[Solved] QBIO401 Assignment8-models to predict proteinDNA binding

We are going to build models to predict proteinDNA binding based on the data obtained from highthroughput in vitro proteinDNA binding assays.

Download and install the following packages from Jupyter Notebook or other platforms:

1. Bioconductor (refer the instructions on https://www.bioconductor.org/install/)
2. The DNA shape prediction tool: DNAshapeR ( refer the instructions on https://www.bioconductor.org/packages/release/bioc/html/DNAshapeR.html)
3. The machine learning tool: caret (refer the instructions on https://github.com/topepo/caret)

Complete the following tasks:

1. Use the DNAshapeR package to predict DNA shape for each sequence bound by transcription factor Mad, Max and Myc. The datasets were obtained from the in vitro gcPBM (genomic context protein binding microarray) assay. The sequence data in FASTA format (Mad.fa, Max.fa and Myc.fa) can be found on Blackboard. Use plotShape() or heatShape() functions of DNAshapeR to generate ensemble plots for the DNA shape parameters of minor groove width (MGW), propeller twist (ProT), Roll, and helix twist (HelT). Discuss what you find from the results [2pt].
2. Write an R function that takes as input a FASTA file. The function returns a feature matrix for 1mer sequence model using one-hot encoding which is a way to represent categorical data as binary vector [3pt].

For DNA, we have four categories A, T, G, and C. Thus a one hot code for DNA is: A: [0,0,0,1]

C: [0,0,1,0]

G: [0,1,0,0]

T: [1,0,0,0]

For example, for the sequence AATTC, the 1-mer one-hot encoding is:

[0,0,0,1,0,0,0,1,1,0,0,0,1,0,0,0,1,0,0,0]

3. Use the function from Q#2 to generate a 1-mer feature vector and combine the shape features generated from Q#1 for 1-mer+shape model with respect to the datasets of Mad, Max and Myc. Use the caret package to build L2-regularized MLR models for 1-mer and 1mer+shape features with 10-fold cross validation (set the L2 lambda value between 2^-15 to 2¹⁵), and print out the average R2 (coefficient of determination) for these two models with respect to

the datasets of Mad, Max and Myc. The corresponding proteinDNA binding affinities can be found in corresponding *.s files. Compare the performance between 1-mer sequence model and 1-mer+shape model and explain what you find [3pt].