This repository is created for 2019 BME203 Intro to Bioinfo course individual project. It uses ResNet to predict splice site.
Dataset
This model uses the dataset specified in HS3D.
true and false splice site files shall be saved in ./data directory
Dependencies
Pytorch for python 3.x should be installed. Please follow instructions on official page.
To install all the other packages:
pip install ipdb, scikit-learn, tqdm, numpy
How to Run
Simply run main.py script:
python main.py
Test Session
To run test session, DSSP pretrained models should be stored in ./DSSP directory. Pretrained models can be downloaded in github.com/DSSP-github/DSSP
Please specify the saved model to test by using -output argument.
python test.py -output output_directory_name
Reports
Several metrics were estimated on the balanced(1:1 ratio of labels) dataset.
| Model | Validation Acc | binary F1 | weighted F1 |
|---|---|---|---|
| resnet18-epoch15 | 82.870982 | 0.830501 | 0.828719 |
| DSSP | 49.229414 | 0.000000 | 0.324807 |