3-state secondary structure predictor.
- Clone my repository
- Go into the folder named scripts
- Choose which predictor you want to run, PSSM or regular
- If you run the script without specifying FASTA-file and/or directory, you will get instructions about this
the prediction will be written to the screen and saved in a output file to be found in the "results" folder
dataparser.py
This cointain four functions:
def parse_fasta(filename):
#writes the dataset for traning into a dictionary called parse_dict with id as key and [sequence, topology] as value. Removes \n and >
def inputvector_X(sequence, window=7):
#Takes one string of amino acids at a time, and creates windows that are encoded into binary and finaly converted to numpy array. The shape is in form of (window, nr of windows) where (row, colum)
#Each numpy array representing one protein
def train_inputvector_X(parse_dict):
#loopes over the dictionary parse_dict and extracts only the sequence to feeds it into the def inputvector_X(sequence, window=7):. each of the numpy arrays created from that function extended into a list called X_vector, so that you get a list of arrays. this list is finaly created into a numpy array.
#this is the final X_vector that is used for traning
def inputvector_y(parse_dict):
#loopes over the dictionary parse_dict and extracts only the topology and encodes it into intergears. Appends them inot a list calle y_vector and turns it into a numpy array.
#this is the final y_vector that is used for traning
train_test.py
#imports the X_vector and the y_vector from the dataparser.py file
#trains the model with deafult parameter except for 'one vs one'
#saves the model with pickle
cross_validation
#imports the X_vector and the y_vector from the dataparser.py file
#for loop, that does fivefold cross validation for every second windowsizes in range 3-20
#stores the result in dictionary. With windowsize as key and the 5 scores as value.
FASTA_to_inputvector.py
def parse_fasta(filename):
#write FASTA file for prediction into a dictionary called seq_dict with id as key and [sequence] as value. Removes \n and >
def inputvector_X(sequence, window=7):
#is the same function as in dataparser.py.
svm_prediction.py
#imports the function inputvector_X() and parse_fasta() from FASTA_to_inputvector
#loades in the models saved with pickle
#runs the prediction on the X_vector created by inputvector_X()
#Converts the prediction from intergears to string (H, S, C).
I have 3 training datsets in my datset folder: data.txt, datamini.txt, dataminix2.txt
data.txt
is my complete datset and contains 399 sequences. I do not recomend running that. Therfore I have created two additional datsets.
datamini.txt
is the one I have used for training the model and doing the cross-validation.
dataminix2.txt
The last dataset called dataminix2.txt is very small containing only 3 sequences. It was used to for easy readabilety when making the scripts.
If you want to change the windowsize, it is important that it is change in all the three places it is beeing specified. Two places in the script dataparser.py and one place in FASTA_to_inputvector.py