This is a tool for faster feature extraction from genome sequences and making efficient prediction. To build efficient prediction model, user can go through the following instructions step by step. If user only wants to predict query sequences from our built model, then just go to step Make prediction for query sequences. Currently, FastFeatGen supports text file as shown in dataset.txt file in datasets directory; however, fasta file can also be used with a simple preprocessing.
Users need to have following softwares/tools installed on their PC.
GCC version >= 4.9
OpenMP version >= 4.5
Python version >= 3.6.6
Python packages: sklearn, mathplotlib, keras
To extract features from a dataset, type the following command:
$ g++ -fopenmp -std=c++11 -O3 featureExtraction.cpp -o featureExtraction
$ ./featureExtraction -in datasets/dataset.txt -out datasets/extractedfile.csv -gen 2 -nt 4 -w 41 -o 2 -p 4 -g 4
Here,
in - name of input data file that contains positive and negative DNA sequences.
All positive sequences are followed by all negative sequences. We shuffle those sequences later.
out - name of output file where extraced features will be saved. A csv file name is recommended.
gen - An integer value to denote genome. 2 for rice genome, 3 for rat genome. 0 (or 1) for only feature
construction from rice (or rat) genome if Positive.csv and Negative.csv files exist in datasets folder.
nt - number of threads to be used to extract features.
w - window size of each DNA sequence.
o - order of position independent features. This is also known as k-mer or n-gram.
p - number of positions to be considered for order $o$-mers while constructing position specific features.
g - number of gap to be considered between two DNA alphabets while constructing di-gapped features.
Extracted features file 'extractedfile.csv' will be saved in datasets directory.
To build best model from SVM, RF or ETC, run the following:
$ python fastfeatgen.py -f datasets/extractedfile.csv -m 1 -p datasets/Positive.txt -n datasets/Negative.txt
Here,
f - name of extracted feature file.
m - an integer to denote rice (1) or rat(2) genome in output file.
p - name of file that contains positive sequences (one sequence per line).
n - name of file that contains negative sequences (one sequence per line).
To predict classification for a set of query sequences, use the following command:
$ python predict.py -q datasets/queryfile.txt -b results/models/bestModelETC10g1.joblib -f results/models/importantfeaturesETC10g1.txt -nt 4
Here,
q - name of query file where all query sequences are present one per line.
b - name of model file based on which prediction will be made.
f - name of important feature files for provided model file.
nt - number of threads to be used to extract features from query sequences.
After running this, our provided model will be used to predict the classification of query sequences.
Some extracted features can be found here. Our results and log files can be found at results folder.
If you find this repository helpful, please cite the following paper:
Md. Khaledur Rahman, "FastFeatGen: Faster parallel feature extraction from genome sequences and efficient prediction of DNA N6-methyladenine sites", In the 9th Proceedings of International Conference on Computational Advances in Bio and medical Sciences (ICCABS 2019), November, 2019, Florida, USA.
This tool is maintained by Md. Khaledur Rahman. If you face any problem or have any query, don't hesitate to send an email to khaled.cse.07@gmail.com or, morahma@iu.edu