SeqMutPlot.mutation_landscape

SeqMutPlot.mutation_landscape(df_scan, entry=None, ax=None, figsize=(12, 6), cmap=None, class_names=None)

Plot the per-position mutation-scan heatmap for one sequence.

Rows are the 20 canonical amino acids, columns are the scanned positions (labelled by the wild-type residue and colored by sequence part — JMD-N / TMD / JMD-C). Each cell is the change the substitution induces: the model prediction shift delta_pred (in percentage points, diverging blue-white-red) when SeqMut.scan() was run with a bound model, otherwise the model-free delta_cpp. With a model, the title reports the wild-type prediction score.

Parameters:

• df_scan (pd.DataFrame) – Mutational landscape produced by SeqMut.scan().

• entry (str, optional) – Protein entry to plot. If None, the first entry in df_scan is used.

• ax (Axes, optional) – Pre-defined Axes object to plot on. If None, a new one is created.

• figsize (tuple, default=(12, 6)) – Figure dimensions (width, height) in inches (used when ax is None).

• cmap (str, optional) – Matplotlib colormap name. If None, a diverging 'bwr' is used for the signed delta_pred and 'viridis' for the non-negative delta_cpp.

• class_names (tuple of str, optional) – (negative, positive) class labels. When given (and a model was bound), the predicted wild-type class is appended to the title.

Returns:

• fig (Figure) – Figure object containing the plot.

• ax (Axes) – Axes object of the mutation-scan heatmap.

Notes

• Returned as a (fig, ax) pair (see SeqMutPlot for the shared return contract).

Examples

:meth:SeqMutPlot.mutation_landscape is the mutation-scan heatmap: the 20 amino-acid substitutions (rows) against the scanned positions (columns, labelled by the wild-type residue and colored by sequence part). When :meth:SeqMut.scan ran with a bound model, each cell is the model prediction shift delta_pred (diverging blue-white-red) and the title reports the wild-type prediction; otherwise the cells show the model-free delta_cpp.

import itertools
import pandas as pd
import matplotlib.pyplot as plt
import aaanalysis as aa
aa.options["verbose"] = False

# Data, CPP features, and a fitted TreeModel that scores each sequence
df_seq = aa.load_dataset(name="DOM_GSEC", n=10)
labels = df_seq["label"].to_list()
sf = aa.SequenceFeature()
df_parts = sf.get_df_parts(df_seq=df_seq)
split_kws = sf.get_split_kws()
df_scales = aa.load_scales()
cpp = aa.CPP(df_parts=df_parts, split_kws=split_kws, df_scales=df_scales, verbose=False)
df_feat = cpp.run(labels=labels, n_filter=25)
X = sf.feature_matrix(features=list(df_feat["feature"]), df_parts=df_parts, df_scales=df_scales)
tm = aa.TreeModel().fit(X, labels=labels)
entry = df_seq["entry"].iloc[0]
ts = int(df_seq.set_index("entry").loc[entry, "tmd_start"])

# Model-guided: per-position prediction shift (the mutation-scan heatmap)
df_scan = aa.SeqMut(model=tm).scan(df_seq=df_seq, df_feat=df_feat, region=None)
aa.plot_settings()
aa.SeqMutPlot().mutation_landscape(df_scan=df_scan, entry=entry, class_names=("NEG", "POS"),
                                   figsize=(12, 6))
plt.tight_layout()
plt.show()

/Users/stephanbreimann/Programming/1Packages/wt-seqmut-ml-guided/aaanalysis/feature_engineering/_backend/cpp_run.py:163: UserWarning: CPP is using the Python kernel fallback — the compiled Cython extension is not available in this install. Output is bit-exact with the Cython path but ~2x slower. Reinstall via pip install --force-reinstall aaanalysis to fetch a prebuilt wheel.
  warnings.warn(

Without a bound model the same method shows the model-free delta_cpp magnitude.

df_scan_free = aa.SeqMut().scan(df_seq=df_seq, df_feat=df_feat, region="tmd")
aa.plot_settings()
aa.SeqMutPlot().mutation_landscape(df_scan=df_scan_free, entry=entry, figsize=(10, 6))
plt.tight_layout()
plt.show()