This project aims to classify true and false intron and exon 5' segregators with a basic classifier. As written, it is mainly used as a toy project to work on data preprocessing and neural network model fitting. This work was performed at KAUST, Saudi Arabia.
The original data is given as a large file of records of the DNA of Arabidopsis thaliana in fasta format.
In this project the libraries Biopython, SciKit-Learn and Keras were used.
The initial data is parsed with the SeqIO module of Biopython. Afterwards a One Hot Encoding of the nucleotide sequences is performed. For the direct preparation for the machine learning model and its robustness a k-fold cross validation is performed.
Different kinds of neural networks like shallow, intermediate and deep were applied and their performance was documented in the form of mean and standard deviation.
The Convert_fa_to_fasta.py script converts the raw data to files that are handable by the SeqIO parser and needs to performed at first.
Second, perform the Prepare_data.py script to store the given, preprocessed data in .npy files.
This data is used in the third step in the simple_classifier.py script to actually execute the ML model.
Results are stored in the results directory.