WIP: To be continued when David and I will work on this

Author: gregorgorjanc

Notes_for_imprinting_work.txt

@@ -0,0 +1,129 @@
+Imprinting Work Restart Notes
+=============================
+
+Date: 2026-04-18
+
+Current setup
+-------------
+
+- Clean restart worktree:
+  `/Users/ggorjanc/Storages/GitBox/AlphaSimR/AlphaSimR-wt-imprinting`
+- Branch in that worktree:
+  `imprinting`
+- Base commit:
+  current `upstream/devel` at the time the worktree was created
+
+What was reviewed
+-----------------
+
+- David's old branch:
+  `https://github.com/david20011999/AlphaSimR/commits/imprinting/`
+- David -> Gregor PR:
+  `https://github.com/gregorgorjanc/AlphaSimR/pull/4`
+- Gregor fork branch:
+  `https://github.com/gregorgorjanc/AlphaSimR/tree/imprinting`
+- Upstream branch:
+  `https://github.com/gaynorr/AlphaSimR/tree/imprinting`
+- Upstream discussion and design record:
+  `https://github.com/gaynorr/AlphaSimR/pull/167`
+
+Main conclusions
+----------------
+
+- The old `imprinting` branches are not a clean place to continue work.
+- `upstream/imprinting` is far behind current `upstream/devel` and should be treated as historical reference, not as the active implementation branch.
+- `origin/imprinting` is effectively the same old line of work and also should be treated as reference only.
+- David's PR to Gregor contains later experiments such as `TraitADI`, `calcGenParamS()`, and `getGvS()`, but it still sits on the stale branch history.
+- The real unresolved technical problem is not the basic API surface for imprinting. It is the variance-component definition and orthogonal decomposition of additive, dominance, and imprinting effects.
+- The upstream discussion in PR #167 is the key design record. The important guidance there was:
+  - move toward regression-based orthogonalization of `bv`, `dd`, and imprinting deviation
+  - do not lock in male/female breeding-value semantics too early
+  - keep F1 / cross-mean prediction outside AlphaSimR for now
+  - keep the first renewed implementation narrow and avoid dragging in polyploids or epistasis immediately
+
+Recommended restart strategy
+----------------------------
+
+- Use the new `imprinting` worktree as the active restart.
+- Treat old code only as a quarry for selective porting.
+- Use `/Users/ggorjanc/Downloads/__TODO_AlphaSimR_ImprintingVarianceComponents.R`
+  as the mathematical scratchpad / oracle for the regression decomposition.
+- First implementation pass should focus on:
+  - tests for orthogonal regressors under arbitrary genotype frequencies
+  - `TraitAI`
+  - updated `calcGenParam()` logic for orthogonal decomposition
+  - matching `getGv()` behavior
+  - only then extend to `TraitADI`
+- Leave out, for now:
+  - polyploids
+  - epistasis-related imprinting combinations
+  - polished vignette work
+  - a final public interpretation of sex-specific breeding values
+
+Useful branch facts
+-------------------
+
+- Local working branch now is `imprinting`.
+- Remote branch status checked on 2026-04-18:
+  - `highlanderlab/imprinting`: does not currently exist
+  - `origin/imprinting`: already exists and points to the old imprinting history
+
+Commands to push this branch
+----------------------------
+
+Work from the imprinting worktree:
+
+```bash
+cd /Users/ggorjanc/Storages/GitBox/AlphaSimR/AlphaSimR-wt-imprinting
+git status --short --branch
+git log --oneline --max-count=5
+```
+
+Push to HighlanderLab for the first time:
+
+```bash
+git push highlanderlab imprinting:imprinting
+```
+
+Push to Gregor's fork under the same branch name:
+
+```bash
+git push origin imprinting:imprinting
+```
+
+Important note about Gregor's fork
+----------------------------------
+
+- `origin/imprinting` already exists on the old history.
+- Because this new local `imprinting` branch is a clean restart from `upstream/devel`, the plain push above will most likely be rejected as non-fast-forward.
+- If the intention is to replace the old `origin/imprinting` branch with this new restart branch, use:
+
+```bash
+git push --force-with-lease origin imprinting:imprinting
+```
+
+Safer alternative for Gregor's fork
+-----------------------------------
+
+If you want to keep the old `origin/imprinting` branch around for archaeology, push this restart under a new branch name instead:
+
+```bash
+git push origin imprinting:imprinting-restart
+```
+
+Suggested practical sequence
+----------------------------
+
+If the goal is to share the restart without losing the old branch immediately:
+
+```bash
+cd /Users/ggorjanc/Storages/GitBox/AlphaSimR/AlphaSimR-wt-imprinting
+git push highlanderlab imprinting:imprinting
+git push origin imprinting:imprinting-restart
+```
+
+If later you decide that `origin/imprinting` should be replaced:
+
+```bash
+git push --force-with-lease origin imprinting:imprinting
+```

__TODO_AlphaSimR_ImprintingVarianceComponents.R

@@ -0,0 +1,75 @@
+# Simulate genotype frequencies for 00, 01, 10, and 11 (maternal-paternal)
+freq = runif(4)
+freq = freq/sum(freq)
+# freq = c(0.25, 0.25, 0.25, 0.25)
+
+# Genotypic effects
+a = 1
+d = 0
+i = 0.5
+
+# Vector of genetic values
+g = c(-a, d+i, d-i, a)
+g = g - sum(freq*g) # Center
+
+## Genotype dosage
+x = c(0, 1, 1, 2)
+x = x - sum(freq*x) # Center
+
+## Breeding value regressor 
+# Genetic contribution of "a"
+x_a = c(-1, 0, 0, 1)
+x_a = x_a - sum(freq*x_a) # Center
+
+## Dominance deviation regressor
+# Genetic contribution of "d"
+x_d = c(0, 1, 1, 0)
+x_d = x_d - sum(freq*x_d) # Center
+
+# Regression coefficient from regressing x_d on x_a
+m = sum(freq*x_a*x_d) / sum(freq*x_a^2)
+
+# Construct orthogonal regressor using lack-of-fit from regression of x_d on x_a
+x_d = x_d - x_a*m
+
+# Check orthogonality (should be zero within numeric precision)
+sum(freq*x_a*x_d)
+
+## Imprinting regressor
+# Genetic contribution of "i"
+x_i = c(0, 1, -1, 0)
+x_i = x_i - sum(freq*x_i) # Center
+
+# Regression coefficient from regressing x_i on x_a
+m_a = sum(freq*x_a*x_i) / sum(freq*x_a^2)
+
+# Regression coefficient from regressing x_i on x_d
+m_d = sum(freq*x_d*x_i) / sum(freq*x_d^2)
+
+# Construct orthogonal regressor using lack-of-fit
+x_i = x_i - x_a*m_a - x_d*m_d
+
+# Check orthogonality (all should be zero within numeric precision)
+sum(freq*x_i*x_a)
+sum(freq*x_i*x_d)
+
+## Calculate variances
+
+# Additive genetic variance
+alpha = sum(freq*x_a*g) / sum(freq*x_a^2)
+bv = x_a*alpha
+sum(freq*bv^2) 
+
+# Dominance genetic variance
+beta = sum(freq*x_d*g) / sum(freq*x_d^2)
+dd = x_d*beta
+sum(freq*dd^2)
+
+# Imprinting genetic variance
+gamma = sum(freq*x_i*g) / sum(freq*x_i^2)
+id = x_i*gamma
+sum(freq*id^2) 
+
+
+
+