protein-ss-pred

GOR method and an SVM-based method for protein secondary structure prediction.
Detailed description of the methods and datasets can be found in the project report.

Getting started

Requirements

The implementations require numpy and scikit-learn packages.

pip install numpy
pip install scikit-learn

In order to play with the notebooks make sure to install Jupyter Notebook.

pip install notebook

Installation

git clone https://github.com/katarinaelez/protein-ss-pred

Usage

GOR

Pretrained model (model.npz) is available.
Prediction from the GOR method can be obtained using:

python gor-predict.py [-h] (--pssm PSSM | --fasta FASTA) filename_model

For example:

python src/gor-predict.py --pssm data/blindTest/pssm/4S1H\:A.pssm models/model.npz

SVM

Pretrained model (model.sav.tar.gz) is available.
Before it can be used it must be extracted in the following way:

tar -xzvf models/model.sav.tar.gz -C models/

Prediction from the SVM-based method can be obtained using:

python svm-predict.py [-h] (--pssm PSSM | --fasta FASTA) [--probs] filename_model

For example:

python src/svm-predict.py --pssm data/blindTest/pssm/4S1H\:A.pssm models/model.sav

Training

GOR

GOR model can be trained using:

python gor-train.py [-h] [--filename_model FILENAME_MODEL]
                    [--window_size WINDOW_SIZE]
                    filename_id_list dir_pssm dir_dssp

For example:

python src/gor-train.py data/training/list.txt data/training/pssm data/training/dssp --filename_model models/model

SVM

SVM model can be trained using:

python svm-train.py [-h] [--filename_model FILENAME_MODEL]
                    [--window_size WINDOW_SIZE]
                    filename_id_list dir_pssm dir_dssp

For example:

python src/svm-train.py data/training/list.txt data/training/pssm data/training/dssp --filename_model models/model

Performance

	GOR		SVM
	CV	Blind test	CV	Blind test
SEN_H	0.86±0.01	0.83	0.80±0.01	0.72
SEN_E	0.62±0.01	0.60	0.58±0.01	0.62
SEN_C	0.42±0.01	0.42	0.82±0.00	0.85
PPV_H	0.58±0.01	0.60	0.82±0.01	0.85
PPV_E	0.54±0.01	0.58	0.75±0.01	0.80
PPV_C	0.80±0.01	0.73	0.72±0.00	0.65
MCC_H	0.50±0.01	0.46	0.71±0.01	0.67
MCC_E	0.45±0.01	0.46	0.58±0.01	0.63
MCC_C	0.40±0.01	0.39	0.58±0.01	0.56
SOV_H	65.48±0.99	62.70	76.39±0.97	68.64
SOV_E	58.64±1.35	63.18	59.20±2.18	67.19
SOV_C	43.09±0.69	45.57	70.18±0.93	70.93
ACC	0.62±0.00	0.62	0.76±0.00	0.75

License

MIT @ Katarina Elez