User Guide: Pulsar™ Weather Station (Columbia Weather Systems)
17 biscarini
1. Eucarpia 2015
Filippo Biscarini
Nelson Nazzicari, Marco Bink, Pere Arus, Maria Jose Aranzana, Ignazio Verde, Sabrina Micali, Thierry
Pascal, Benedicte Quilot-Turion, Patrick Lambert, Igor Pacheco Cruz, Daniele Bassi, and Laura Rossini
Modelling genome-
wide predictions in
peach: first results
and perspectives
FruitBreedomics International Conference
2. Motivation
A little background!
● Genome-enabled predictions: popular in human (e.g. disease
risk) and livestock (e.g. dairy cattle) genetics
● Relevant also in plant genetics: crops and trees
● Interest in peach breeding, too!
3. Plant material
populations and traits
● 11 populations from 4
sites
● 3 traits: fruit weight,
sugar content (Brix)
and titrable acidity
● max n. records per
population x trait
combination
● yellow/red: done/to-be-
done
4. Phenotypes
description
● wide phenotypic
variability
● range: 51.7 – 191.8
(fruit weight); 12.04 –
16.21 (sugar content);
7.78 – 16.50 (acidity)
● larger for fruit weight
and acidity, lower for
sugar content
● red: max values;
yellow: min values
σ
μ
5. Genotypes
description
● all populations genotyped with:
peach 9K SNP-chip
● different data editing (so far): call-
rate, monomorphic SNP
● n. SNP after editing
● low residual missing rate: 0.23 –
5.56%
σ
μ
6. Imputation
results
● “Beagle”
● Little imputation
experiment
● Complete datasets, MAF
0.25-0.45
● 10 replicates per
threshold
● Imputation accuracy:
decreases with amounts
of missing genotypes
● Full data: 0.62 – 0.97
● Worst-case scenario:
MB1.73 x EarlyGold:
5.5% missing genotypes
x 0.95 imputation
accuracy = 0.3% errors
7. Genomic predictions
Let's get to the meat!
● Use SNP genotypes to predict unobserved (e.g. future) performances
● Phenotypic data over multiple years (1 to 5) → repeated observations model
● A.k.a. “repeatability model”
● 5-fold cross-validation: used to estimate the accuracy of genomic predictions
● 10-50 repetitions, so far (depending on population-trait)
9. Genomic predictions
The model
Var( y)=ZGZ ' σa
2
+WIW ' σpe
2
+I σe
2
● G: matrix of estimated genomic relationships (IBS)
● À la Van Raden (2008)
● G-BLUP: BLUP model in which the genomic information was introduced
through the variance-covariance matrix
● ASREML (REML), BGLR (MCMC)
10. Genomic predictions
The model
● Heritability: h
2
=
σa
2
σa
2
+σpe
2
+σe
2
● Repeatability: rep=
σa
2
+σ pe
2
σa
2
+σpe
2
+σe
2
● Predictive ability: r=cor( y , ̂y) ̂yik=μ+̂yeark +̂ai+̂pei
14. Next steps
● Genomic predictions in peach trees are feasible
● Preliminary results: very variable!
● Sample size, specific crosses, n. of replicates, trait definition ...
● Methods and (bio)informatics pipelines ready: 100%!
● Align data: retrieve all genotypes and phenotypes, apply same editing
policies etc …: almost there!
● Same n. of repetitions all over
● Interpret results
● Start drafting paper