kth-signal-peptide-prediction

This is a repository for project at KTH relate to subject DD2404 Description:

Repository structure

/bin . All python code are stored under /bin folder,

/data . All data are stored under /data folder

/results . All Results are stored under /results folder, with each test-run stored into corresponding date, log books are here too

/doc . All documentations relate to this project are store under /doc

Logbooks are stored under results

Requirements

• Python 3
• Numpy
• glob
• sklearn
• matplotlib

Training data

Training data should be located under /data/negative_examples and /data/positive_examples The program will look for the training data under those path, please ensure the program is run under the bin folder of this repo

How to use the code to train and predict signal peptides?

Under /bin folder, run ./runall SAVEPATH [argumemts]

usage: runall [-h] [-pt] [--nfold NFOLD] [-f] [-p]
                            [-t THRESHOLD] [-i INPUT_MODE] [-q]
                            save_path

There are 3 function groups, they are 1.Performance Test, 2.F-score test, 3. Prediction

1. Perfromance test (-pt) train the model based on the data specified by [-i INPUT_MODE], and produce a graph thay compares all models. The graph is saved in the path speicifed by SAVEPATH.

   • --nfold can be used as an additional argument to spcify the number of fold for cross-validation. Default to 3.

   example: ./runall ../results/2017-12-26 -pt --nfold 5
   

   Above command will run performance test with nfold of 5 and save to the results folder

2. F-score test (-f) train the models with data specified by INPUT_MODE based on 80/20 data split, and save precision/recall and f-beta score into SAVEPATH

   example: ./runall ../results/2017-12-26 -f -i 1
   

   Above code will run f-score test with non_TM data only

3. Prediction (-p), Train Logistic regression and run predictions on files located under /data/proteomes/ ,files must be in fasta format.

   • -t can be used to indicate the THRESHOLD for classify a sample into positive. Default to 0.5

[-i INPUT_MODE] is the argument for specify which data is used for training.

• 1 = non-TM only,
• 2 = TM only,
• All other numbers = TM + non-TM will be usesd

[-q] quiet mode, will not print anything on screen.