solGS:  A  Web-­based  Solution  for  
Genomic  Selection
Isaak  Y  Tecle,  Naama  Menda,  Guillaume  
Bauchet,  Lukas  Mueller
Tecle  et  al.  Bioinformatics  2014,  15:398
Phenotyped  
&  
genotyped  individuals
Genomic  selection…
Prediction  model
Predicted  
breeding
Values  (GEBVs)
Genotyped  selection  
candidates
Training  population
Challenges…
n Data  volume,  storage
n Data  structuring,  cleaning,  imputation
n Statistical  analysis  complexity
n visualization  and  sharing
solGS  webtool
http://cassavabase.org/solgs
What  you  can  do  with  solGS…
n Store  data
n Chado  Natural  Diversity  schema
n Compose  training  populations
n Build  models  and  predict  breeding  
values  of  selection  candidates
n Test  model  accuracy  
What  you  can  do  with  solGS…
n Explore  phenotype  data,  population  
structure
n Check  on  relationship  between  GEBVs  
vs  observed  phenotypes
n Calculate  selection  indices,  correlation  
n Visualize  data  on  interactive  plots
What  is  the  statistical  approach  
behind  solGS?
…preparing  data
n Omits  individuals  completely  missing  
phenotype  values
n Adjusts  phenotype  values  for  block  
effects
n Averages  across  multiple  trials  after  
adjusting  for  block  effects
n Imputes  missing  marker  data
n Median  substitution
…statistical  modeling
n Univariate
n RR-­BLUP
n Endelman,  Plant  Genome  (2010)
n GBLUP  
n Marker-­based  realized  relationship  matrix
n Prediction  accuracy
n Based  on  10-­fold  cross-­validation
How  does  solGS  work?
Composing  a  training  population:  
Fitting  a  prediction  model...
3  options
Fitting  a  prediction  model…
Option  1:  
Search  using  a  trait  name
Estimating  breeding  values  of  
selection  candidates
Applying  the  model…
Fitting  a  prediction  model…
Option  2:  
Search  for  trials
Estimating  breeding  values  of  a  
selection  candidates  for  multiple  
traits
Applying  the  models…
Estimating  genetic  correlations
Calculating  selection  indices
Fitting  a  prediction  model…
Option  3:  
use  your  own  list  of  individuals
To  sum  up…
n Store  data
n Build  prediction  models
n Estimate  breeding  values
n Additional  analyses:  
n Correlation  analysis
n Population  structure
n Selection  indices
n http://cassavabase.org/solgs
n Open  source  code
Thanks  to…
Many  thanks!!
Background  image:  nextgencassava.org

Cassavabase SolGS presentation PAG 2016