mTIM

transcript identification method

A) PREPARATION

• software package comes without parallelization software package but training and prediction are ready for it

  • see general_settings.m : CFG.grid_use = 0;

• for generating mex-files you need to adapt the makefiles in the corresponding directories (just change the path to your matlab) mTIM/src/utils : change Makefile, then type 'make all' mTIM/src/sotool/native : change Makefile, then type 'make all' mTIM/src/sotool/losses : change Makefile, then type 'make all' mTIM/src/model : just type 'mex convert_states2labels.cpp' in your matlab command line

B) TESTING

• get the mini test example which is a driectory consisting of

  • acceptor and donor splice-sites
  • genome
  • annotation
  • rna-seq reads

• you have to manually adapt the path in mini.gio/genome.config

• you also need to adapt the paths in mTIM/src/mini_settings.m

• generate data by typing 'mTIM_prepare_data('mini')'

  • you can find the generated files under mTIM/out/mini

• training and prediction with 'mTIM_predict('mini')'

  • training output files are stored in mTIM/out/mini/{DATE}/ where the trained predictor is in mTIM/out/mini/{DATE}/xval_fold{X}/sosvm_final.mat
  • predicted annotation is in mTIM/out/mini/{DATE}/prediction.mat

MISC

• feature generation files are in /src/data_preparation
• viterbi is in /src/sotool/native/best_path.cpp