Skip to main content

Table 3 Dissimilarity based partial least squares (DPLS) prediction results over all dataset in a 10-fold CV setup

From: Dissimilarity based Partial Least Squares (DPLS) for genomic prediction from SNPs

Distance

PQc (R2d)

RMSEa

Optimal LVsb

Euclidean

0.62 ± 0.005

370 ± 2.7

4

Gower

0.60 ± 0.0052

380 ± 2.8

6

Allele share

0.61 ± 0.005

380 ± 2.8

6

Nei

0.59 ± 0.005

390 ± 2.9

6

Bray

0.63 ± 0.004

370 ± 2.6

4

Jaccard

0.64 ± 0.0043

360 ± 2.8

4

Kulczynski

0.61 ± 0.0053

380 ± 2.8

4

GRM

0.62 ± 0.005

370 ± 2.9

5

GBLUP

0.61 ± 0.001

369.9 ± 0.66

NA

  1. All the results presented in table are significant (with respect to p-value computed from permutation analysis). The results are averaged over 10-fold CV scheme. The 10-fold CV procedure was repeated 50 times. The standard error (se) calculated over 10-fold CV repetition. The last row present prediction results obtained from GBLUP. The PQ (R2d), RMSE and LVs represents prediction quality, root mean square error and latent variables respectively
  2. a RMSE stands for root mean square error
  3. b LVs stands for latent variables used for model building
  4. c PQ represent prediction quality
  5. d R 2 presented in the table are estimated for testset and not from training model. The value is calculated in a cross validation setup (some time indicated as Q2). This value is refer as prediction quality in this study