From 7b63a9e271af7b6ed18179a9f174e90519a1a552 Mon Sep 17 00:00:00 2001
From: Blaise Li <blaise.li__git@nsup.org>
Date: Mon, 2 May 2022 12:16:22 +0200
Subject: [PATCH] Optional codon influence in PCA.
---
libcodonusage/__init__.py | 2 +-
libcodonusage/libcodonusage.py | 87 ++++++++++++++++++++++++----------
2 files changed, 63 insertions(+), 26 deletions(-)
diff --git a/libcodonusage/__init__.py b/libcodonusage/__init__.py
index 4fcad87..a79d9a5 100644
--- a/libcodonusage/__init__.py
+++ b/libcodonusage/__init__.py
@@ -1,6 +1,6 @@
__copyright__ = "Copyright (C) 2022 Blaise Li"
__licence__ = "GNU GPLv3"
-__version__ = "0.25"
+__version__ = "0.26"
from .libcodonusage import (
aa2colour,
aa_usage,
diff --git a/libcodonusage/libcodonusage.py b/libcodonusage/libcodonusage.py
index 49d5c6b..3bb2a5e 100644
--- a/libcodonusage/libcodonusage.py
+++ b/libcodonusage/libcodonusage.py
@@ -674,10 +674,62 @@ methionine (M) and tryptophan (W).
all_nan_cols)
+def codon_influence_in_components(
+ components, colnames,
+ figs_dir=None, formats=None):
+ """
+ Plot the influence of the columns in the first 4 principal axes of a PCA.
+
+ Each column should correspond to a codon, and will be represented as a
+ colour bar whose coulour is based on the last letter of the codon.
+
+ *components* should be a numpy array representing the principal
+ axes of a PCA, such as the `components_` attribute of a fitted
+ `sklearn.decomposition.PCA` object.
+
+ *colnames* should be the names of the columns in the *components*
+ array and are expected to match the following pattern:
+ <aa>_<codon>, where <aa> is a single-letter code for an amino-acid,
+ and <codon> is one of the 3-letter codons for this amino-acid,
+ in capital letters among A, T, G and C (i.e. in the DNA alphabet).
+ The last letter will be used to set the colour of a bar in the plots
+
+ *figs_dir* should be a path to a directory that will be used
+ to save graphics representing the influence of each data column
+ on the first four principal components.
+
+ *formats* should be a list of formats in which the figures should
+ be saved, such as "svg" or "png".
+ """
+ render_md(
+ "Vizualizing the influence of codons in the first 4 components\n")
+ # TODO: *figsize* could be adapted depending on the number of columns
+ (fig, axes) = plt.subplots(4, 1, figsize=(16, 16))
+ for (component, axis) in enumerate(axes):
+ pd.Series(
+ components[component],
+ index=colnames).plot.bar(
+ ax=axes[component],
+ # colname is supposed to end with the 3-letters codon
+ color=[
+ nuc2colour[colname[-1]]
+ for colname in colnames])
+ axis.set_ylabel(f"weight in component {component}")
+ # axis.set_xticklabels(axis.get_xticklabels(), rotation=90)
+ fig.subplots_adjust(hspace=.5)
+ if figs_dir is not None and formats is not None:
+ for ext in formats:
+ plt.savefig(
+ figs_dir.joinpath(f"PCA_components.{ext}"),
+ metadata=fmt_metadata[ext])
+ display(fig)
+ plt.close(fig)
+
+
def codon_usage_pca(
usage_data,
figs_dir=None, hue="chrom", exclude_cols=None,
- formats=None):
+ formats=None, cols_are_codons=True):
"""
Perform Principal Component Analysis on *usage_data*.
@@ -696,9 +748,12 @@ def codon_usage_pca(
If *figs_dir* is not None, this path to a directory will be used
to save graphics representing the projection of the observations
- in the first four principal components (0 vs. 1 and 2 vs. 3)
- as well as graphics representing the influence of each data column
+ in the first four principal components (0 vs. 1 and 2 vs. 3).
+
+ Unless *cols_are_codons* is set to False, there will also be
+ graphics representing the influence of each data column
on the first four principal components.
+
*formats* should be a list of formats in which the figures should
be saved, such as "svg" or "png".
@@ -730,28 +785,10 @@ def codon_usage_pca(
metadata=fmt_metadata[ext])
display(fig)
plt.close(fig)
- render_md(
- "Vizualizing the influence of codons in the first 4 components\n")
- (fig, axes) = plt.subplots(4, 1, figsize=(16, 16))
- for (component, axis) in enumerate(axes):
- pd.Series(
- pca.components_[component],
- index=usage_data.columns).plot.bar(
- ax=axes[component],
- # colname is supposed to end with the 3-letters codon
- color=[
- nuc2colour[colname[-1]]
- for colname in usage_data.columns])
- axis.set_ylabel(f"weight in component {component}")
- # axis.set_xticklabels(axis.get_xticklabels(), rotation=90)
- fig.subplots_adjust(hspace=.5)
- if figs_dir is not None and formats is not None:
- for ext in formats:
- plt.savefig(
- figs_dir.joinpath(f"PCA_components.{ext}"),
- metadata=fmt_metadata[ext])
- display(fig)
- plt.close(fig)
+ if cols_are_codons:
+ codon_influence_in_components(
+ pca.components_, usage_data.columns,
+ figs_dir=figs_dir, formats=formats)
return (pca, transformed_data)
--
GitLab