feat: implement posterior prob filter for COLOC at small overlaps N<10 #977
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -208,11 +208,15 @@ class Coloc(ColocalisationMethodInterface): | |
|
|
||
| Attributes: | ||
| PSEUDOCOUNT (float): Pseudocount to avoid log(0). Defaults to 1e-10. | ||
| OVERLAP_SIZE_CUTOFF (int): Minimum number of overlapping variants bfore filtering. Defaults to 5. | ||
| POSTERIOR_CUTOFF (float): Minimum overlapping Posterior probability cutoff for small overlaps. Defaults to 0.5. | ||
| """ | ||
|
|
||
| METHOD_NAME: str = "COLOC" | ||
| METHOD_METRIC: str = "h4" | ||
| PSEUDOCOUNT: float = 1e-10 | ||
| OVERLAP_SIZE_CUTOFF: int = 5 | ||
| POSTERIOR_CUTOFF: float = 0.5 | ||
|
|
||
| @staticmethod | ||
| def _get_posteriors(all_bfs: NDArray[np.float64]) -> DenseVector: | ||
|
|
@@ -277,7 +281,15 @@ def colocalise( | |
| ) | ||
| .select("*", "statistics.*") | ||
| # Before summing log_BF columns nulls need to be filled with 0: | ||
| .fillna( | ||
| 0, | ||
| subset=[ | ||
| "left_logBF", | ||
| "right_logBF", | ||
| "left_posteriorProbability", | ||
| "right_posteriorProbability", | ||
| ], | ||
| ) | ||
| # Sum of log_BFs for each pair of signals | ||
| .withColumn( | ||
| "sum_log_bf", | ||
|
|
@@ -305,9 +317,18 @@ def colocalise( | |
| fml.array_to_vector(f.collect_list(f.col("right_logBF"))).alias( | ||
| "right_logBF" | ||
| ), | ||
| fml.array_to_vector( | ||
| f.collect_list(f.col("left_posteriorProbability")) | ||
| ).alias("left_posteriorProbability"), | ||
| fml.array_to_vector( | ||
| f.collect_list(f.col("right_posteriorProbability")) | ||
| ).alias("right_posteriorProbability"), | ||
| fml.array_to_vector(f.collect_list(f.col("sum_log_bf"))).alias( | ||
| "sum_log_bf" | ||
| ), | ||
| f.collect_list(f.col("tagVariantSource")).alias( | ||
| "tagVariantSourceList" | ||
| ), | ||
| ) | ||
| .withColumn("logsum1", logsum(f.col("left_logBF"))) | ||
| .withColumn("logsum2", logsum(f.col("right_logBF"))) | ||
|
|
@@ -327,10 +348,39 @@ def colocalise( | |
| # h3 | ||
| .withColumn("sumlogsum", f.col("logsum1") + f.col("logsum2")) | ||
| .withColumn("max", f.greatest("sumlogsum", "logsum12")) | ||
| .withColumn( | ||
| "anySnpBothSidesHigh", | ||
| f.aggregate( | ||
| f.transform( | ||
| f.arrays_zip( | ||
| fml.vector_to_array(f.col("left_posteriorProbability")), | ||
| fml.vector_to_array( | ||
| f.col("right_posteriorProbability") | ||
| ), | ||
| f.col("tagVariantSourceList"), | ||
| ), | ||
| # row["0"] = left PP, row["1"] = right PP, row["tagVariantSourceList"] | ||
| lambda row: f.when( | ||
| (row["tagVariantSourceList"] == "both") | ||
|
There was a problem hiding this comment. I think checking for
To do this you would have to make an overlap example with at least 2 variants on one side and 3 variants on the other side There was a problem hiding this comment. Hi @project-defiant , tests added, let me know if it addresses your concerns! |
||
| & (row["0"] > Coloc.POSTERIOR_CUTOFF) | ||
| & (row["1"] > Coloc.POSTERIOR_CUTOFF), | ||
| 1.0, | ||
| ).otherwise(0.0), | ||
|
Comment on lines
+365
to
+368
There was a problem hiding this comment. I am not 100% sure you are comparing 2 the same variants here (since they can be left or right oriented as well) |
||
| ), | ||
| f.lit(0.0), | ||
| lambda acc, x: acc + x, | ||
| ) | ||
| > 0, # True if sum of these 1.0's > 0 | ||
| ) | ||
| .filter( | ||
| (f.col("numberColocalisingVariants") > Coloc.OVERLAP_SIZE_CUTOFF) | ||
| | (f.col("anySnpBothSidesHigh")) | ||
| ) | ||
| .withColumn( | ||
| "logdiff", | ||
| f.when( | ||
| (f.col("sumlogsum") == f.col("logsum12")), | ||
| Coloc.PSEUDOCOUNT, | ||
| ).otherwise( | ||
| f.col("max") | ||
| + f.log( | ||
|
|
@@ -382,6 +432,10 @@ def colocalise( | |
| "lH2bf", | ||
| "lH3bf", | ||
| "lH4bf", | ||
| "left_posteriorProbability", | ||
| "right_posteriorProbability", | ||
| "tagVariantSourceList", | ||
| "anySnpBothSidesHigh", | ||
| ) | ||
| .withColumn("colocalisationMethod", f.lit(cls.METHOD_NAME)) | ||
| .join( | ||
|
|
||
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am not so sure that the converting to vectors and back is a correct way to handle the posteriorProbabilibies, I wish to understand the logic, why the
VectorUDTwas used initially here. @ireneisdoomed do you know the reason behind the vectorization of the logBF values (is it the sparsity or default numpy conversion?)