philippmuench

#!/usr/bin/env Rscript

# Script to find exact matches between evaluation and training datasets
# Identifies which samples from evaluation set are identical to samples in training set

# Load necessary libraries
if (!requireNamespace("Matrix", quietly = TRUE)) {
  install.packages("Matrix")
}
library(Matrix)

philippmuench

## Setup
Sys.setenv(CUDA_VISIBLE_DEVICES='')
options(tensorflow.extract.warn_tensors_passed_asis = FALSE)

library(dplyr, warn.conflicts = FALSE)
library(purrr)
library(glue)
library(envir)

library(tensorflow)

philippmuench

import os
import hashlib
from pathlib import Path
import requests
import logging
from colorama import Fore, Style, init
import gzip
import shutil
import time
import random

philippmuench

import os
import random

# Get a list of all FASTA files in the bacdive_gff folder
FASTA_FILES, = glob_wildcards("fasta/{fasta_file}.fasta")

rule all:
    input:
	expand("gff/{fasta_file}.gff", fasta_file=FASTA_FILES),
        expand("reformatted_gff_shuffled/{fasta_file}.gff", fasta_file=FASTA_FILES),

philippmuench

import os
import argparse
import random

def is_fasta(filename):
    try:
        with open(filename, 'r') as f:
            first_line = f.readline().strip()
            if not first_line:
                return 'empty'

philippmuench

## app.R ##
library(shinydashboard)
library(shiny)
library(keras)
library(deepG)
library(ggplot2)
library(dplyr)
library(DT)
library(hdf5r)
library(plotly)

philippmuench

---
title: "GenomeNet Viewer"
output: 
  flexdashboard::flex_dashboard:
  orientation: rows
social: menu
theme: united #cerulean
source_code: embed
runtime: shiny
---

philippmuench

#' @title Trains a (mostly) LSTM model on genomic data. Designed for developing genome based language models (GenomeNet)
#'
#' @description
#' Depth and number of neurons per layer of the netwok can be specified. First layer can be a Convolutional Neural Network (CNN) that is designed to capture codons.
#' If a path to a folder where FASTA files are located is provided, batches will ge generated using an external generator which
#' is recommended for big training sets. Alternative, a dataset can be supplied that holds the preprocessed batches (generated by \code{preprocessSemiRedundant()})
#'  and keeps them in RAM. Supports also training on instances with multiple GPUs and scales linear with number of GPUs present.
#' @param train_type Either "lm" for language model, "label_header" or "label_folder". Language model is trained to predict next character in sequence.
#' label_header/label_folder are trained to predict a corresponding class, given a sequence as input. If "label_header", class will be read from f

philippmuench

trainMinimalFunctionalAPI <- function(path = "example_files/fasta") {
  library(wavenet)
  message("Initialize model! This can take a few minutes.")

  maxlen <- 1000
  input <- keras::layer_input(batch_shape = c(64, maxlen, 6))
  

  # https://github.com/ibab/tensorflow-wavenet/blob/master/wavenet/ops.py#L46
  first <- keras::layer_conv_1d(

philippmuench

trainMinimalFunctionalAPI <- function(path = "example_files/fasta") {
  
  message("Initialize model! This can take a few minutes.")

  input <- keras::layer_input(batch_shape = c(256, 50, 6))
  
  cnn <- 
    keras::layer_conv_1d(
      object = input,
      kernel_size = 3,