Hi
the INFER.NET engine seems to be very promising in my field of research (plant genetics) and I would love to get on board but have some start up issues.
In a pilot project I produced some code (in VB) to generate plants and (by linking them up) a pedigree structure.
I currently use recursive code to obtain the expected average fraction of a certain founding plant in each of my derived progeny. I would like to replace this (point) estimate with a proper INFER.NET posterior distribution, setting prior values (.observedvalue?)
to the founders and deriving posterior probabilities for mean and variance for the expected fraction of founder genome at each of the progeny.
It seems to me this must be possible using infer.net ; the probability to derive genome from a parent could be defined as a bernoulli(0.5)? but I am new to infer.net and need some help to get started.
Can someone give me some input how to implement this in a way similar to my getIBDwithTarget function (see code below)?
Thanks a lot
Ralph van Berloo
Imports MicrosoftResearch.Infer
Imports MicrosoftResearch.Infer.Models
Imports MicrosoftResearch.Infer.Distributions
Public Class Plant
Property Parent1 As Plant
Property Parent2 As Plant
Property Name As String
Sub New(ByVal parent1 As Plant, ByVal parent2 As Plant, ByVal name As String)
_Parent1 = parent1
_Parent2 = parent2
_Name = name
End Sub
Function IsFounder() As Boolean
Return (_Parent1 Is Nothing AndAlso _Parent2 Is Nothing)
End Function
End Class
Public Class DeriveIBD
' interest is in expectation AND variance for fraction of genome derived from Founder1
Dim founder1 As New Plant(Nothing, Nothing, "Founder1")
Dim founder2 As New Plant(Nothing, Nothing, "Founder2")
Dim F1 As New Plant(founder1, founder2, "F1") ' so expected fraction = 0.5
Dim BackCross1 As New Plant(F1, founder1, "BackCross1") ' so expected fraction = 0.75
Dim BackCross2 As New Plant(BackCross1, founder1, "BackCross2") ' so expected fraction = 0.875
Dim F2 As New Plant(F1, F1, "F2") ' expected fraction = again 0.5 but with larger variance
Sub Main()
Dim F1_IBD As Double = getIBDwithTarget(F1, founder1)
Dim BackcrossIBD As Double = getIBDwithTarget(BackCross1, founder1)
Dim Backcross1IBD As Double = getIBDwithTarget(BackCross1, founder1)
Dim Backcross2IBD As Double = getIBDwithTarget(BackCross2, founder1)
Dim F2_IBD As Double = getIBDwithTarget(F2, founder1)
Console.WriteLine("IBD of different progeny with founder 1:")
Console.WriteLine("F1: " + F1_IBD.ToString)
Console.WriteLine("BackCross1: " + Backcross1IBD.ToString)
Console.WriteLine("BackCross2: " + Backcross2IBD.ToString)
Console.WriteLine("F2: " + F2_IBD.ToString)
End Sub
Function getIBDwithTarget(ByVal tester As Plant, ByVal reference As Plant) As Double ' to be replaced with a Discrete?
Dim result As Double ' Discrete?
If tester.IsFounder Then
If tester Is reference Then 'in case our tester is a founder it can either be the reference or not
result = 1
Else
result = 0
End If
Else
' in case tester is not a founder it will have received half of its genome from either parent
result = 0.5 * getIBDwithTarget(tester.Parent1, reference) + 0.5 * getIBDwithTarget(tester.Parent2, reference)
End If
Return result
End Function
End Class
Running this gives as output:
IBD of different progeny with founder 1:
F1: 0.5
BackCross1: 0.75
BackCross2: 0.875
F2: 0.5
So my code yields the expected answers, but as point estimates only. I would like to use proper distributions allowing me to sample these distributions (also in more complex pedigrees) and get estimates for variance, extremes and perform tests for exceeding
thresholds etc.
