Term
|
Definition
ΔBV/t = ( h2 * σp *i ) / L
*predicts the response to selection
*the chage in BV OR pheno Δ in a pop |
|
|
Term
|
Definition
| ACCURACY VS L *older ind have highest accuracy *w/ old bulls ↑ accuracy BUT ↑L INTENSITY VS GEN INTERVAL *Choosing mostly old bulls and few new would inc i but also inc L INTENSITY VS RISK *can control i by mating 1 bull to lots cows *if only keep new bulls, ↑i BUT but very risky (b/c new bulls have not been tested and could possibly prod low quality prog) |
|
|
Term
|
Definition
*predict ind BV using whatever info available (i.e mean prog pheno) but need to use accuracy to see how reliable our prediction
*predicting the genetic merit (BVs) using available info
*IMPORTANT B/C: better GP --> ↑h --> R (ΔBV)
if selecting ind based on BVs |
|
|
Term
| single source genetic prediction |
|
Definition
*uses trait parameters to measure behavior of a trait in a pop (h2, H2, rep)
*making a prediction from 1 source of info
Prediction = Regression (b) * info
Info = P-bar,P(p) [aka mean progeny pheno] |
|
|
Term
| Factors needed to make a prediction |
|
Definition
1) SLOPE OF RELATIONSHIP BETWEEN INFO SOURCE & VALUE BEING PREDICTED
*h2 is the slope of a specific prediction (ind BV from P)
*need a regression (b) ==> to make a prediction
2) STRENGTH/ ACCURACY OF RELATION BET INFO & VALUE BEING PREDICTED
*to quantify and understand accuracy of prediction
*ned a correlation (r) ==> to interpret a prediction |
|
|
Term
Visual representation of making a prediction
1) WHAT IS THE SLOPE
2) HOW TO PREDICT BV GIVEN INFO
3) WHAT IS THE ACCURACY/ WHAT IS OUR CONFIDENCE FOR THE PREDICTION |
|
Definition
1) given by the true BVs (yax) that come from info(xax) --> best line
2) use info given --> follow up till hit slope --> go left until read BV on y axis
3)ACCURACY OF PREDICTION- represented by how spread apart the pts are from the best fit line
*best prediction would be on the line ==> the higher the spread, the more errors in pred (↓h)
*if CI big, an ind may have a BV way diff than what was predicted |
|
|
Term
| 2 ways to solve for regression |
|
Definition
1) EMPIRICAL APPROACH
2) ANALYTICAL APPROACH |
|
|
Term
|
Definition
*generate lots of data (w/ lrg sample size) --> use to determine slope and accuracy
*use lots of data to find appropriate b for any # of progeny for a specific trait
ASSUMPTIONS:
*assume BV is true(b.c lots of info)
mean progeny pheno = true PD (so 2*P-bar = BV)
WHAT ARE THE BVs FOR THE BULLS
*use 2 * P-bar to find BV for each
WHAT IS THE APPROPRIATE REGRESS
*for e. bull, grab a random # of progeny
*now can find b b/c have BV and info |
|
|
Term
|
Definition
*using logic, algebra, etc to solve for the appropriate b
IF BULL HAD p PROGENY, HOW WOULD YOU PREDICT THE BV IN AN OPTIMAL WAY?
1) WHAT IS THE APPRPRAITE b FOR ____ PROGENY & h2=____?
*FIND THE SLOPE
* b = 2 [(p * h2) / (4 + (p-1)h2)]
2)HOW PREDICT THE BEST EST BV IF P-bar,P(p) =____?
*MAKE THE PREDICTION/ EST BV
* BV-cap = b * info
3) HOW ACCURATE IS THE EST OF IND BV?
* rbv,p = √ [ph2 / 4+(p-1)h2] |
|
|
Term
factors that influence b
(for empirical app) |
|
Definition
*depending on p and h2, b should be between 0-2
1) h2
* IF h2=0 ==> BV-cap=0 ==> b=0
*b/c h2=0 means no additive gentic variation for that trait --> so every ind in pop has BV=0 (no variation)
*there can still be pheno variation b/c E
2) # OF PROGENY
*IF h2 = 0, # PROG=∞ ==> P-bar=0
*P would only be influenced by E and overall P-bar=0 (noise and randomness goes away)
*PD is the truth ==> if have more progeny, move closer towards the true dev of the ind progeny from the pop mean
*IF h2>0, # PROG=∞ ==> BV-cap=2PD ==> b=2
*PD = P-bar (the men progeny pheno would be the true PD BV =2PD)
3) TYPE AND AMOUNT OF INFO
*so can rank ind based on BVs --> more h --> able to choose best ind --> better R |
|
|
Term
factors that influence b
(for analytical app) |
|
Definition
b= 2[(ph2)/(4+(p-1)h2)]
1) h2
*h2=0 ==> b=0
*b/c no matter mean prog pheno, we know its due to noise (not ind BV)
*if h2=0, no variation in BV
*0 in numerator
2) # PROGENY
*h2>0, p=∞ ==> b=2
*b/c as h2 OR #p ↑, the ratio for b approached 2
3) TYPE AND AMOUNT OF DATA
*when have a limited amount of data, attritbute some deviation to random effects b/c they could also create a + P-bar
*when have a limited # of ind and low h2, more noise contrib to dev, so less confident
*the more info (↑#prog, ↑h2) --> the more can be attributed to genetics itself as opposed to outside noise |
|
|
Term
| Sampling formulas from a single source of info for b and h |
|
Definition
* if you are trying to predict BV-cap from A SINGLE (NONREPEATED) PERFORMANCE RECORD ON THE IND, then b is h2 and accuracy is h
*if you are trying to predict BV-cap from THE AVERAGE OF SINGLE RECORDS ON p PROGENY, then b is 2ph2/4+(p-1)h2 and h is √[ph2 / 4+(p-1)h2] |
|
|
Term
|
Definition
68% COFIDENCE
CI = BV-cap ± √[(1-r^2)(σ^2bv)]
95% CONDIFDENCE
CI = BV-cap ± 2√[(1-r^2)(σ^2bv)] |
|
|
Term
| Forming a confidence range |
|
Definition
h2 = σ^2bv / σ^2p
WHAT WOULD THE CI INTERVAL BE/ WHAT IS THE RANGE OF POSSIBLE VALUES
*use IC formulas |
|
|
Term
|
Definition
*theres a relationship between h and infosource (correlation bet BV-cap)
*theres a relationship between h and CI for the estimate
*if know b and h for any partic type or amount of info, you can make and interpret estimates b/c can quantify the CI around the estimate |
|
|
Term
| Predict an ind BV from own pheno value |
|
Definition
*single source genetic prediction
* I = b * x
I: predicion
b: slope
x: own info (EX: pheno value) |
|
|
Term
| Multiple source genetic prediction |
|
Definition
*have mult sources and slopes --> sum them
I = b1x1 + b2x2 + ... + bnxn
ASSUMPTIONS:
*b depends on e.other ==> the larger thb the more wt being put on the source
*b for e ind source depends on other sources and amount of info
*when have diff amounts of info, diff wts are put on the info sources; some sources are more valuable than others |
|
|
Term
| Problem with MSGP (index that combines multiple traits) |
|
Definition
*an ind has multp Vs (one for e. trait) so how do you rank an ind if you give equal consid to all traits
SOLN: you can take mult BVs and come up with an index that ranks ind w/ combined estimates across mult traits (get 1 # while considering mult traits) |
|
|
Term
|
Definition
*how you rank ind in a quant way
*what is an ind overall merit when consider mult traits
H = VT1BV-capT1 + ...
V: relative economic value
PATTERNS
+BV and -V
*make product of the two - ==> ↓H
-BV and -V
*make the product of the two + ==> ↑H
*by having a - V you are still emphasizing ind w/ -BVs |
|
|
Term
| BVs with different magnitudes |
|
Definition
*if BVs are in diff magnitudes (cm, lbs, etc), fix it by normalizing --> convert them to SD and putting them all on same scale
1) CONVERT TO SD
* BV-cap / σ(whole) |
|
|
Term
|
Definition
*diff indexes put diff weights on traits depending on emphasis in selection
*for realtive wt, i values means emphasizing ind w/ low P (so take abs value to find most and least emphasized)
*if want to dec P-bar in a pop, choose - V |
|
|
