Difference between revisions of "Team:NCTU Formosa/Improve"

     The way we use SCM to cure fungal diseases is just a part for its ability. In fact, the peptide prediction system based on the SCM can be specialized in different cases of evaluating sequences.

     We decided to apply the method to NCTU_Formosa 2016, which utilized spider toxin to kill the pests. We introduced the scoring card to the insecticidal protein to see whether we could also predict invertebrate proteins from ion channel impairing toxins, improving their searching tool while finding more candidates for the project last year.

     First, we collected the insecticidal and ion channel impairing toxins by 2016 selection database. After deleting peptides which contained non-standard amino acids, we randomly chose positive and negative data to our datasets and divided them into two datasets, training datasets and testing datasets.

一張表格

     For training parts, after initializing the first scorecard, we used IGA to optimize the scorecard for ten generations.

Results

     FullTrain_acc=91.70454568181819
     CV acc(train)=93.8636343698348
     CV auc(train)=95.44599143143164
     Best theshold=498.75
     Best_acc(test)=88.86363681818182
     Sensitivity(test)=0.7031249936523439
     Specitivity(test)=0.9202127637222726

一張表格

一張表格

Discussion

     To improve the project of NCTU_Formosa 2016, we introduced the scoring card method to the insecticidal proteins. By using the method, we can predict more new insecticidal proteins.

     We collected about three thousands of ion channel impairing toxins.

     Below is the excerpt of the peptide list.

一個表格