_PROBLEM CoEPrA-2006_Regression_003_Dataset_2

_GROUP_NAME Reiji Teramoto

_GROUP_MEMBERS
Reiji Teramoto

_ADDRESS
 Fundamental and Environmental Research Laboratories, NEC
Corporatopn.

_MODELING_PROCEDURE
  We used marginalized count kernels for amino acid
sequences to generate
  new descriptors and suppor vector regression as a
learning machine.[1,2].
  To select descriptors and calibrate model, we performed
10-fold 
  cross-validation via tuning types of marginalized count
kernel,  
  gaussian kernel parameter g, coefficient of slack
variables C and 
  epsilon-insensitive p using the calibration dataset.
  We examined 6 types of marginalized count kernel as
follows:
  1. 1st-order maginalized count kernel with hidden markov
models that hidden states is 3(abbreviation MCK1(h=3)).
  2. 2nd-order maginalized count kernel with hidden markov
models that hidden states is 3(abbreviation MCK2(h=3)).
  3. 1st-order maginalized count kernel with hidden markov
models that hidden states is 3(abbreviation MCK1(h=4)).
  4. 2nd-order maginalized count kernel with hidden markov
models that hidden states is 4(abbreviation MCK2(h=4)).
  5. 1st-order maginalized count kernel with hidden markov
models that hidden states is 5(abbreviation MCK1(h=5)).
  6. 2nd-order maginalized count kernel with hidden markov
models that hidden states is 5(abbreviation MCK2(h=5)).

  We examined gaussian kernel parameter g defined as
follows:
  K(u,v) = exp(-g*|u-v|^2)

  We determined the final model based on the best
performance of 10-fold cross-validation as follows:
  MCK1(h=5): c=64.0, g=0.5, p=0.5, mean squared
error=0.0.53697. 

1. Tsuda K, Kin T, Asai K., Marginalized kernels for
biological sequences, Bioinformatics. 2002;18 Suppl
1:S268-75. 
2. Vapnik V, The Nature of Statistical Learning Theory,
Springer, 1999.

_PREDICTION
Obj_00001 7.397
Obj_00002 7.959
Obj_00003 7.382
Obj_00004 7.771
Obj_00005 7.612
Obj_00006 7.995
Obj_00007 6.718
Obj_00008 7.214
Obj_00009 7.153
Obj_00010 7.621
Obj_00011 6.561
Obj_00012 6.291
Obj_00013 6.618
Obj_00014 7.087
Obj_00015 6.885
Obj_00016 6.960
Obj_00017 6.028
Obj_00018 6.749
Obj_00019 7.014
Obj_00020 6.669
Obj_00021 6.957
Obj_00022 7.148
Obj_00023 7.148
Obj_00024 7.729
Obj_00025 6.739
Obj_00026 6.873
Obj_00027 6.909
Obj_00028 7.212
Obj_00029 7.148
Obj_00030 7.655
Obj_00031 7.067
Obj_00032 7.387
Obj_00033 6.901
Obj_00034 7.358
Obj_00035 6.556
Obj_00036 6.555
Obj_00037 6.739
Obj_00038 6.958
Obj_00039 7.144
Obj_00040 6.663
Obj_00041 6.921
Obj_00042 7.344
Obj_00043 7.450
Obj_00044 6.743
Obj_00045 7.394
Obj_00046 6.888
Obj_00047 7.477