Introdcution

This is an implementation of DNA gene family classification. To run the code:

python train_and_eval.py

Requirements

Please install the required packages in requirements.txt:

pip install -r requirements.txt

Data Preprocessing

The data preprocessing is implemented in data_process.py. Please see DNADataset class for details.

The DNA sequences are:

1. Sliced into subsequences of maximum length of 512 nucleotides.
2. Optionally, before slicing and padding, the sequences can be augmented by their reverse complement.
3. The sequences are tokenized into 5-mers, and then converted to one-hot encoding.

Model

A Convolutional Neural Network (CNN) is implemented in model.py.

Data splits and evaluation

The input data is split into 5 folds of 20% test sequences and 80% training. These splits are iterated over for 5 times, allowing each fold to be used as a test set once. For any iteration, the training set is further split into 80% training and 20% validation.

The final test score is computed over the predictions of all the sequences when they were used in a test set. We report the accuracy and f1 score of the model on the whole dataset.