Flu-Prediction

Predicting Future Flu Virus Strains with Machine Learning. These programs predict future influenza virus strains based on previous trends in flu mutations.

Talks

Check out my talks at PyData and PyGotham.

License

Flu-Prediction is available under the GPLv3 License.

Dependencies

Python 2 or 3 with Numpy, Biopython, and Scikit-learn libraries installed.

To use:

Clone/download the repository. Install the dependencies by doing pip install -r requirements.txt.

Input any HA (hemagglutinin) or NA (neuraminidase) flu protein sequence and it's corresponding child sequence into the program and it will output a predicted offspring of that specific flu strain.

Reading in a FASTA file with Biopython

Use the Biopython library to import a sequence (a FASTA file format). You can use any flu FASTA file of your choosing, or you can use the ones in the Flu-Data folder. The data in the Flu-Data folder contain a wide variety of flu FASTA files, from single flu strains up to 1000 flu strains, which are grouped by flu subtype and protein. Data was obtained from the Influenza Research Database (IRD).

from Bio import SeqIO
sequence = SeqIO.parse('myfasta.fasta','fasta') # put your FASTA file here
parent_fasta = parent.fasta 
parent_seq = parent.seq

child_fasta = parent.fasta 
child_seq = child.seq

Encoding

Then encode it with the Encoding_v2 module:

from Encoding_v2 import encoding
parent = []
for k in range(len(X0)):
    encoded_parent = encoding(parent_seq[k])
    parent.append(encoded_parent)
    
child = []
for l in range(len(y0)):
    encoded_child = encoding(child_seq[l])
    child.append(encoded_child)

This turns the sequence into a list of float64's. Then, give the X and y to the machine learning algorithm. Enter any machine learning algorithm (eg, RandomForestsRegressor, DecisionTreeRegressor, etc.) in the 'algorithm' parts of the code.

Fitting the model

Substitute algorithm for any scikit-learn model of your choosing.

from sklearn.algorithms import algorithm()
alg = algorithm()
alg.fit(X,y)
alg.predict(new_X)

The algorithm I use in this project is a Random Forests Regressor model:

from sklearn.ensemble import RandomForestRegressor()
rfr = RandomForestRegressor() # Specify and parameters in the parenthesis
rfr.fit(X,y)
rfr.predict(new_X)

Computing accuracy using K-Fold cross-validation:

from sklearn import cross_validation
algorithm_scores = cross_validation.cross_val_score(algorithm,X,y,cv=2)
print 'Algorithm Trees',algorithm_scores
print("Average Accuracy: %0.2f (+/- %0.2f)" % (algorithm_scores.mean()*100, algorithm_scores.std() *100))

Computing accuracy using R² (for linear models):

from sklearn import metrics
y_pred = algorithm.predict(X_test)
print 'Algorithm R2 score:', metrics.r2_score(y_test,y_pred,multioutput='variance_weighted')

Computing accuracy using Mean Squared Error (MSE):

from sklearn import metrics
y_pred = algorithm.predict(X_test)
print 'Algorithm mean squared error:', metrics.mean_squared_error(y_test,y_pred,multioutput='variance_weighted')

Predicting Flu Strains:

y_pred = algorithm.predict(X)
print y_pred

The prediction output is a list of floats. Each number in the float corresponds to a base pair: A to 1, T to 2, G to 3, and C to 4.