDIR/cleanFASTQ/3c/SRR5511062.reverse_paired.fastq"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now have the gene expression in 6 different samples in terms of TPM. A common first step is to validate our data through visualization. We can imagine each sample as a single data point on our graph, where we have n-dimensions for the number of transcripts in yeast. We can expect the \"before-caffeine samples\" to form one cluster and the \"after-caffeine experiments\" to form another cluster. However, how do we transform our high-dimensional transcript expression data into 2-dimensional data?\n",
    "\n",
    "We will explore two popular algorithms for dimensionality reduction: [t-SNE](https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding)(t-distributed stochastic neighbor embedding) and [PCA](https://en.wikipedia.org/wiki/Principal_component_analysis)(principal component analysis). \n",
    "\n",
    "We will use Python and use several different libraries. Downloading python libraries is very easy using the [pip](https://pip.pypa.io/en/stable/quickstart/) command."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import dependencies\n",
    "import pandas as pd\n",
    "from sklearn.manifold import TSNE\n",
    "from sklearn.decomposition import PCA\n",
    "import time\n",
    "import math\n",
    "from plotnine import *\n",
    "import numpy as np\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>target_id</th>\n",
       "      <th>VIMSS6580644</th>\n",
       "      <th>VIMSS6580645</th>\n",
       "      <th>VIMSS6580646</th>\n",
       "      <th>VIMSS6580647</th>\n",
       "      <th>VIMSS6580648</th>\n",
       "      <th>VIMSS6580649</th>\n",
       "      <th>VIMSS6580651</th>\n",
       "      <th>VIMSS6580653</th>\n",
       "      <th>VIMSS6580656</th>\n",
       "      <th>VIMSS6580657</th>\n",
       "      <th>...</th>\n",
       "      <th>VIMSS6586892</th>\n",
       "      <th>VIMSS6586893</th>\n",
       "      <th>VIMSS6586898</th>\n",
       "      <th>VIMSS6580614</th>\n",
       "      <th>VIMSS6580621</th>\n",
       "      <th>VIMSS6580622</th>\n",
       "      <th>VIMSS6580627</th>\n",
       "      <th>VIMSS6580628</th>\n",
       "      <th>VIMSS6580629</th>\n",
       "      <th>VIMSS6580612</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>a1</th>\n",
       "      <td>2.001253</td>\n",
       "      <td>2.314166</td>\n",
       "      <td>1.591324</td>\n",
       "      <td>1.138631</td>\n",
       "      <td>1.495095</td>\n",
       "      <td>1.540466</td>\n",
       "      <td>1.596420</td>\n",
       "      <td>1.880289</td>\n",
       "      <td>1.723808</td>\n",
       "      <td>2.099069</td>\n",
       "      <td>...</td>\n",
       "      <td>1.387208</td>\n",
       "      <td>1.639635</td>\n",
       "      <td>1.821620</td>\n",
       "      <td>2.330458</td>\n",
       "      <td>1.653391</td>\n",
       "      <td>1.996965</td>\n",
       "      <td>1.362009</td>\n",
       "      <td>1.409023</td>\n",
       "      <td>1.801380</td>\n",
       "      <td>1.902422</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>a2</th>\n",
       "      <td>1.871936</td>\n",
       "      <td>2.351112</td>\n",
       "      <td>1.458062</td>\n",
       "      <td>1.090011</td>\n",
       "      <td>1.459961</td>\n",
       "      <td>1.393377</td>\n",
       "      <td>1.563595</td>\n",
       "      <td>1.805389</td>\n",
       "      <td>1.777433</td>\n",
       "      <td>2.074169</td>\n",
       "      <td>...</td>\n",
       "      <td>1.261981</td>\n",
       "      <td>1.582633</td>\n",
       "      <td>1.857878</td>\n",
       "      <td>2.350058</td>\n",
       "      <td>1.817173</td>\n",
       "      <td>1.938477</td>\n",
       "      <td>1.001145</td>\n",
       "      <td>1.465667</td>\n",
       "      <td>1.930689</td>\n",
       "      <td>1.939737</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>a3</th>\n",
       "      <td>1.955481</td>\n",
       "      <td>2.315626</td>\n",
       "      <td>1.437268</td>\n",
       "      <td>1.211801</td>\n",
       "      <td>1.492533</td>\n",
       "      <td>1.504729</td>\n",
       "      <td>1.595578</td>\n",
       "      <td>1.846698</td>\n",
       "      <td>1.660973</td>\n",
       "      <td>2.140822</td>\n",
       "      <td>...</td>\n",
       "      <td>1.275938</td>\n",
       "      <td>1.631825</td>\n",
       "      <td>1.801223</td>\n",
       "      <td>2.336622</td>\n",
       "      <td>1.524715</td>\n",
       "      <td>2.082688</td>\n",
       "      <td>1.119444</td>\n",
       "      <td>1.549779</td>\n",
       "      <td>1.980582</td>\n",
       "      <td>1.883486</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c1</th>\n",
       "      <td>1.767758</td>\n",
       "      <td>1.876159</td>\n",
       "      <td>1.507000</td>\n",
       "      <td>1.136264</td>\n",
       "      <td>1.357550</td>\n",
       "      <td>1.263331</td>\n",
       "      <td>1.548698</td>\n",
       "      <td>1.813714</td>\n",
       "      <td>1.963342</td>\n",
       "      <td>2.113850</td>\n",
       "      <td>...</td>\n",
       "      <td>1.284956</td>\n",
       "      <td>1.490526</td>\n",
       "      <td>1.831983</td>\n",
       "      <td>2.325329</td>\n",
       "      <td>1.676088</td>\n",
       "      <td>1.928189</td>\n",
       "      <td>1.233260</td>\n",
       "      <td>1.569479</td>\n",
       "      <td>1.906061</td>\n",
       "      <td>1.853782</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c2</th>\n",
       "      <td>1.921927</td>\n",
       "      <td>2.002891</td>\n",
       "      <td>1.441230</td>\n",
       "      <td>1.141130</td>\n",
       "      <td>1.332368</td>\n",
       "      <td>1.349048</td>\n",
       "      <td>1.579315</td>\n",
       "      <td>1.799521</td>\n",
       "      <td>1.921784</td>\n",
       "      <td>2.192283</td>\n",
       "      <td>...</td>\n",
       "      <td>1.253885</td>\n",
       "      <td>1.496107</td>\n",
       "      <td>1.850982</td>\n",
       "      <td>2.337631</td>\n",
       "      <td>1.754243</td>\n",
       "      <td>1.941409</td>\n",
       "      <td>1.286061</td>\n",
       "      <td>1.776610</td>\n",
       "      <td>1.915668</td>\n",
       "      <td>1.832984</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>c3</th>\n",
       "      <td>1.895282</td>\n",
       "      <td>1.935375</td>\n",
       "      <td>1.514434</td>\n",
       "      <td>1.191018</td>\n",
       "      <td>1.364104</td>\n",
       "      <td>1.251660</td>\n",
       "      <td>1.655679</td>\n",
       "      <td>1.814967</td>\n",
       "      <td>1.772066</td>\n",
       "      <td>2.164014</td>\n",
       "      <td>...</td>\n",
       "      <td>1.322463</td>\n",
       "      <td>1.495933</td>\n",
       "      <td>1.912814</td>\n",
       "      <td>2.340575</td>\n",
       "      <td>1.693179</td>\n",
       "      <td>1.943146</td>\n",
       "      <td>1.423976</td>\n",
       "      <td>1.710578</td>\n",
       "      <td>1.997486</td>\n",
       "      <td>1.607802</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>6 rows × 4305 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "target_id  VIMSS6580644  VIMSS6580645  VIMSS6580646  VIMSS6580647  \\\n",
       "a1             2.001253      2.314166      1.591324      1.138631   \n",
       "a2             1.871936      2.351112      1.458062      1.090011   \n",
       "a3             1.955481      2.315626      1.437268      1.211801   \n",
       "c1             1.767758      1.876159      1.507000      1.136264   \n",
       "c2             1.921927      2.002891      1.441230      1.141130   \n",
       "c3             1.895282      1.935375      1.514434      1.191018   \n",
       "\n",
       "target_id  VIMSS6580648  VIMSS6580649  VIMSS6580651  VIMSS6580653  \\\n",
       "a1             1.495095      1.540466      1.596420      1.880289   \n",
       "a2             1.459961      1.393377      1.563595      1.805389   \n",
       "a3             1.492533      1.504729      1.595578      1.846698   \n",
       "c1             1.357550      1.263331      1.548698      1.813714   \n",
       "c2             1.332368      1.349048      1.579315      1.799521   \n",
       "c3             1.364104      1.251660      1.655679      1.814967   \n",
       "\n",
       "target_id  VIMSS6580656  VIMSS6580657      ...       VIMSS6586892  \\\n",
       "a1             1.723808      2.099069      ...           1.387208   \n",
       "a2             1.777433      2.074169      ...           1.261981   \n",
       "a3             1.660973      2.140822      ...           1.275938   \n",
       "c1             1.963342      2.113850      ...           1.284956   \n",
       "c2             1.921784      2.192283      ...           1.253885   \n",
       "c3             1.772066      2.164014      ...           1.322463   \n",
       "\n",
       "target_id  VIMSS6586893  VIMSS6586898  VIMSS6580614  VIMSS6580621  \\\n",
       "a1             1.639635      1.821620      2.330458      1.653391   \n",
       "a2             1.582633      1.857878      2.350058      1.817173   \n",
       "a3             1.631825      1.801223      2.336622      1.524715   \n",
       "c1             1.490526      1.831983      2.325329      1.676088   \n",
       "c2             1.496107      1.850982      2.337631      1.754243   \n",
       "c3             1.495933      1.912814      2.340575      1.693179   \n",
       "\n",
       "target_id  VIMSS6580622  VIMSS6580627  VIMSS6580628  VIMSS6580629  \\\n",
       "a1             1.996965      1.362009      1.409023      1.801380   \n",
       "a2             1.938477      1.001145      1.465667      1.930689   \n",
       "a3             2.082688      1.119444      1.549779      1.980582   \n",
       "c1             1.928189      1.233260      1.569479      1.906061   \n",
       "c2             1.941409      1.286061      1.776610      1.915668   \n",
       "c3             1.943146      1.423976      1.710578      1.997486   \n",
       "\n",
       "target_id  VIMSS6580612  \n",
       "a1             1.902422  \n",
       "a2             1.939737  \n",
       "a3             1.883486  \n",
       "c1             1.853782  \n",
       "c2             1.832984  \n",
       "c3             1.607802  \n",
       "\n",
       "[6 rows x 4305 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#import data\n",
    "a1 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/1a/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "a2 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/2a/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "a3 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/3a/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "c1 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/1c/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "c2 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/2c/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "c3 = pd.read_csv(os.environ['WORKDIR'] + '/kallisto/3c/abundance.tsv', sep='\\t', header=0, usecols=['target_id', 'tpm'])\n",
    "\n",
    "#parse each sample's data\n",
    "a1.rename(columns = {'tpm':'a1'}, inplace = True)\n",
    "a2.rename(columns = {'tpm':'a2'}, inplace = True)\n",
    "a3.rename(columns = {'tpm':'a3'}, inplace = True)\n",
    "c1.rename(columns = {'tpm':'c1'}, inplace = True)\n",
    "c2.rename(columns = {'tpm':'c2'}, inplace = True)\n",
    "c3.rename(columns = {'tpm':'c3'}, inplace = True)\n",
    "a2 = a2.drop('target_id', axis=1)\n",
    "a3 = a3.drop('target_id', axis=1)\n",
    "c1 = c1.drop('target_id', axis=1)\n",
    "c2 = c2.drop('target_id', axis=1)\n",
    "c3 = c3.drop('target_id', axis=1)\n",
    "\n",
    "#combine samples into one table, where each row represents a sample and columns represent transcript\n",
    "dataframe = pd.concat([a1,a2,a3,c1,c2,c3], axis=1)\n",
    "dataframe = dataframe.T\n",
    "dataframe.columns = dataframe.iloc[0]\n",
    "dataframe = dataframe.drop('target_id')\n",
    "\n",
    "#clear memory, don't need references anymore\n",
    "del a1,a2,a3,c1,c2,c3\n",
    "\n",
    "#clean data by removing possible noise\n",
    "#filter out columns where any value is less than 10 (lowly expressed)\n",
    "min_threshold = 10\n",
    "for column in dataframe:\n",
    "    for value in dataframe[column]:\n",
    "        if value < min_threshold:\n",
    "            dataframe.drop(column, axis=1, inplace=True)\n",
    "            break\n",
    "\n",
    "#log base 10 all values in dataframe\n",
    "dataframe.applymap(lambda x: math.log(x, 10))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Explained variation per principal component [0.60461493 0.20639627]\n",
      "[t-SNE] Computing 5 nearest neighbors...\n",
      "[t-SNE] Indexed 6 samples in 0.000s...\n",
      "[t-SNE] Computed neighbors for 6 samples in 0.007s...\n",
      "[t-SNE] Computed conditional probabilities for sample 6 / 6\n",
      "[t-SNE] Mean sigma: 7615.190083\n",
      "[t-SNE] KL divergence after 250 iterations with early exaggeration: 55.712395\n",
      "[t-SNE] KL divergence after 1550 iterations: 0.008925\n",
      "t-SNE done! Time elapsed: 0.17518281936645508 seconds\n"
     ]
    }
   ],
   "source": [
    "# perform PCA calculations\n",
    "pca = PCA(n_components=2)\n",
    "pca_result = pca.fit_transform(dataframe.values)\n",
    "\n",
    "print ('Explained variation per principal component {}'.format(pca.explained_variance_ratio_))\n",
    "\n",
    "# perform t-SNE calculations\n",
    "time_start = time.time()\n",
    "tsne = TSNE(n_components=2, verbose=1, perplexity=2, n_iter=5000, learning_rate = 10)\n",
    "tsne_results = tsne.fit_transform(dataframe.values)\n",
    "\n",
    "print ('t-SNE done! Time elapsed: {} seconds'.format(time.time()-time_start))\n",
    "\n",
    "dataframe['PC1'] = pca_result[:,0]\n",
    "dataframe['PC2'] = pca_result[:,1] \n",
    "\n",
    "dataframe['tSNE1'] = tsne_results[:,0]\n",
    "dataframe['tSNE2'] = tsne_results[:,1]\n",
    "\n",
    "# add label column to table to identify each sample\n",
    "dataframe['label'] = pd.Series(['before', 'before', 'before', 'after', 'after', 'after'], index = dataframe.index)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "PCA is a deterministic algorithm that looks for most variant variables across samples. It does this by trying to find a best-fit linear line to the data, which is called the first principal component. The first principal component always accounts for the most variation in the data. Subsequent principal components are the best fitting lines perpendicular to previous principal components. The final points are simply the position of the respective point after being projected onto the principal components. This [video](https://www.youtube.com/watch?v=FgakZw6K1QQ) by StatQuest is very helpful in understanding the step-by-steps of PCA and gaining intuition about how it works. From our example, we can see that 81.1% of the variation in transcript expression in our samples are accounted for by the first and second principal components.\n",
    "\n",
    "t-SNE is a non-deterministic algorithm for dimensionality reduction. Each point is initiated in a position in the wanted lower-dimension. At each step, t-SNE will measure the distances of each point with each other point, and compare these with the similarity values from the original data. The ultimate goal is to make the similarity values between points on the lower dimension to match with the values from the original dimension. This [video](https://www.youtube.com/watch?v=NEaUSP4YerM) by StatQuest helps explain t-SNE. Keep in mind that t-SNE is non-deterministic, meaning that each run will behave differently and result in different results.\n",
    "\n",
    "Note: to re-run t-SNE calculations, you have to first reload the data in the previous code cell before running code cell with t-SNE calculations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<ggplot: (7548741994)>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#graph PCA \n",
    "chart = ggplot(dataframe, aes(x='PC1', y='PC2', color='label') ) \\\n",
    "        + geom_point(size=3,alpha=1) \\\n",
    "        + ggtitle(\"First and second principal component\")\n",
    "chart"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAp0AAAHICAYAAAAMflHnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAAPYQAAD2EBqD+naQAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAIABJREFUeJzs3XmczXX///Hn55w5c2Yfw8xYY4aiESFZspOkqyRJEVcowsXVIq7U5bJrVWkTrdr4/i4p0dWqUNIm1ZWIyppIwzD7ds7794drTp1mXz5zDI/77TY3zufzPp/P65x5mXn6LO9jGWOMAAAAABs5Al0AAAAATn2ETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROoAS9evVSQkJCqctOFaNGjZJlWYEuI6Asy9KoUaMCXUaJ9uzZI8uyNGvWrFLHLl26VJZlaf369bbXVZr169fLsiwtXbq0xGXlNWvWLFmWpT179lS6RgD2IXSiwvbs2aNZs2bp66+/LnJ9bm6ulixZos6dOys+Pl4hISE644wz1KdPH82YMUM5OTm+sQW/GC3L0muvvVbk9izLUt++ff2W9erVy/e8or5O9vAAwB5ff/21Zs2aRRAFTiJBgS4ANdeePXs0e/ZsJSQkqG3btn7rPB6P+vXrpw0bNqhHjx6aOnWqYmJitG/fPn3++ee66667dNNNN8ntdhfa7rRp0zRgwAAFBZWtPR0Oh55//vki1zVr1qz8L6wU7777rowxVb7dk8FTTz2lxYsXB7oMnEZ69OihrKwsuVyuCm9j+vTpmjZtmt/Pk6+//lqzZ88+pc9MADUNoRO2WL16tTZs2KArr7xSK1euLLT+119/VXR0dKHlHTp00BdffKEnn3xSf/vb38q0L8uyNGLEiErXXFbBwcHVtq/q5nK5KvXLH0XLzs5WUFBQmf8jdTpxOBwKCQmp1DZ4b4GagdPrqJBZs2apd+/ekqTRo0f7Tmf36tVLkrRjxw5J8o35s7p16xYZbsaPH6+mTZtq9uzZSk9Pt6f4Ivz4448aNGiQoqOjFRkZqX79+umbb74pcmxJ13nu27dPQ4YMUUxMjKKjozV48GAdPnxYkvTss8+qVatWCgkJUWJiop555pkit79u3TpdcskliomJkdvtVlJSku699155PJ4i9/nLL7/o2muvVe3atRUaGqqePXvqyy+/9BtrjNGjjz6qdu3aKTo6WuHh4UpISNCwYcP066+/+sYVd03n999/r6FDh6pu3bpyu91q2rSppkyZotTUVL9xBZdJfPDBB1qwYIHOOuss3/iHHnqo0HY//fRTDRgwQA0aNJDb7VZ8fLx69uyp1atXF/ne/Fl6erpmzZqlVq1aKTQ0VDExMerQoYMee+wxv3HHjh3T5MmTlZiYKLfbrbp162rYsGH64YcfyrQfSXr55ZfVqVMnRUREKDw8XJ07d9b//d//FRpX8H3Zu3evhg4dqtjYWIWGhurnn3/2jVm5cqV69uypqKgohYaGql27dnr66aeL3O/y5cvVpk0bhYSEqGHDhpo8ebIyMzPLXHeB/Px8zZkzx/cenH322Vq0aJHfmIkTJ8qyLG3btq3Q87Ozs1W7dm1169atTPt75JFH1KJFC7ndbiUmJmru3LnKz88vNK64azpTU1M1adIk1atXT6GhoTrvvPO0YsWKIq/f/POyUaNGafTo0ZJO/Azichvg5MB/DVEhV155pfLy8nTXXXfpxhtvVPfu3SWdCJOS1LRpU0nSihUrNHz4cMXExJRpu8HBwZo/f76GDRum++67T3PmzCnT85KTk4tcHhUVVeqRyf3796tLly46duyYxo8fr6SkJH3yySfq1auX6tSpU6b9S1JGRoZ69eqlrl276u6779b333+vxx57TIcOHdKgQYP0+OOPa+zYsYqMjNRTTz2lMWPG6Oyzz1bXrl1923j22Wc1ZswYtWvXTtOmTVOtWrX08ccf64477tBXX31VKORkZGSoR48eOv/88zVv3jz9+uuveuihh9S/f3/t2rVLkZGRkqS77rpL06dP16WXXqoxY8bI5XJp//79euutt3To0CHf960oX3/9tXr06CGPx6MJEyaoadOm2rhxox544AGtXbtWmzZtUlhYmN9z7rzzTqWlpen6669XRESEXnjhBU2ePFn169fX0KFDJUk7d+5U3759FRcXp4kTJ6p+/fr67bfftHnzZn366ae6/PLLS3y/jx8/ru7du+vbb7/VwIEDdcMNN8jpdGrr1q169dVXNWnSJElSWlqaunbtqm3btmn48OHq0qWLfvrpJy1atEhvv/22Pv74Y7Vs2bLEfc2YMUNz587Vueeeq3/961+SpJdeeknDhg3Trl27dOedd/qNT09PV/fu3XX++edr9uzZSktLU0REhKQTAWn27Nnq3bu3Zs6cqdDQUL3zzjsaO3asfvzxR91zzz2+7SxevFgTJkxQ8+bNNXPmTLlcLr388sv68MMPS6y3KLfffrtSU1M1duxYud1uLVu2TBMnTtThw4d9NySNHz9eixYt0tNPP60HH3zQ7/krV65USkqKxo4dW+q+pk2bpnvvvVfnn3++7r77bmVlZenZZ5/V66+/XqZa8/Pz1b9/f33yyScaMmSIevXqpf379+uGG25QixYtSn3+uHHj5Ha79eSTT+rOO+9UUlKSJHsutwFQDgaooHXr1hlJ5rnnniu0Licnx5x33nlGkgkLCzN9+/Y1//znP82aNWtMRkZGofHPPfeckWRefPFF4/V6TYcOHUx4eLj55ZdffGMkmQsvvNDveT179jSSiv1avnx5qa/jr3/9q5FkXnvtNb/ld999t5FkmjRpUmifRS2TZO6++26/5bfeequRZBo2bGiOHTvmW37o0CHjdrvN0KFDfcsOHjxoQkJCzBVXXGG8Xq/fdhYsWGAkmfXr1xfa51133eU3dvny5UaSWbJkiW9Zu3btTFJSUqnvxciRI82ffyx0797dWJZlPvnkE7/ls2fPNpLM3LlzfcsKvo/nnnuuyc7O9i1PT083derUMRdccIFv2cMPP2wkmc8++6zUuooyceJEI8ksXLiw0DqPx+P7+7/+9S8jydx7771+Y9avX19kT0kyI0eO9D3euXOncTgc5rzzzjOZmZl+r6lVq1bG6XSa3bt3+5YXfF9uv/32QnVt2bLFWJZlbrrppkLrJk2aZBwOh/npp5+MMcYcO3bMREREmMaNG/v1TmZmpmnbtq2RZGbOnFn0m/MHBd+TRo0amZSUFN/yrKws0759+0L1d+nSxcTGxpqcnBy/7fTs2dNER0cX+e/3j3744QfjcDhMx44d/XrgyJEjpn79+oV+ZhT1c+TJJ580kszUqVP9tr1582ZjWZaR5FfzzJkzCy0reN3r1q0rsV4A1YfT67BFcHCwNmzYoLvuuktnnnmm3n//fc2fP18DBgxQvXr19MADDxT7XMuydP/99ysjI0MzZ84sdV8Oh0PvvfdekV/Fnd4v4PV6tWrVKrVs2VJXXHGF37pbbrnFd3SqLBwOh2655Ra/ZQVHgEeOHOl3DWvdunXVokULv9O7r7zyirKzszVmzBgdOXJEycnJvq/LLrtMkvTOO+8U2uett97qt+yiiy6SdOJIYoGYmBj9/PPP2rBhQ5lfjyT99ttv+uijj3TxxRerc+fOfuumTJmi8PDwIq/ZnTRpkt9NHeHh4brgggsK1SRJr732mrKysspVl9fr1bJly9S0aVP9/e9/L7Te4fj9R9vKlSsVHR2tm2++2W9Mz5491bt3b33wwQdKSUkpdl+rVq2S1+vVP/7xD4WGhvq9pqlTp8rj8RR5BO/2228vtOzll1+WMUY33HCD3/c3OTlZl19+ubxer9auXSvpxA1r6enpmjRpkl/vhIaGasqUKSW8O0X729/+plq1avkeh4SEaPLkyYXqHz9+vJKTk/1mkdi5c6c2bNigESNGFDqq/WevvfaavF6vpkyZ4tcDtWvX1sSJE8tUa8G+//wetm/fXv369SvTNgCcfAidsE1ERITuuOMOffPNN0pNTdVHH32k22+/3fcLafny5cU+t2fPnrrsssv07LPPFnl92R8VTKVU1FdJp40l6fDhw0pLSyvy9GpISEi5Tsc1aNCg0A0RBcGq4HKDP687cuSI7/H27dslSZdddpni4uL8vs4++2xJ8rv+srh9FlwS8Mdt33PPPYqMjFSvXr1Ur149DRkyRIsXL9bx48dLfE27du2SJJ1zzjmF1oWFhalZs2b66aefCq0r6vXWqVPHr6ahQ4fq0ksv1T333KOYmBj16NFD06dP19atW0usSTpxOUVKSoratGnjFzCLew1nnnlmkTMltG7dWsYY7d69u8TnS0W/B61bt5akQu9BXFxckZeUFHyP27RpU+h7XBCmCr7HBdssqjeLqqU0RW2nYNmPP/7oWzZkyBDVqVPH7xrTgr+X5dR6VdS9a9cuxcbGFnl5S8G/BQA1D9d0olpERESoW7du6tatm3r37q3+/fvrmWee0bBhw4p9zr333qu33npLt99+u9asWVON1f7OlGNqJKfTWe51f9y+1+uVdOIXfJMmTYoc36BBgzLv84/b7tChg3744Qe99957WrdunTZs2KBXXnlFM2bM0EcffVTsdXIF2yjvhPEl1VXA5XLpjTfe0JYtW/TOO+9o48aNeuihh3TXXXfp/vvv12233Vam11YWlZnwviLvQXFHAwu+x2+88UaRIVgqOrAXV1NV+eNrCwkJ0ciRI/XQQw9p9+7datSokZ5//nl17NhRbdq0qdR+ylP36f4hBcCpiNCJCqvoL4UuXbpIkg4cOFDiuJYtW2r06NF6+umny31auKzi4+MVGRlZ7N26u3btKtfNRJXRvHlzSSeOgP55EvyqEBYWpoEDB2rgwIGSTgSfAQMG6L777iv2TvqCI71FHX3MysryHUWsjPPOO0/nnXeeJOno0aPq3Lmz/vnPf+rmm28udhqcgiOJ33zzjbxeb4lHO5s1a6YffvhBOTk5hYLe1q1bZVmWEhMTi31+wevbunVroSN1Be9LWY+IN2/eXG+//bbq16/ve80l1S1J27Zt06WXXuq3rrSj/0XZtm2b73v/5+38uf5x48bpwQcf1DPPPKO2bdvq8OHDmj9/fpn288e6//x+lbXupk2baseOHTpy5Eihf38FR4tLQ2gFTj6cXkeFFVzvePTo0ULrvvnmm2JDZcH1WqXdMSxJc+bMUVhYmP7xj39UotLiORwODRw4UNu2bdOqVav81i1cuLBap226+uqrFRISolmzZhW536ysLKWlpVVo27/99luhZe3bt5dU9PevQFxcnLp376533nlHn3/+ud+6Bx54QOnp6Ro8eHCFaipqxoHatWsrMTFROTk5ysjIKPa5DodD1157rXbt2qVHH3200PqCI4rSiZkWjh8/Xmjcxo0b9cEHH6hPnz4lzq5wxRVXyOFwaMGCBcrOzvYtz8zM1P333y+n01kozBXnuuuukyTdcccdysvLK7T++PHjvk/q6tevn8LDw/XYY4/5XQaRnZ2tBQsWlGl/f7Ro0SIdO3bM9zgnJ0cPPvignE5noZkCmjdvrt69e2vp0qVavHixIiMjfbMOlOaKK66QZVlasGCBcnNzfcuPHj2qxx9/vMzbkE6c7fijL7/8Uu+9916ZtlHSzycAgcGRTlRYy5YtFRkZqUWLFiksLEy1atVSfHy8+vTpo3Xr1mnq1Knq27evunXrpgYNGigtLU2ffvqpXnnlFUVFRZXpJqH69etr8uTJmjdvXrFjjDF66aWXilwXFxeniy++uMR9zJs3T2+//bauueYajR8/XmeffbY+/fRTrV69Ws2aNStybkE7NGzYUEuWLNH111+vFi1aaOTIkWratKmOHj2q7du367XXXtOqVat8c6GWR1JSkjp16qROnTqpYcOGOnLkiG9OzZEjR5b43EceeUQ9evRQnz59/KZMWrZsmdq0aaPJkydX6PUWvO+XXXaZmjZtKofDofXr1+vdd9/1zZla2vPXr1+vW265RevWrVOvXr3kcrn03XffaceOHXr//fclSVOnTtXKlSs1depUffPNN35TJkVHR+uRRx4pcT9nnnmm/vnPf2ru3Lnq3Lmzhg8f7uu5b7/9VvPnzy/zJ960b99e8+bN0/Tp09WqVSsNGzZMjRo10uHDh/Xf//5Xq1ev1rZt25SQkKDo6Gjde++9mjRpkjp27KjRo0fL5XLppZdeKtPlC38WHx+vDh066Prrr1dwcLCWLVumLVu2aPr06UUe6R0/fryuueYaHThwQDfeeGOZb6o766yzdNttt2nBggXq2rWrhg0bpuzsbD3zzDNq0KCBDh48WOo2Ro8erWeeeUb333+/9uzZ45syadGiRWrfvr02b95c6pHMDh06yOFwaP78+UpJSVF4eLgSExPVqVOnMr0OADYIzE3zOFX85z//Me3atTNut9tIMj179jTGGLNnzx5z1113mQsvvNCcccYZxu12m5CQENO8eXMzYcIE37QwBf44ZdKfpaammvj4+ApNmdSpU6cyvY4dO3aYgQMHmsjISBMREWEuuugi89VXXxU7PVJZlhlT8rRSxT3n008/NVdddZWpW7eucblcpm7duuaCCy4wc+fONUeOHCn1+cYUnvbn7rvvNj179jTx8fHG5XKZ+vXrm0suucSsXbvW73lFTZlkjDHbtm0zV199tYmNjTUul8s0adLETJ482W8qH2NKnqbmz9tet26dueaaa0xCQoIJDQ01UVFRpk2bNua+++4zWVlZRb6uPzt+/Li58847TfPmzU1wcLCpVauW6dixo1m0aJHfuKNHj5pbbrnFNGnSxLhcLhMbG2uGDh1qduzYUWibf37vCrz44oumY8eOJjQ01ISGhppOnTqZZcuWFRpX0velwNtvv23+8pe/mDp16hiXy2UaNGhgevfubR544IFCr/2ll14yrVu3NsHBwaZ+/frm1ltvNd999125p0x67733zKxZs0yTJk1McHCwad68uXnkkUeKfV5ubq6pW7eukWQ2b95c6n7+yOv1moceesiceeaZxuVymYSEBDNnzhzz3nvvlWnKJGOMSUlJMRMmTDDx8fHG7Xab8847z7z66qtm8uTJRpL59ddffWOLmjLJGGOWLl1qkpKSjMvlKvb7CqD6WMacoh8iDQCoMI/HoyZNmig+Pl5btmwJdDk+l156qTZs2KDU1NRSZy4AcHLhXywAoJAVK1bowIEDmjBhQkD2X9RHfW7evFlvv/22+vbtS+AEaiCOdAIAfNasWaN9+/Zp7ty5CgkJ0Y4dO4qd3slOI0eOVGpqqrp27arIyEht3bpVTz/9tJxOpz755BPfHKkAag5CJwDAJyEhQb/88ovatm2rxYsXlzq1k11efvllPf7449qxY4dSU1NVu3Zt9ejRQzNnzlSrVq0CUhOAyiF0AgAAwHZcFAMAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2Cwp0AXZITk62ZbuWZSk0NFRZWVlipqmKCQ4OVm5ubqDLqJHov8qj/yqO/qs8+q90sbGxgS4BNuJIZzk4HA6FhYXx8WuVEIhPNjlV0H+VR/9VHP1XefQfTnf89AAAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALDdKfnZ61XNm5am3M83ybNnt7JDQqQzmyuobXtZfKQZAABAmRA6S5H79RZlLn9ekiV58pUnSf/9StYbqxR+40QFndEkwBUCAACc/Di9XoL8fXuV+fJzkscjefJ/X+HxyGRnKX3Jo/JmpAeuQAAAgBqC0FmCnHXvSZZV9EpjpLx85X72SfUWBQAAUAMROothjFHetm8lr7f4QZ585X79ZfUVBQAAUEMROotjzInT6qXJy7W/FgAAgBqO0FkMy+GQIzau5EEOh5wNz6ieggAAAGowQmcJ3N17SU5n8QO8Xrm79ay2egAAAGoqQmcJgjt3kzOxWdHB03IouHtvBSU0rf7CAAAAahhCZwksp1MRY/6mkIsukRUe4VvuiItX6JBhCr38ygBWBwAAUHMwOXwprKAghfTtL3effrLS01Srdm2leo28Jd3VDgAAAD8c6Swjy+GQM6a2gmrFyCpu7k4AAAAUidAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROAAAA2I7QCQAAANsROgEAAGA7yxhjAl1EVUtNTZXb7a7y7VqWpeDgYOXm5uoUfNuqRVBQkPLz8wNdRo1E/1Ue/Vdx9F/l0X+ls+N3N04eQYEuwA65ubnKzc2t8u06nU4FBwcrIyNDHo+nyrd/OoiMjFRaWlqgy6iR6L/Ko/8qjv6rPPqvdITOUxun1wEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtgsKdAGAyctV/s4dMtlZctSuI2dCU1mWFeiyAABAFSJ0ImCMMcrZ8L6y33tLysuTHA4pP1+O2rEKHTJMrrNaBLpEAABQRTi9joDJfvdNZb+5WsrJkbxeKT9fkuQ9mqyMpx5X/k8/BLhCAABQVQidCAhvWqpy1r5zImwWOcCrzNdXVm9RAADANoROBETuV5tPnE4vgfeXn+U5dLCaKgIAAHYidCIgzLFjkjElD7IseY8fq56CAACArQidCAgrIlJylHKHujFyREZWT0EAAMBW3L2OgAhu117Zb75e4hhHXLwc9RtWU0U1gzFGudu/U9aH6+T55WdZLpdcbc+T+4LuctSKCXR5AAAUi9CJgHDE1FZw1x7K/eRjyespPMCyFDpgEPN1/oExRr8996TSPlpfsEBGUs76D5Tz4XpF3DhRQYnNAlkiAADF4vQ6AiZ04FUK7t7zxA1FziDJ5Trx99BQhY0YLVfL1oEu8aSSs+lDpX784YlrYf94PawnX8rLVfrTT8hkZwWuQAAASsCRTgSM5XAobMCVCunTT3lb/yuTlSlH7TpytWwtK4jW/CNjjLI+eE/yFHFUuEB+nnK//Fzurj2rrzAAAMqI3+wIOEd4hNydugS6jJOaST0u79EjJQ/yeJS383tCJwDgpMTpdaAmKG16qQLFTbYPAECAETqBGsCKipYVGVXyIKdTQc3Oqp6CAAAoJ0InUANYDodCe10oOZ0lDLIU3KFz9RUFAEA5EDqBGiKk54UKa9228MeHOp2Sw6Hw626QIzwiMMUBAFAKbiQCagjL6VT9m27Tr2vfVeaG9+U9fEhyBsl1bluF9OgjZ736gS4RAIBiETqBGsRyOOTu0ElB550f6FIAACgXTq8DAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACw3UkVOlNTUzV8+HBNmTLFt2zv3r2aMmWKrrrqKk2aNEnfffddACsEAABARZxUofPZZ59VQkKC73F+fr7mzZunCy64QMuXL9fgwYM1f/58paenB65IAAAAlNtJEzq//fZbHTp0SL179/ZblpOTo0GDBsnlcql3796qW7euNm3aFMBKAQAAUF5BgS5AkvLy8rRkyRJNmTJFP/30k2/5vn371KRJEzkcv2fjxMRE7du3z+/5ycnJSk5O9j12OByKi4ur8jqdTqffnyg/y7J4/yqI/qs8+q/i6L/Ko/9wujspQueKFSvUrl07JSQk+IXOrKwshYeH+40NDw9XZmam37KVK1fqqaee8j0eNWqUJk2aZFu9UVFRtm37dBAcHBzoEmo0+q9y6L/Kof8qh/7D6SzgofOXX37R+vXr9fDDDxdaFxoaWihgZmZmKjQ01G/Z4MGD1bNnT99jh8OhlJSUKq/V6XQqKipKqamp8ng8Vb7900F4eLgyMjICXUaNRP9VHv1XcfRf5dF/pYuJiQl0CbBRwEPn9u3bdeTIEY0ZM0bSiZuHcnNzNXz4cP3973/X3r175fV6fafYd+/erf79+/ttIzY2VrGxsb7HycnJtv5Q9Hg8/NCtIGMM710l0X8VR/9VHv1XcfQfTncBD53dunVTmzZtfI8//vhjrVu3TtOnT1etWrXkcrm0atUqDRgwQJs2bdKhQ4d0wQUXBLBiAAAAlFfAQ6fb7Zbb7fY9joiIUFBQkO/I5fTp0/XYY49p2bJlqlu3ru68805FRkYGqlwAAABUgGWMMYEuoqr98U72quR0OhUTE6OUlBROkVRQZGSk0tLSAl1GjUT/VR79V3H0X+XRf6X746VyOPWcNPN0AgAA4NRF6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA21nGGBPoIqpaamqq3G53lW/XsiwFBwcrNzdXp+DbVi2CgoKUn58f6DJqJPqv8ui/iqP/Ko/+K50dv7tx8ggKdAF2yM3NVW5ubpVv1+l0Kjg4WBkZGfJ4PFW+/dNBZGSk0tLSAl1GjUT/VR79V3H0X+XRf6UjdJ7aOL1T3ygKAAAgAElEQVQOAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAACctmbNmqWIiIhyPSchIUGTJk0K2P5rKkInAAAAbEfoBAAAgO0InQAAAJIyMjI0adIktWjRQmFhYUpISND48eN1/PjxIsfff//9atiwocLCwjRw4EAdPHjQb31OTo7uvPNONWnSRG63W0lJSVq2bFl1vJST0in5iUQAAADllZmZKY/Ho/nz5ysuLk779+/X/PnzNWjQIH3wwQd+Y1977TU1adJETzzxhFJSUjRt2jRdeeWV+uSTT3xjrr76am3cuFEzZ85UUlKS3nzzTY0YMUIxMTG65JJLqvvlBRyhEwAAQFJcXJyeeOIJ3+P8/HwlJiaqW7du2rlzp5o3b+5bl5aWpjfffFO1atWSJDVq1Eh9+/bVu+++q379+mndunVavXq13nnnHfXr10+SdNFFF+nAgQOaOXPmaRk6Ob0OAADwPy+++KLatWuniIgIuVwudevWTZK0c+dOv3G9e/f2BU5JuvDCCxUVFaVPP/1UkvTuu++qdu3a6tOnj/Lz831fF154ob766it5PJ7qe1EnCY50AgAA6MQp8+uuu0433nij5s+frzp16ujgwYMaNGiQsrOz/cbGx8cXen58fLzvus7k5GQdPXpULperyH0dPHhQjRo1qvoXcRIjdAIAAEhasWKF2rZtqyVLlviWbdiwocixhw8fLnJZ/fr1JUm1a9dWXFyc3nzzzSKfX1RoPdVVaejMy8vTwYMH1bhx46rcLAAAgO2ysrIUHBzst+zll18ucuy6det0/PhxRUdHS5Lef/99paamqlOnTpKkvn376r777lNwcLDOPfdcewuvIcp8TecTTzyhZs2aKTQ0VOedd56WL19eaMyWLVuUmJhYpQUCAABUh4suukiff/655syZo7Vr1+q2227T+++/X+TYyMhIXXLJJVq9erVeeOEFjRgxQh07dtTFF1/s29aAAQPUv39/LVy4UB988IHWrFmje+65R2PGjKnOl3XSKNORzhUrVmjixIkaNGiQRo8erQ8//FDDhw/X6tWr9fzzzxf6XwEAAEBNM27cOO3atUuPPfaYFixYoIsvvljLli1T586dC40dNGiQGjVqpPHjxyslJUV9+/b1Oy0vSa+88oruueceLVq0SHv37lV0dLRatWql0aNHV9dLOqlYxhhT2qCOHTuqS5cuWrhwoW/Z66+/rtGjR6tNmzZatWqVoqOj9dlnn6lLly4BvyMrOTnZlu06nU7FxMQoJSUl4K+xpoqMjFRaWlqgy6iR6L/Ko/8qjv6rPPqvdLGxsYEuATYq0+n1HTt26PLLL/dbNnDgQG3cuFE//fSTevToUWgWfgAAAKBAmUJneHh4kf87a9mypTZt2qS8vDx16dJFO3bsqPICAQAAUPOVKXSee+65euONN4pc16hRI23cuFFxcXEaO3ZslRYHAACAU0OZbiQaNGiQ5s+fryNHjqhOnTqF1teuXVvr1q3TlVdeqffee6/KiwQAAKhKBZ+zXtUiIyOrfJunijId6Rw3bpz27dtXZOAsEB4ernfeeUder7fKigMAAMCpgc9eBwAAgO3KFDr79evnd5OQMUZz5szRoUOH/MZ9/fXXatq0adVWCAAAgBqvTKFz7dq1On78uO+x1+vV7Nmz9csvv/iNy8nJ0d69e6u2QgAAANR4FT69XoY55QEAAFCKJ598UvXr11dERIR27doV6HJsU6a71wEAAE43Ji9POdu+lTc1VY6oKLlbtpblclXpPvLy8nTzzTfrww8/VIcOHSRJlmVp+/btOvvss6t0X4FG6AQAAPiTzE0fKf21FTK5OZLTKXk8soLdihg0RGFdulfZfg4dOqTs7Gy1bt26yraZl5cnVxWH46pQ5tPrO3bs0JYtW3xfkvT999/7Lfv+++9tKxQAAKA6ZG76SGn/96JMdpbk9Up5eZLXK5OdpbT/e1GZmz4q9zbvv/9+nXnmmYqMjFRSUpJWrlzpdzQzNjZW7du3V48ePSRJ7du3V0REhJ588klJ0hdffKHu3bsrJiZGSUlJevXVV33bHjVqlMaNG6fLL79cERERWrlyZRW8C1WvzEc6R40aVWjZiBEjZFmW77Exxu8xAABATWLy8pT+2gqpuHtXjFH6qhUK7dC5XKfaExMTtWHDBtWvX18rVqzQiBEj9MMPP+i7775TYmKikpOTFRISIunE6fUvv/zSF0gPHjyo/v3765lnntGAAQO0efNmXXLJJUpKSlJSUpIk6eWXX9aaNWv0+uuvKzs7u3Jvgk3KFDrXrVtndx0AAAABl7Pt2xOn1EtgcnKUs22rQtq0K/N2r7rqKt/fr7nmGt1999367LPP1L59+1Kf++KLL6pv37664oorJEmdOnXSoEGDtGLFCs2YMUOSdNlll6l3796SpNDQ0DLXVZ3KFDp79uxpdx0AAAAB501NPXENZ0mfsOh0ypt6vPj1RXjhhRf04IMPas+ePZKk9PR0JScnl+m5e/bs0euvv65atWr5luXn5+uvf/2r73Hjxo3LVU8gcCMRAADA/ziioqTSPpPd45EjKrrM29y7d6/GjBmjtWvXqmvXrnI6nWrXrl2Zp59s3Lixhg4dqqVLlxZft+Pk/5DJMoXOyy+/vMwbtCxLr7/+eoULAgAACBR3y9aygt0nbiIqhuV2y92yVZm3mZGRIUmKi4uTdOKo59atW4sdX7duXf3000++azpHjBih9u3ba82aNbrkkkvk9Xr11VdfKSoqyndNZ01QplicmpqqtLS0Er8OHDigN954Q2+88YbdNQMAANjCcrkUMWiIVNyN0ZaliCuGlOsmopYtW2rq1Knq2rWr6tatq6+//lpdunQpdvzs2bM1ZswY1apVS0899ZQaNWqk//znP1q4cKHq1q2r+vXr64477lBOTsnXnp5sLFPJjxZKS0vTI488ooceekgZGRkaO3asHnnkkaqqr0LKeo1EeTmdTsXExCglJUWe0g69o0iRkZFKS0sLdBk1Ev1XefRfxdF/lUf/lS42Nrba9pWZmVliL2du+kjpq1bI5Pxhnk63WxFXlDxPZ2RkpB3lnhIqfE1nWlqaHn74YS1cuFBZWVkaO3asbr/9dtWvX78q6wMAAKh2YV26K7RDZ+Vs2ypv6nE5oqLlbtmqyj+R6HRS7tCZmpqqhQsX6uGHH1Z2drbGjRunf/zjH6pXr54d9QEAAASE5XKVa1oklKzMofP48eO+sJmbm6sJEyZo6tSpio+Pt7M+AAAAnALKFDpnzpypRx55RB6PR3/72980ZcqUar3uAgAAADVbmULn3LlzZVmWOnXqpG3btun6668vdixTJgEAAODPyhQ6e/To4ftMde68AwAANZ1V3JRIsE2ZQuf69ettLgMAAKD6nKyfT34qO/k/MwkAAAA1XplCZ3Jysv773/8WWv7f//5XV111lc455xxdeOGFWrNmTZUXCAAAgJqvTKfX77jjDn355ZfasmWLb9nevXvVvXt3ZWZmqk2bNtq6dasGDRqkDz74QD169LCtYAAAgMoq7ROJKopPJCpemULnxx9/rBtuuMFv2UMPPaT09HS99dZb6tevn7KysnTRRRfp3nvvLVfozMvL0+LFi/XNN98oLS1NsbGxGjJkiHr16iXpRLh99NFHtWfPHtWrV08TJkzQOeecU/ZXCAAAgIAr0+n1AwcOqFWrVn7L1qxZo7Zt26pfv36STlyQ+/e//73I0/Al8Xg8ql27tubNm6fly5dr4sSJWrx4sb7//nvl5+dr3rx5uuCCC7R8+XINHjxY8+fPV3p6ern2AQAAgMAqU+i0LMtvaoFff/1Vu3fvVs+ePf3GNWzYUMnJyeUqICQkRMOHD1e9evXkcDjUsmVLJSUlafv27fr222+Vk5OjQYMGyeVyqXfv3qpbt642bdpUrn0AAAAgsMp0er1FixZau3at76jmG2+8IcuyfI8LHDx4UHFxcZUqKDs7Wz/++KMGDBigffv2qUmTJnI4fs/GiYmJ2rdvn99zkpOT/cKuw+GodB1FcTqdfn+i/CzL4v2rIPqv8ui/iqP/Ko/+w+muTKHzpptu0nXXXaeUlBTVq1dPTzzxhM4880z17dvXb9w777yj1q1bV7gYY4wefvhhnXXWWWrXrp127typ8PBwvzHh4eHKzMz0W7Zy5Uo99dRTvsejRo3SpEmTKlxHaaKiomzb9ukgODg40CXUaPRf5dB/lUP/VQ79d/pKSEjQ4sWL1b9//3I/d+zYsVq5cqXi4uK0Y8cOG6qrHmUKncOHD9f+/fv12GOP6dixY2rfvr0WLVqkoKDfn3748GGtWbNGs2fPrlAhxhgtWrRIR44c0Zw5c2RZlkJDQwsFzMzMzEITug4ePNjvVL/D4VBKSkqF6iiJ0+lUVFSUUlNTbbnj7XQQHh6ujIyMQJdRI9F/lUf/VRz9V3n0X+liYmICXYKfPGP0bVaOUj1eRTkdah3qlquaP8lo48aNWrNmjfbu3Vvj74wvU+iUpGnTpmnatGkaPXq0ZsyYocTERL/18fHx+uyzzzRnzhyNHz++XEUYY7R48WLt2rVLc+fOVUhIiCSpcePGevXVV+X1en2n2Hfv3l3ofwmxsbGKjY31PU5OTrb1h6LH4+GHbgUZY3jvKon+qzj6r/Lov4qj/2qWj9IytSIlXTnGyGlJHiO5LUtDYiLUPTKs2urYvXu3EhISKhQ4PR6PLMvyu0wxkMpdxQsvvKDffvutyHVHjhzR888/X+4ilixZoh07dmj27NkKC/v9G9m6dWu5XC6tWrVKeXl52rBhgw4dOqQLLrig3PsAAAAoi4/SMvXi0TRlGSOvpDwjeSVlGaMXj6bpo7TM0jZRpK+++kqtW7dWdHS0Bg8erGPHjkmSvvjiC3Xv3l0xMTFKSkrSq6++Kkl68sknNXbsWH3xxReKiIjQLbfcIklaunSpWrRooZiYGPXt21c7d+707SMhIUH33HOP2rVrp/DwcB0+fFipqakaN26cGjVqpHr16mnSpEnKzs6u1HtUEeUOncYYvzvZ/+iHH35QnTp1yrW9w4cP680339T+/ft1/fXX6+qrr9bVV1+tf//73woKCtL06dO1adMmDRs2TP/+979155131vjDywAA4OSUZ4xWpKTLFLPeSFqRkq48U9yI4i1dulSvv/66fv75Z+Xk5Ojmm2/WwYMH1b9/f912221KTk7W0qVLNWbMGG3fvl033nijFi9erA4dOig9PV0LFy7U+vXrdeutt+r555/Xr7/+qi5duuiyyy5TXl6ebz8vvfSSVq5cqdTUVMXFxWn06NHKycnRtm3b9P333+uHH37Q3LlzK/YGVUKZTq8/8cQTeuKJJySduPvu2muvLXRdZXZ2tvbs2aMhQ4aUq4D4+HitXr262PUJCQlasGBBubYJAABQEd9m5SinlECZY4y2ZuWoXVhIubY9adIkNW3aVJI0f/58dezYUeecc4769u2rK664QpLUqVMnDRo0SCtWrNCMGTMKbeOll17SqFGj1LlzZ0nSjBkztGjRIn322Wfq1q1bof0cPnxYq1ev1tGjR30H7aZPn66RI0dq/vz55aq/ssoUOhs0aKD27dtLkrZu3aoWLVoUmpIoODhYSUlJhT65CAAAoKZI9XjltCRvCbnTaUnHPd5yb7tx48a+vzdp0kS5ubnavXu3Xn/9ddWqVcu3Lj8/X3/961+L3MaBAwf87m0JCgpSo0aNdODAgSL3s2fPHnk8Hp1xxhm+ZYG6vrhMoXPgwIEaOHCg73FRNxIBAADUdFFOhzylnDn3GCnaWf6bc/44z/i+ffvkcrnUqFEjDR06VEuXLi3TNho2bKi9e/f+XovHo59//lkNGzb0LfvjjUONGzdWUFCQDh8+HPApu8r9jj333HMETgAAcEpqHeqWu5RpkdyWpVah7nJve9GiRdq9e7fS0tI0ffp0XXPNNRo5cqTeeustrVmzRvn5+crNzdVnn32m7du3F7mN4cOH6/nnn9fmzZuVm5uruXPnKiYmRp06dSpyfL169XTppZfq5ptvVkpKiowx2r9/v95+++1y119ZJ8c99AAAACcB1/+mRSoudlqShsREVGi+zuuuu06XX365GjVqJKfTqYcffliNGjXSf/7zHy1cuFB169ZV/fr1dccddygnJ6fIbfTu3Vv33Xefrr32WsXHx+vDDz/UmjVr5HK5it3v888/r6CgILVt21bR0dG6+OKL/e54ry6WMRW4/eokV97Pfy8rp9OpmJgYpaSkMNdaBUVGRiotLS3QZdRI9F/l0X8VR/9VHv1Xuj/OuW23zMzMEnu5ovN0MsNO8co8OTwAAMDpontkmDpHhGprVo6Oe7yKdjrUKgCfSHQqIXQCAAAUwWVZ5Z4WCcXjmk4AAADYjtAJAAAA2xE6AQAAYDuu6QQAAKcdp9MZ6BJOO4ROAABw2nG7yz+5OyqH0+sAAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdpYxxgS6iKqWmpoqt9td5du1LEvBwcHKzc3VKfi2VYugoCDl5+cHuowaif6rPPqv4ui/yqP/SmfH726cPIICXYAdcnNzlZubW+XbdTqdCg4OVkZGhjweT5Vv/3QQGRmptLS0QJdRI9F/lUf/VRz9V3n0X+kInac2Tq8DAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGwXFOgCAABAzZCc79Ev+V4FWVIzV5DcDivQJaEGIXQCAIASJed7tCw1UztzPXJK8kpySeoV7talESFyWoRPlI7QCQAAipXi8WrBkXRlGiNJ8vxvea6k9zNydNTj1cjoMFkET5SCazoBAECx3k7PVpYx8haxziNpc3ae9uZ5ilgL+CN0AgCAIuUbo8+zclVSpHRI2piVU10loQYjdAIAgCJleo3yShnjlZScX9RxUMAfoRMAABQpxGGptCs1LUmRDuIESkeXAACAIgVbllq7g0oNCx1Dg6ulHtRshE4AAFCsv0SEFHu00ynpjCCnznEzGQ5KR+gEAADFauQK0sSYcEVYlhw6MddiQcQ8MzhIk2qHy8F0SSgD/msCAABK1Nzt0vz4KG3NydPPeR65LEut3C41cDkDXRpqEEInAAAoldOy1CYkWG1CAl0JaipOrwMAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALBdUKALAACgKmR5jbbm5CnTa1TH6VCSO0hOywp0WQD+h9AJAKjRvMbozfRsrc3IkdGJU3geSWGWpWHRoWoTEhzgCgFInF4HANRwq9Oy9W5GjvJ1ImzmSfJKSjdGTx/L1Hc5eYEtEIAkQicAoAY77vHq/cwceYtZbyStTM2SMaY6ywJQBEInAKDG2pydK2cpYw57vPo531Mt9QAoHqETAFBjpXmNSjuG6ZCU6uVIJxBohE4AQI0V7bBU2v3pXknRDn7dAYFWI+5eT09P1+OPP64tW7YoNDRUV199tf7yl78EuiwAQIC1DwnWa2nZJY6p53SoYRChEwi0GhE6lyxZIo/Ho+eee04HDx7UjBkz1KhRI5177rmBLg0AEEBRTocuCnfrvYyibyayJA2OCpXFfJ1AwJ30//XLzs7Wxx9/rBEjRigsLEzNmjVTnz59tHbt2kCXBgA4CVwWEaK/RLjlkuSU5NKJsBntsDSuVriS3K7AFghAUg040nngwAFJUuPGjX3LmjZtqlWrVvkeJycnKzk52ffY4XAoLi6uymtxOp1+f6L8LMvi/asg+q/y6L+KO9n779LoCPWNDNd3Obm+TyRq4XbJcRId4aT/cLo76UNndna2QkND/ZaFh4crKyvL93jlypV66qmnfI9HjRqlSZMm2VZTVFSUbds+HQQH8+kglUH/VQ79Vzkne//VC3QBpaD/cDo76UNnSEiIX8CUpIyMDL8gOnjwYPXs2dP32OFwKCUlpcprcTqdioqKUmpqqjwe5nyriPDwcGVkZAS6jBqJ/qs8+q/i6L/Ko/9KFxMTE+gSYKOTPnQ2bNhQkrR//36dccYZkqTdu3erSZMmvjGxsbGKjY31PU5OTrb1h6LH4+GHbgUZY3jvKon+qzj6r/Lov4qj/3C6O+lvJAoJCVHXrl318ssvKzMzU7t379b777+vCy+8MNClAQAAoIxO+iOdkjRu3Dg99thjGjVqlMLCwjR8+HC1adMm0GUBAACgjGpE6IyIiNC0adMCXQYAAAAq6KQ/vQ4AAICaj9AJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbEfoBAAAgO0InQAAALAdoRMAAAC2I3QCAADAdoROAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADYjtAJAAAA2xE6AQAAYDtCJwAAAGxH6AQAAIDtCJ0AAACwHaETAAAAtiN0AgAAwHaETgAAANiO0AkAAADbEToBAABgO0InAAAAbBcU6AIAAEjxeLUpM0d78zxyWZbahLjULsQll2UFujQAVYTQCQAIqI2ZOfp/qVlySsr/37Jvc/K0Ks3STbUjVC/IGcjyAFQRTq8DAAJmW06e/l9qlox+D5yS5JWU7jV65Gi6co0JUHUAqhKhEwAQMG+nZxe7zispw2v0ZVZu9RUEwDaETgBAQOR4jXbleVTScUyPpC3ZedVVEgAbEToBAAGRV2Lc/B2n14FTA6ETABAQYZal8FLuTndKOsPFjUTAqYDQCQAICIdlqUdYsEqKlB5J3cPc1VUSABsROgEAAdM3PEQNgpxFBk9L0oAIt+oyZRJwSiB0AgACxu2wdEvtCF0Y7lboH860NwhyaHStMF0cERq44gBUKSaHBwAElNth6fLIUF0aEaLjXiOXpEgnx0SAU41lzKl3W2Bqaqrc7qq/BsiyLAUHBys3N1en4NtWLYKCgpSfn1/6QBRC/1Ue/Vdx9F/l0X+ls+N3N04ep+SRztzcXOXmVv1kwk6nU8HBwcrIyJDH46ny7Z8OIiMjlZaWFugyaiT6r/Lov4qj/yqP/isdofPUxvkLAAAA2I7QCQAAANsROgEAAGA7QicAAABsR+gEAACA7QidAAAAsB2hEwAAALYjdAIAAMB2hE4AAADY7pT8GEy7JCcna+XKlRo8eLBiY2MDXQ5OM/QfAon+A1BZHOksh+TkZD311FNKTk4OdCk4DdF/CCT6D0BlEToBAABgO0InAAAAbOecNWvWrEAXUZOEhobq/PPPV1hYWKBLwWmI/kMg0X8AKoMbiQAAAGA7Tq8DAADAdoROAAAA2I7QCQAAANsFBbqAmuLVV1/VunXrdPjwYYWHh6t379669tpr5XQ6JUnp6el6/PHHtWXLFoWGhurqq6/WX/7ylwBXjVMF/YXqkpeXp8WLF+ubb75RWlqaYmNjNWTIEPXq1UuStHfvXj366KPas2eP6tWrpwkTJuicc84JbNEAagRCZxkZY3TTTTcpMTFRR48e1bx58xQWFqbBgwdLkpYsWSKPx6PnnntOBw8e1IwZM9SoUSOde+65Aa4cpwL6C9XF4/Godu3amjdvnuLj4/X9999rzpw5qlevns4880zNmzdP/fv31913362NGzdq/vz5evLJJxURERHo0gGc5Di9XkaDBw/WWWedpaCgIMXHx6tnz57atm2bJCk7O1sff/yxRowYobCwMDVr1kx9+vTR2rVrA1w1TgX0F6pTSEiIhg8frnr16snhcKhly5ZKSkrS9u3b9e233yonJ0eDBg2Sy+VS7969VbduXW3atCnQZQOoAQidFfTdd9+pcePGkqQDBw5Iku+xJDVt2lR79+4NSG04tdBfCKTs7Gz9+OOPatKkifbt26cmTZrI4fj9V0diYqL27dsXwAoB1BSETp04nVTS15+98cYb2rNnjwYNGiTpxA/l0NBQvzHh4eHKysqqlvpxaqO/ECjGGD388MM666yz1K5dO2VlZSk8PNxvDL0IoKy4plPSv/71L23durXIdbVq1dILL7zge7xu3Tq98sormjdvnqKioiSdOB315x+6GRkZhYICUBH0FwLBGKNFixbpyP9v7/5Cmvr/OI6/Jt+ZetEoljYrDCVosYwodNGNUWCtuqtumzQNBVlJFgQjnRDdRNscohRJwegmZBBddCE0IoSS6mIjumkDQUwTLGIrNPe7iN9gX7U/fne0vt/nAw5sn/M5n70PHLYX53PO2fS0/H6/TCaTSktLlU6n8/ql02mORQA/hdAp6erVqz/VLxaLaXBwUD09Pdq8eXOufdOmTZKksbExbdmyRZKUTCZVVVVV+GLxn8PxhZWWzWbV39+vt2/fqqenRyUlJZK+XeIxNDSk+fn53BR7MpnU4cOHV7NcAH8Iptd/UiwW082bN3XlypUFP/YlJSXav3+/IpGI0um0ksmkhoeHdfDgwVWqFv8mHF9YaQMDA3rz5o26u7vz/uLO/4MAAAR6SURBVGd9586dMpvNikajmp2dVSwW08TEhPbt27eK1QL4U/Df6z/J4/FoenpaZrM517Zjxw51dXVJ+vYcxXA4rBcvXqisrIznKKKgOL6wUiYnJ+XxeGQ2m3PPIZakEydO6NSpU0qlUgqHw0qlUqqoqFBra6scDscqVgzgT0HoBAAAgOGYXgcAAIDhCJ0AAAAwHKETAAAAhiN0AgAAwHCETgAAABiO0AkAAADDEToBAABgOEInAAAADEfoBLAs0WhUfX19C9ojkYjq6upksVi0du1a2e12eTweTU5O5vo0NDTIZDLJ5/Mt2L6hoUHHjh3LvX/8+LFMJtOiy6FDh3L9RkdH1dTUJLvdrqKiorwxAACr76/VLgDAnykajWp0dFRtbW25tmvXruny5cs6f/68/H6/stms4vG4IpGIxsfHVV5enjdGKBRSR0eH1q1b98PPGxwc1Pbt2/PaLBZL7vXTp0/15MkT1dfXK5PJ/MO9AwAUGqETQMH09vbK7Xbr+vXrubYjR46os7NT8/PzeX3r6+sVj8cVCATU3d39w7EdDof27t275Pr29nZ5vV5J386WAgB+L0yvA/hlbrdbd+7cUSKRyE11u91uzczMyGazLbpNUVH+143ValVbW5uCwaBmZmb+cU1/Hx8A8HvhWxrAL/P5fHK5XKqurtbIyIhGRkbk8/m0Z88e9ff369atW5qYmPjhOBcuXNDs7KwCgcAP+379+lVzc3N5y9/PngIAfl+ETgC/rKamRhs2bFBpaamcTqecTqdqamrU19en9evXq7m5WTabTdXV1fJ6vUqlUouOU15ertbWVgWDQX348OG7n+l0OmU2m/MWv99vwN4BAIxA6ARQMA6HQ4lEQg8fPpTX65XFYlEoFFJtba1evXq16DadnZ368uWLgsHgd8e+e/eunj9/nre0tLQYsRsAAANwIxGAgiouLpbL5ZLL5ZIkPXr0SEePHpXf79fQ0NCC/hUVFTp79qwCgYDOnTu35Lh2u/27NxIBAH5vnOkEYKjGxkbt2rVLr1+/XrLPxYsXlclkFAqFVrAyAMBKInQCWJbi4mJ9/vw5r+3du3cL+mUyGY2NjWnjxo1LjmWz2dTS0qIbN27o48ePBa8VALD6mF4HsCx2u123b9/WvXv3tG3bNlmtVtXV1en48eNqbGyUzWbT+Pi4ent79f79+9wzNJdy6dIlDQwM6OXLl6qsrFywPh6Pa25uLq9tzZo12r17tyRpampKsVgs9/rTp0+6f/++JMnlcqmsrKwQuw0AWCZCJ4BlOXPmjJ49e6b29nZNT0/r9OnT6urq0oMHD9TR0aGpqSlZrVbV1tZqeHhYBw4c+O54lZWVam5uVjgcXnR9U1PTgraqqqrcnfGJREInT57MW///98lkUlu3bv31nQQAFIwpm81mV7sIAAAA/LtxTScAAAAMR+gEAACA4QidAAAAMByhEwAAAIYjdAIAAMBwhE4AAAAYjtAJAAAAwxE6AQAAYDhCJwAAAAxH6AQAAIDhCJ0AAAAw3P8AqFKGCB1EiKEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<ggplot: (7549237203)>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# graph t-SNE\n",
    "chart = ggplot(dataframe, aes(x='tSNE1', y='tSNE2', color='label') ) \\\n",
    "        + geom_point(size=3,alpha=1) \\\n",
    "        + ggtitle(\"tSNE dimensions colored by digit\")\n",
    "chart"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In both our PCA and our t-SNE graph, we can see a clear distinction between the before-caffeine cluster and the after-caffeine cluster! This indicates that caffeine is changing transcriptome state as the yeast is reacting to the caffeine in its environment on the transcriptome level.\n",
    "\n",
    "Although PCA and t-SNE is often only helpful to indicate that transcripts are being differentially expressed, it is still a crucial first step in data analysis. After this, scientists will often use [heatmaps](https://en.wikipedia.org/wiki/Heat_map), [multidimensional scaling](https://en.wikipedia.org/wiki/Multidimensional_scaling), [statistical tests] (https://pachterlab.github.io/sleuth_walkthroughs/trapnell/analysis.html), or other techniques to continue exploring the secrets hidden within the data. We invite you to continue learning about different crucial methods used by researches and to continue investigating how yeast uses gene regulation to respond to chemical changes in its environment."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Additional Resources\n",
    "\n",
    "- If you are interested in annotation files (`.gff3`), which identify the locations of genes on the genome and the functions of genes, [UCSC Genome Bioinformatics](http://hgdownload.soe.ucsc.edu/downloads.html) is a great starting place to find annotations for your organism.\n",
    "- more technical information on the paired-end sequencing samples used in the original paper can be found [here](https://www.ncbi.nlm.nih.gov/bioproject/PRJNA385168) and [here](https://trace.ncbi.nlm.nih.gov/Traces/sra/sra.cgi?run=SRR5511057). \n",
    "- another common pipline you can explore is: Trimmomatic/STAR/[HTseq](https://htseq.readthedocs.io/en/release_0.11.1/), this provides the expression of transcripts in *counts*, which is great for learning how to do statistical analysis on sequencing data\n",
    "  "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}