S.N.U. EECS Jeong-Jin Lee Eui-Taik Na Locating exons S.N.U. EECS Jeong-Jin Lee Eui-Taik Na
Using Multilayer Neural Network Approach Part I. Using Multilayer Neural Network Approach
System Architecture Output Input Representation Learning Finding exon starting region and ending region Input Representation Orthogonal encoding Using windows [0, 0, 0, 0] ~ outside the sequences Learning Backpropagation
Parameters 1 Interpretations of inputs Case 1 - aligning the center - aligning the end Case 3, 4 - biased
Parameters 2 The size of window Number of units in the hidden layer
Experimental Results 1 Three datasets from UCSC Data set 1 Data set 2 Having 7 subsets Data set 1 3447 trainings, 251 evaluations Data set 2 3448 trainings, 250 evaluations Evaluation Score Sums of success hits’ percentages
Experimental Results 2 According to window type
Experimental Results 3 According to window size
Experimental Results 4 According to number of hidden units
Part 2 Using HMMs
Using HMMs The Tied Model The Wheel Model
Future considerations HMM + NN mixed models