genal-python 1.5.0__tar.gz → 1.5.1__tar.gz

{genal_python-1.5.0 → genal_python-1.5.1}/.gitignore

@@ -1,14 +1,18 @@
+.DS_Store
 __pycache__/
+.pytest_cache/
+.coverage
+htmlcov/
 dist/
 .ipynb_checkpoints/
 ipynb_checkpoints/
 genal/.ipynb_checkpoints/
 test_data/
 cursor/
-tests/
 tmp_GENAL/
 docs/build/
 docs/_build/
 REVIEW_REPORT.md
 TASKS.md
 code_concatenated
+tests/

genal_python-1.5.0/README.md → genal_python-1.5.1/PKG-INFO

@@ -1,3 +1,35 @@
+Metadata-Version: 2.4
+Name: genal-python
+Version: 1.5.1
+Summary: A python toolkit for polygenic risk scoring and mendelian randomization.
+Author-email: Cyprien Rivier <riviercyprien@gmail.com>
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: GNU General Public License v3 or later (GPLv3+)
+Classifier: Operating System :: OS Independent
+License-File: LICENSE
+Requires-Dist: aiohttp>=3.7
+Requires-Dist: nest_asyncio>=1.5
+Requires-Dist: numpy>=1.17.3
+Requires-Dist: pandas>=1.0
+Requires-Dist: plotnine>=0.9
+Requires-Dist: psutil>=5.0
+Requires-Dist: requests>=2.0
+Requires-Dist: pyliftover>=0.4
+Requires-Dist: scikit_learn>=0.24
+Requires-Dist: scipy>=1.7,<1.13
+Requires-Dist: statsmodels>=0.13,<0.15
+Requires-Dist: tqdm>=4.38
+Requires-Dist: wget>=3.0
+Requires-Dist: fastparquet>=0.4
+Requires-Dist: pyarrow>=3.0
+Requires-Dist: pytest>=7.0 ; extra == "test"
+Requires-Dist: pytest-cov ; extra == "test"
+Requires-Dist: pytest-xdist ; extra == "test"
+Project-URL: Home, https://github.com/CypRiv/genal
+Provides-Extra: test
+
 <div align="center">
   <img src="genal_logo.png" height="80" alt="genal logo" />
   <h1>genal</h1>
@@ -123,7 +155,7 @@ G_instruments.query_outcome(G_outcome, proxy=True, reference_panel="EUR_37")
 mr_results = G_instruments.MR(action=2, heterogeneity=True, odds=False)
 
 # 5.5) Plot MR results
-G_instruments.MR_plot(filename="mr_scatter")
+G_instruments.MR_plot(filename="mr_scatter", figure_size=(10, 6))
 ```
 ## Core concept: the `Geno` object
 
@@ -133,6 +165,7 @@ G_instruments.MR_plot(filename="mr_scatter")
 - `G.phenotype`: stored after `G.set_phenotype(...)` (phenotype DataFrame + metadata)
 - `G.MR_data`: stored after `G.query_outcome(...)` (exposure/outcome association tables used by MR)
 - `G.MR_results`: stored after `G.MR(...)` (results table + harmonized SNP table; used by plotting)
+- `G.MRpresso_subset_data`: stored after `G.MRpresso(...)` (outlier-removed harmonized table)
 
 Most methods either:
 - return a **new `Geno`** object (e.g., `clump()`), or
@@ -147,6 +180,7 @@ Most methods either:
 - **Align rsIDs to a target genotype dataset**: `Geno.update_snpids(path=..., replace=...)`
 - **Extract genotype subset**: `Geno.extract_snps(path=...)` → writes extracted files under `tmp_GENAL/`
 - **Two-sample MR pipeline**: `Geno.query_outcome(...)` → `Geno.MR(...)` (+ `MR_plot`)
+- **Leave-one-out MR**: `Geno.MR_loo(...)` → `Geno.MR_loo_plot(...)` (identify influential variants)
 - **MR-PRESSO**: `Geno.MRpresso(...)` (parallel; outlier + distortion tests)
 - **Colocalization**: `Geno.colocalize(...)` (approx Bayes factors; returns posterior probabilities)
 - **Association testing (individual-level)**: `Geno.set_phenotype(...)` → `Geno.association_test(...)`
@@ -298,7 +332,17 @@ About `action` (palindromic SNP handling during harmonization):
 Plot the MR scatter:
 
 ```python
-G_clumped.MR_plot(filename="mr_scatter")
+G_clumped.MR_plot(filename="mr_scatter", figure_size=(10, 6))
+```
+
+You can also draw a funnel plot of single-SNP ratio estimates (Wald ratios):
+
+```python
+G_clumped.MR_funnel(
+    methods=["IVW", "WM", "Egger"],  # vertical reference lines (optional)
+    filename="mr_funnel",
+    figure_size=(10, 6),
+)
 ```
 
 ### 6) Sensitivity: MR-PRESSO
@@ -311,7 +355,16 @@ mod_table, GlobalTest, OutlierTest, BiasTest = G_clumped.MRpresso(
     cpus=-1,   # use all CPU cores
 )
 ```
-If you want to rerun all MR methods after removing outliers with MR-PRESSO, you can use the `use_mrpresso_data=True` argument in `MR()`:
+
+To highlight MR-PRESSO outliers on plots, pass `use_mrpresso_data=True` (outliers are colored in red):
+
+```python
+G_clumped.MR_plot(filename="mr_scatter_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_clumped.MR_funnel(filename="mr_funnel_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_clumped.MR_loo_plot(filename="loo_forest_mrpresso", figure_size=(10, 8), use_mrpresso_data=True)
+```
+
+If you want to rerun MR methods after removing outliers with MR-PRESSO, you can use the `use_mrpresso_data=True` argument in `MR()`:
 ```python
 res = G_clumped.MR(
     action=2,
@@ -325,7 +378,29 @@ res = G_clumped.MR(
 res
 ```
 
-### 7) Single-SNP association tests (individual-level data)
+### 7) Sensitivity: Leave-One-Out MR
+Leave-one-out MR helps identify influential variants that may be driving the causal estimate.
+
+```python
+# Run leave-one-out analysis (default uses IVW)
+loo_results = G_clumped.MR_loo(method="IVW", heterogeneity=False, odds=False)
+```
+
+Visualize the results with a forest plot:
+
+```python
+# Default: show top influential instruments
+G_clumped.MR_loo_plot(filename="loo_forest", figure_size=(10, 8))
+```
+
+Tips:
+- `MR_loo` accepts the same `action`, `use_mrpresso_data`, and method parameters as `MR`.
+- `MR_loo_plot` supports `top_influential=True` (default) for a compact figure showing the most influential SNPs, or `top_influential=False` for paginated output with all instruments.
+- `MR_loo_plot(..., use_mrpresso_data=True)` colors MR-PRESSO outliers in red (requires running `MRpresso()` first). When outliers exist, an extra summary row ("MR-PRESSO corrected") is added using the same MR method as the leave-one-out analysis.
+- `MR_loo_plot(..., methods=["WM", "Egger"])` adds extra overall estimates for the requested methods (computed on all instruments).
+- Set `odds=True` in `MR_loo` if you want odds ratio scaling on the plot.
+
+### 8) Single-SNP association tests (individual-level data)
 
 Use individual-level data to re-estimate SNP–trait effects in a specific cohort (e.g., different ancestry, different measurement protocol).
 
@@ -354,7 +429,7 @@ G_adj.association_test(
 
 This updates `G_adj.data[["BETA","SE","P"]]` with cohort-specific estimates and recomputes `FSTAT` to be consistent with the updated values.
 
-### 8) Lift to a different build
+### 9) Lift to a different build
 
 Lift variants between builds (e.g., hg19 → hg38):
 
@@ -365,7 +440,7 @@ lifted.head()
 
 For large datasets, you can provide a UCSC LiftOver executable via `liftover_path`.
 
-### 9) Query the GWAS Catalog
+### 10) Query the GWAS Catalog
 
 Attach a per-SNP list of associated traits using the GWAS Catalog API:
 
@@ -520,3 +595,4 @@ If you use methods derived from other packages (e.g., MR-PRESSO), please also ci
 
 ## License
 GPL-3.0-or-later (see `LICENSE`).
+

genal_python-1.5.0/PKG-INFO → genal_python-1.5.1/README.md

@@ -1,31 +1,3 @@
-Metadata-Version: 2.4
-Name: genal-python
-Version: 1.5.0
-Summary: A python toolkit for polygenic risk scoring and mendelian randomization.
-Author-email: Cyprien Rivier <riviercyprien@gmail.com>
-Requires-Python: >=3.8
-Description-Content-Type: text/markdown
-Classifier: Programming Language :: Python :: 3
-Classifier: License :: OSI Approved :: GNU General Public License v3 or later (GPLv3+)
-Classifier: Operating System :: OS Independent
-License-File: LICENSE
-Requires-Dist: aiohttp>=3.7
-Requires-Dist: nest_asyncio>=1.5
-Requires-Dist: numpy>=1.17.3
-Requires-Dist: pandas>=1.0
-Requires-Dist: plotnine>=0.9
-Requires-Dist: psutil>=5.0
-Requires-Dist: requests>=2.0
-Requires-Dist: pyliftover>=0.4
-Requires-Dist: scikit_learn>=0.24
-Requires-Dist: scipy>=1.7,<1.13
-Requires-Dist: statsmodels>=0.13,<0.15
-Requires-Dist: tqdm>=4.38
-Requires-Dist: wget>=3.0
-Requires-Dist: fastparquet>=0.4
-Requires-Dist: pyarrow>=3.0
-Project-URL: Home, https://github.com/CypRiv/genal
-
 <div align="center">
   <img src="genal_logo.png" height="80" alt="genal logo" />
   <h1>genal</h1>
@@ -151,7 +123,7 @@ G_instruments.query_outcome(G_outcome, proxy=True, reference_panel="EUR_37")
 mr_results = G_instruments.MR(action=2, heterogeneity=True, odds=False)
 
 # 5.5) Plot MR results
-G_instruments.MR_plot(filename="mr_scatter")
+G_instruments.MR_plot(filename="mr_scatter", figure_size=(10, 6))
 ```
 ## Core concept: the `Geno` object
 
@@ -161,6 +133,7 @@ G_instruments.MR_plot(filename="mr_scatter")
 - `G.phenotype`: stored after `G.set_phenotype(...)` (phenotype DataFrame + metadata)
 - `G.MR_data`: stored after `G.query_outcome(...)` (exposure/outcome association tables used by MR)
 - `G.MR_results`: stored after `G.MR(...)` (results table + harmonized SNP table; used by plotting)
+- `G.MRpresso_subset_data`: stored after `G.MRpresso(...)` (outlier-removed harmonized table)
 
 Most methods either:
 - return a **new `Geno`** object (e.g., `clump()`), or
@@ -175,6 +148,7 @@ Most methods either:
 - **Align rsIDs to a target genotype dataset**: `Geno.update_snpids(path=..., replace=...)`
 - **Extract genotype subset**: `Geno.extract_snps(path=...)` → writes extracted files under `tmp_GENAL/`
 - **Two-sample MR pipeline**: `Geno.query_outcome(...)` → `Geno.MR(...)` (+ `MR_plot`)
+- **Leave-one-out MR**: `Geno.MR_loo(...)` → `Geno.MR_loo_plot(...)` (identify influential variants)
 - **MR-PRESSO**: `Geno.MRpresso(...)` (parallel; outlier + distortion tests)
 - **Colocalization**: `Geno.colocalize(...)` (approx Bayes factors; returns posterior probabilities)
 - **Association testing (individual-level)**: `Geno.set_phenotype(...)` → `Geno.association_test(...)`
@@ -326,7 +300,17 @@ About `action` (palindromic SNP handling during harmonization):
 Plot the MR scatter:
 
 ```python
-G_clumped.MR_plot(filename="mr_scatter")
+G_clumped.MR_plot(filename="mr_scatter", figure_size=(10, 6))
+```
+
+You can also draw a funnel plot of single-SNP ratio estimates (Wald ratios):
+
+```python
+G_clumped.MR_funnel(
+    methods=["IVW", "WM", "Egger"],  # vertical reference lines (optional)
+    filename="mr_funnel",
+    figure_size=(10, 6),
+)
 ```
 
 ### 6) Sensitivity: MR-PRESSO
@@ -339,7 +323,16 @@ mod_table, GlobalTest, OutlierTest, BiasTest = G_clumped.MRpresso(
     cpus=-1,   # use all CPU cores
 )
 ```
-If you want to rerun all MR methods after removing outliers with MR-PRESSO, you can use the `use_mrpresso_data=True` argument in `MR()`:
+
+To highlight MR-PRESSO outliers on plots, pass `use_mrpresso_data=True` (outliers are colored in red):
+
+```python
+G_clumped.MR_plot(filename="mr_scatter_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_clumped.MR_funnel(filename="mr_funnel_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_clumped.MR_loo_plot(filename="loo_forest_mrpresso", figure_size=(10, 8), use_mrpresso_data=True)
+```
+
+If you want to rerun MR methods after removing outliers with MR-PRESSO, you can use the `use_mrpresso_data=True` argument in `MR()`:
 ```python
 res = G_clumped.MR(
     action=2,
@@ -353,7 +346,29 @@ res = G_clumped.MR(
 res
 ```
 
-### 7) Single-SNP association tests (individual-level data)
+### 7) Sensitivity: Leave-One-Out MR
+Leave-one-out MR helps identify influential variants that may be driving the causal estimate.
+
+```python
+# Run leave-one-out analysis (default uses IVW)
+loo_results = G_clumped.MR_loo(method="IVW", heterogeneity=False, odds=False)
+```
+
+Visualize the results with a forest plot:
+
+```python
+# Default: show top influential instruments
+G_clumped.MR_loo_plot(filename="loo_forest", figure_size=(10, 8))
+```
+
+Tips:
+- `MR_loo` accepts the same `action`, `use_mrpresso_data`, and method parameters as `MR`.
+- `MR_loo_plot` supports `top_influential=True` (default) for a compact figure showing the most influential SNPs, or `top_influential=False` for paginated output with all instruments.
+- `MR_loo_plot(..., use_mrpresso_data=True)` colors MR-PRESSO outliers in red (requires running `MRpresso()` first). When outliers exist, an extra summary row ("MR-PRESSO corrected") is added using the same MR method as the leave-one-out analysis.
+- `MR_loo_plot(..., methods=["WM", "Egger"])` adds extra overall estimates for the requested methods (computed on all instruments).
+- Set `odds=True` in `MR_loo` if you want odds ratio scaling on the plot.
+
+### 8) Single-SNP association tests (individual-level data)
 
 Use individual-level data to re-estimate SNP–trait effects in a specific cohort (e.g., different ancestry, different measurement protocol).
 
@@ -382,7 +397,7 @@ G_adj.association_test(
 
 This updates `G_adj.data[["BETA","SE","P"]]` with cohort-specific estimates and recomputes `FSTAT` to be consistent with the updated values.
 
-### 8) Lift to a different build
+### 9) Lift to a different build
 
 Lift variants between builds (e.g., hg19 → hg38):
 
@@ -393,7 +408,7 @@ lifted.head()
 
 For large datasets, you can provide a UCSC LiftOver executable via `liftover_path`.
 
-### 9) Query the GWAS Catalog
+### 10) Query the GWAS Catalog
 
 Attach a per-SNP list of associated traits using the GWAS Catalog API:
 
@@ -548,4 +563,3 @@ If you use methods derived from other packages (e.g., MR-PRESSO), please also ci
 
 ## License
 GPL-3.0-or-later (see `LICENSE`).
-

{genal_python-1.5.0 → genal_python-1.5.1}/docs/source/concepts.md

@@ -10,6 +10,7 @@ Key attributes (you don't need to manipulate these directly):
 - `G.phenotype`: set by {py:meth}`genal.Geno.set_phenotype` (phenotype DataFrame + metadata)
 - `G.MR_data`: set by {py:meth}`genal.Geno.query_outcome` (exposure/outcome tables used by MR)
 - `G.MR_results`: set by {py:meth}`genal.Geno.MR` (results table + harmonized SNP table; used by plotting)
+- `G.MR_loo_results`: set by {py:meth}`genal.Geno.MR_loo` (leave-one-out results tuple; used by `MR_loo_plot`)
 - `G.MRpresso_results` / `G.MRpresso_subset_data`: set by {py:meth}`genal.Geno.MRpresso`
 
 ## Standard columns
@@ -38,7 +39,7 @@ This is a *practical guide*, not an exhaustive contract. When a method can work
 | {py:meth}`genal.Geno.clump` | `SNP`, `P` | LD clumping via PLINK; returns a new `Geno` (or `None` if nothing passes). |
 | {py:meth}`genal.Geno.prs` | `EA`, `BETA`, plus `SNP (or CHR+POS)` | If `CHR+POS` are available, genal will prefer position-based matching to your genotype dataset to reduce ID-mismatch losses. |
 | {py:meth}`genal.Geno.query_outcome` | `SNP`, `EA`, `NEA`, `BETA`, `SE` (exposure and outcome) | Outcome querying is rsID-based; proxy search is optional. If you plan to use `action=2` later, `EAF` in both datasets is strongly recommended. |
-| {py:meth}`genal.Geno.MR` / {py:meth}`genal.Geno.MRpresso` | `MR_data` | Both consume `MR_data` produced by `query_outcome()`. |
+| {py:meth}`genal.Geno.MR` / {py:meth}`genal.Geno.MR_loo` / {py:meth}`genal.Geno.MRpresso` | `MR_data` | All consume `MR_data` produced by `query_outcome()`. |
 | {py:meth}`genal.Geno.colocalize` | `BETA`, `SE`, plus `CHR+POS` (preferred) **or** `SNP` (in both datasets) | If `EA/NEA` are present in both datasets, effects are allele-aligned; otherwise results assume both GWAS use the same reference allele. For quantitative traits, provide `sdY` or (`EAF` + `n`) to avoid the default `sdY=1` assumption. |
 | {py:meth}`genal.Geno.update_eaf` | `EA`, plus `CHR+POS` **or** `SNP` | Uses PLINK to compute allele frequencies from a reference panel; coordinate-based matching is faster when available. |
 | {py:meth}`genal.Geno.filter_by_gene` / {py:meth}`genal.Geno.lift` | `CHR`, `POS` | Genomic coordinate operations. |
@@ -65,8 +66,11 @@ A helpful mental framework:
 | `association_test()` | `None` | runs PLINK `--glm`; mutates `G.data` (`BETA/SE/P`) |
 | `query_outcome()` | `None` | sets `G.MR_data` (exposure/outcome tables used by MR) |
 | `MR()` | `pd.DataFrame` | sets `G.MR_results` and returns the results table |
-| `MR_plot()` | plot object | requires `G.MR_results`; writes `.png` if `filename=...` |
-| `MRpresso()` | tuple | sets `G.MRpresso_results` and `G.MRpresso_subset_data` (outlier-removed harmonized table) |
+| `MR_plot()` | plot object | requires `G.MR_results`; writes `.png` if `filename=...`; supports `use_mrpresso_data=True` for outlier highlighting |
+| `MR_funnel()` | plot object | requires `G.MR_results`; writes `.png` if `filename=...`; supports `use_mrpresso_data=True` for outlier highlighting |
+| `MR_loo()` | `pd.DataFrame` | sets `G.MR_loo_results` and returns the LOO results table |
+| `MR_loo_plot()` | plot object(s) | requires `G.MR_loo_results`; writes `.png` if `filename=...`; may return a list for multi-page output; supports `methods=[...]` overall rows and `use_mrpresso_data=True` for outlier highlighting |
+| `MRpresso()` | tuple | sets `G.MRpresso_results` and `G.MRpresso_subset_data` (outlier-removed harmonized table; SNP-indexed) |
 | `prs()` | `None` | writes `<name>.csv` and uses PLINK temp files |
 | `query_gwas_catalog()` | `pd.DataFrame` | adds an `ASSOC` column (network-bound); `replace=True` overwrites `G.data` |
 | `filter_by_gene(replace=False)` | `Geno` | returns a new `Geno` filtered to a locus |
@@ -80,7 +84,7 @@ Be aware of these common side effects:
 
 - `~/.genal/config.json` is created/updated as you configure PLINK, reference folders, or default genotype paths.
 - `tmp_GENAL/` is used as a scratch directory for PLINK commands and is **not** automatically deleted.
-- Some methods generate output files in your current directory (notably `prs()`, and plot saving in `MR_plot()`).
+- Some methods generate output files in your current directory (notably `prs()`, and plot saving in `MR_plot()`, `MR_funnel()`, and `MR_loo_plot()`).
 
 ## Resource usage (`ram`, `cpus`)
 

{genal_python-1.5.0 → genal_python-1.5.1}/docs/source/methods.md

@@ -132,6 +132,30 @@ The bandwidth uses a modified Silverman rule multiplied by the user-provided fac
 
 The sign method tests whether exposure and outcome effects tend to have the same sign across variants. `genal` performs a binomial test against the null of 50% sign agreement.
 
+## Leave-one-out MR
+
+Implementation: {py:func}`genal.MR_tools.MR_loo_func`, wrapped by {py:meth}`genal.Geno.MR_loo`.
+
+Leave-one-out MR iterates over all instruments, sequentially removing each SNP and re-estimating the causal effect using the remaining instruments. This identifies variants that have a disproportionate influence on the overall estimate.
+
+For each SNP $i$ in the instrument set:
+
+1. Remove SNP $i$ from the harmonized data.
+2. Re-run the selected MR method on the remaining $J-1$ SNPs.
+3. Store the resulting estimate $\hat{\theta}_{-i}$.
+
+The "influence" of SNP $i$ is defined as:
+
+```{math}
+\text{influence}_i = \left| \hat{\theta}_{-i} - \hat{\theta}_{\text{all}} \right|
+```
+
+where $\hat{\theta}_{\text{all}}$ is the estimate using all instruments.
+
+Notes:
+- Any single MR method can be used (IVW, Egger, weighted median, mode-based, etc.).
+- At least 3 instruments are required (so that each LOO subset has ≥2 instruments).
+
 ## MR-PRESSO (summary)
 
 Implementation: {py:func}`genal.MRpresso.mr_presso`, wrapped by {py:meth}`genal.Geno.MRpresso`.
@@ -144,11 +168,26 @@ At a high level, `genal`'s MR-PRESSO implementation:
   - an outlier test (per-variant p-values, Bonferroni-corrected),
   - an optional distortion test (whether the causal estimate changes materially after removing outliers).
 
+### Distortion test
+
+The distortion test assesses whether detected outliers materially bias the causal estimate. `genal` implements the following version:
+
+1. **Observed distortion**: $D_\text{obs} = (\hat{\theta}_\text{all} - \hat{\theta}_\text{no outliers}) / |\hat{\theta}_\text{no outliers}|$
+2. **Expected distortion null**: bootstrap resampling is performed *exclusively on the non-outlier subset*. For each iteration:
+   - Sample with replacement $J-k$ SNPs from the non-outlier data (where $k$ is the number of detected outliers).
+   - Fit the IVW model on the sampled data and record $\hat{\theta}_\text{exp}$.
+   - Compute $D_\text{exp} = (\hat{\theta}_\text{all} - \hat{\theta}_\text{exp}) / |\hat{\theta}_\text{exp}|$.
+3. **P-value**: $p = \text{mean}(|D_\text{exp}| > |D_\text{obs}|)$.
+
+This differs from the original MR-PRESSO R implementation, which in some cases samples from the full dataset (including outliers) for the expected-bias regressions and was inconsistent with the paper's description.
+
+### Output structure
+
 `Geno.MRpresso()` returns four objects:
 - `mod_table`: a small results table (`Raw` and `Outlier-corrected` rows; IVW model),
 - `GlobalTest`: RSS and global p-value,
-- `OutlierTest`: per-variant outlier p-values (empty if the global test is not significant),
-- `BiasTest`: distortion test result dictionary (empty if distortion test was not run).
+- `OutlierTest`: per-variant outlier p-values (empty if the global test is not significant); SNP IDs as row labels,
+- `BiasTest`: distortion test result dictionary containing `"outliers_indices"` (SNP IDs), `"distortion_test_coefficient"`, and `"distortion_test_p"` (empty if distortion test was not run).
 
 If outliers are found, `genal` stores the outlier-removed harmonized table and allows rerunning MR with `Geno.MR(use_mrpresso_data=True)`.
 

{genal_python-1.5.0 → genal_python-1.5.1}/docs/source/workflows.md

@@ -175,10 +175,20 @@ Key arguments you commonly tune:
 After `MR()`, you can generate a scatter plot with method lines:
 
 ```python
-G_instruments.MR_plot(filename="mr_scatter")  # saves mr_scatter.png
+G_instruments.MR_plot(filename="mr_scatter", figure_size=(10, 6))  # saves mr_scatter.png
 ```
 
-## 6) MR-PRESSO (outlier detection and distortion testing)
+You can also draw a funnel plot of single-SNP ratio estimates (Wald ratios):
+
+```python
+G_instruments.MR_funnel(
+    methods=["IVW", "WM", "Egger"],  # vertical reference lines (optional)
+    filename="mr_funnel",
+    figure_size=(10, 6),
+)
+```
+
+## 6a) MR-PRESSO (outlier detection and distortion testing)
 
 {py:meth}`genal.Geno.MRpresso` runs a parallel MR-PRESSO implementation.
 
@@ -196,14 +206,71 @@ What you typically tune:
 - `significance_p`: threshold for global/outlier tests.
 - `outlier_test` / `distortion_test`: disable if you only want the global test.
 
+Output structure:
+- `OutlierTest`: DataFrame with per-SNP outlier p-values; SNP identifiers (rsIDs) are used as row labels (not numeric indices).
+- `BiasTest`: dictionary containing `"outliers_indices"` (list of SNP IDs), `"distortion_test_coefficient"`, and `"distortion_test_p"`.
+
 If outliers are found, you can rerun MR using the outlier-removed subset:
 
 ```python
 res_no_outliers = G_instruments.MR(use_mrpresso_data=True)
 ```
 
+To highlight MR-PRESSO outliers on plots, pass `use_mrpresso_data=True` (outliers are colored in red and shown in the legend):
+
+```python
+G_instruments.MR_plot(filename="mr_scatter_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_instruments.MR_funnel(filename="mr_funnel_mrpresso", figure_size=(10, 6), use_mrpresso_data=True)
+G_instruments.MR_loo_plot(filename="loo_forest_mrpresso", figure_size=(10, 8), use_mrpresso_data=True)
+```
+
 See {doc}`methods` for algorithm details and outputs.
 
+## 6b) Leave-one-out MR (sensitivity analysis)
+
+{py:meth}`genal.Geno.MR_loo` iteratively removes each SNP and re-estimates the causal effect. This helps identify influential variants that may be driving the overall result.
+
+```python
+loo_df = G_instruments.MR_loo(
+    method="IVW",        # any single MR method key (see MR method map)
+    action=2,
+    heterogeneity=False, # set True to include Q statistics
+    odds=False,          # set True for OR-scale output
+)
+```
+
+Key arguments:
+- `method`: a single MR method key (e.g., `"IVW"`, `"Egger"`, `"WM"`); must not be `"all"`.
+- `use_mrpresso_data=True`: use the outlier-removed dataset from MR-PRESSO instead of all instruments.
+
+### Visualizing leave-one-out results
+
+{py:meth}`genal.Geno.MR_loo_plot` creates a forest plot from the stored `MR_loo_results`:
+
+```python
+# Default: show top influential instruments
+G_instruments.MR_loo_plot(filename="loo_forest", figure_size=(10, 8))
+```
+
+```python
+# Or paginate all instruments
+G_instruments.MR_loo_plot(
+    top_influential=False,  # show all, not just influential
+    snps_per_page=30,
+    page=1,                 # or None for all pages
+    filename="loo_forest_all",
+    figure_size=(10, 12),
+)
+```
+
+Key arguments:
+- `top_influential=True` (default): select the `snps_per_page` most influential SNPs (largest change in estimate when removed) and render a single compact figure.
+- `top_influential=False`: paginate all instruments; use `page=N` to select a specific page or `page=None` to render all pages.
+- `snps_per_page`: number of SNPs per page (minimum 5).
+- `use_mrpresso_data=True`: color MR-PRESSO outliers in red (requires `MRpresso()` first). When outliers exist, an extra summary row ("MR-PRESSO corrected") is added using the same MR method as the leave-one-out analysis.
+- `methods=["WM", "Egger"]`: add extra overall estimates for the requested methods (computed on all instruments).
+
+
 ## 7) Additional capabilities (beyond the core pipeline)
 
 ### Single-SNP association tests (individual-level data)