A minimal CPP analysis

The shortest complete loop in AAanalysis: load a dataset, run Comparative Physicochemical Profiling (CPP), and read out the physicochemical signature that distinguishes the two groups.

We use the domain-level γ-secretase dataset (DOM_GSEC: substrates vs. non-substrates), so this is the domain level row of the Prediction tasks table — the unit of comparison is the TMD part-set and the reference is built from labeled A-vs-B groups. For the broader tour (machine learning, SHAP feature impact, the comparison harness) see the Quick start tutorial.

import matplotlib.pyplot as plt
import numpy as np

import aaanalysis as aa
aa.options["verbose"] = False
aa.options["random_state"] = 42

1. Load a domain-level dataset

load_dataset returns a sequence table (df_seq) with the TMD bounds; load_scales returns the amino-acid scales CPP profiles over. The binary label column is the A-vs-B grouping CPP contrasts.

df_seq = aa.load_dataset(name="DOM_GSEC", n=50)
labels = df_seq["label"].to_list()
df_scales = aa.load_scales()
aa.display_df(df=df_seq, n_rows=10, show_shape=True)

DataFrame shape: (100, 8)
  entry sequence label tmd_start tmd_stop jmd_n tmd jmd_c
1 Q14802 MQKVTLGLLVFLAGF...PGETPPLITPGSAQS 0 37 59 NSPFYYDWHS LQVGGLICAGVLCAMGIIIVMSA KCKCKFGQKS
2 Q86UE4 MAARSWQDELAQQAE...SPKQIKKKKKARRET 0 50 72 LGLEPKRYPG WVILVGTGALGLLLLFLLGYGWA AACAGARKKR
3 Q969W9 MHRLMGVNSTAAAAA...AIWSKEKDKQKGHPL 0 41 63 FQSMEITELE FVQIIIIVVVMMVMVVVITCLLS HYKLSARSFI
4 P53801 MAPGVARGPTPYWRL...GLFKEENPYARFENN 0 97 119 RWGVCWVNFE ALIITMSVVGGTLLLGIAICCCC CCRRKRSRKP
5 Q8IUW5 MAPRALPGSAVLAAA...EVPATPVKRERSGTE 0 59 81 NDTGNGHPEY IAYALVPVFFIMGLFGVLICHLL KKKGYRCTTE
6 P01135 MVPSAGQLALFALGI...LLKGRTACCHSETVV 0 99 121 AVVAASQKKQ AITALVVVSIVALAVLIITCVLI HCCQVRKHCE
7 O43914 MGGLEPCSRLLLLPL...SDVYSDLNTQRPYYK 0 42 64 DCSCSTVSPG VLAGIVMGDLVLTVLIALAVYFL GRLVPRGRGA
8 P05556 MNLQPIFWIGLISSV...KSAVTTVVNPKYEGK 0 729 751 ENPECPTGPD IIPIVAGVVAGIVLIGLALLLIW KLLMIIHDRR
9 P16234 MGTSHPAFLVLGCLL...DIGIDSSDLVEDSFL 0 527 549 VAPTLRSELT VAAAVLVLLVIVIISLIVLVVIW KQKPRYEIRW
10 P50895 MEPPDAPAQARGAPR...SGGARGGSGGFGDEC 0 549 571 TVSPQTSQAG VAVMAVAVSVGLLLLVVAVFYCV RRKGGPCCRQ

2. Reduce scale redundancy with AAclust

Hundreds of amino-acid scales are highly correlated. AAclust clusters them and keeps one representative (medoid) per cluster, so CPP profiles a compact, non-redundant scale set. Here we fix n_clusters=50 for a quick, deterministic run; AAclust can also choose the cluster count automatically.

aac = aa.AAclust()
X = np.array(df_scales).T
scales = aac.fit(X, names=list(df_scales), n_clusters=50).medoid_names_
df_scales = df_scales[scales]

3. Run CPP

SequenceFeature builds the parts (the TMD and its juxtamembrane flanks) that CPP profiles. CPP.run then creates all Part × Split × Scale features, contrasts the two groups, and returns the ranked, non-redundant feature table df_feat.

sf = aa.SequenceFeature()
df_parts = sf.get_df_parts(df_seq=df_seq)
cpp = aa.CPP(df_scales=df_scales, df_parts=df_parts)
df_feat = cpp.run(labels=labels)
aa.display_df(df=df_feat, n_rows=10, show_shape=True)

DataFrame shape: (100, 13)
  feature category subcategory scale_name scale_description abs_auc abs_mean_dif mean_dif std_test std_ref p_val_mann_whitney p_val_fdr_bh positions
1 TMD_C_JMD_C-Seg...4,5)-ZIMJ680104 Energy Isoelectric point Isoelectric point Isoelectric poi...n et al., 1968) 0.373000 0.220000 0.220000 0.124000 0.137000 0.000000 0.000000 33,34,35,36
2 TMD_C_JMD_C-Seg...6,9)-ZIMJ680104 Energy Isoelectric point Isoelectric point Isoelectric poi...n et al., 1968) 0.341000 0.264000 0.264000 0.187000 0.172000 0.000000 0.000002 32,33
3 TMD-Segment(11,12)-ROBB760113 Conformation β-turn β-turn Information mea...n-Suzuki, 1976) 0.337000 0.319000 -0.319000 0.175000 0.256000 0.000000 0.000003 27,28
4 TMD_C_JMD_C-Seg...2,2)-ZIMJ680104 Energy Isoelectric point Isoelectric point Isoelectric poi...n et al., 1968) 0.337000 0.106000 0.106000 0.071000 0.082000 0.000000 0.000002 31,32,33,34,35,36,37,38,39,40
5 TMD_C_JMD_C-Seg...5,7)-FAUJ880103 ASA/Volume Volume Volume Normalized van ...e et al., 1988) 0.334000 0.174000 0.174000 0.109000 0.139000 0.000000 0.000002 32,33,34
6 TMD_C_JMD_C-Pat...4,8)-CHOC760104 ASA/Volume Buried Buried Proportion of r...(Chothia, 1976) 0.326000 0.309000 -0.309000 0.165000 0.274000 0.000000 0.000002 33,37
7 TMD_C_JMD_C-Pat...,12)-PALJ810102 Conformation α-helix α-helix Normalized freq...u et al., 1981) 0.325000 0.159000 0.159000 0.119000 0.136000 0.000000 0.000002 24,28,32
8 TMD_C_JMD_C-Seg...,15)-MITS020101 Polarity Amphiphilicity Amphiphilicity Amphiphilicity ...u et al., 2002) 0.324000 0.237000 0.237000 0.174000 0.171000 0.000000 0.000002 33
9 TMD_C_JMD_C-Pat...4,8)-GUYH850101 Composition MPs (anchor) Partition energy Partition energy (Guy, 1985) 0.323000 0.242000 0.242000 0.135000 0.219000 0.000000 0.000002 33,37
10 TMD_C_JMD_C-Seg...,10)-MITS020101 Polarity Amphiphilicity Amphiphilicity Amphiphilicity ...u et al., 2002) 0.323000 0.235000 0.235000 0.174000 0.171000 0.000000 0.000002 33,34

4. Rank and read out the signature

A TreeModel scores how much each feature contributes to telling the two groups apart (group-level feature importance). CPPPlot.ranking then shows the top features — each an interpretable, residue-grounded Part × Split × Scale combination — which together form the physicochemical signature of the substrates.

X = sf.feature_matrix(df_parts=df_parts, features=df_feat["feature"])
tm = aa.TreeModel()
tm.fit(X, labels=labels)
df_feat = tm.add_feat_importance(df_feat=df_feat, sort=True)

cpp_plot = aa.CPPPlot()
aa.plot_settings(short_ticks=True, weight_bold=False)
cpp_plot.ranking(df_feat=df_feat, n_top=10)
plt.tight_layout()
plt.show()
../_images/tutorial0_minimal_1_output_9_0.png

5. Map the full signature

CPPPlot.ranking reads out the top features one at a time; CPPPlot.feature_map charts the whole signature in a single figure — every selected Part × Split × Scale feature placed by scale subcategory (y-axis) and residue position (x-axis), colored by the group mean difference and marked by the same feature importance. It is the most complete single-figure read-out of a CPP analysis, and the canvas the Quick start tutorial later overlays SHAP feature impact onto (shap_plot=True).

aa.plot_settings(font_scale=0.65, weight_bold=False)
cpp_plot.feature_map(df_feat=df_feat)
plt.show()
../_images/tutorial0_minimal_2_output_11_0.png

That is the whole loop: data → CPP → signature. To pick the right setup for your task — residue, domain, or protein level — see the Prediction tasks page, and the Protocols for end-to-end workflows.