SU-project

Beta barrel (4 state) individual assignment

1. clean_CV includes functions for:

   • parser
   • window list
   • binary list of aa
   • numerical list of topology
   • topology back to ch
   • save into .npz file
   • SVM
   • cross validation

2. cv

   • tested on a range of window sizes
   • window size 17

3. trained_model

   • saves trained model

4. final_predictor

   • predicts topology using model
   • output looks like this for each protein:
     *ID
     *amino acid sequence
     *predicted topology

Trained my model with all the proteins I was given (42 proteins) and window size 17 with a linear kernel. Used "testingdata_pred.txt" as my test file for predictor.

For PSSM portion:

1. all_pssm includes functions to:

   • parse and prepare input for predictor
   • optimized version (PSSM_model.sav)

2. PSSM_predictor

   • predicts topology on 50 proteins using model, currently split as 80% train and 20% test on original dataset (membrane-beta_4state.3line1.txt)
   • "PSSM_prediction.txt" is the output for the predicted topology of 50 proteins, currently 80% train and 20% test results
   • "PSSM_prediction_scores.txt" is all the data of my scores

Trained my model with the PSSMs found in folder "PSSM" and tested on the PSSMs found in folder "PSSM_50test", both located in "PSI-BLAST". But my predictor is currently set-up to train 80% of my dataset and test 20% of it.