AAclust: Selecting redundancy-reduced scale sets

The Amino Acid clustering (AAclust) class is k-optimized clustering wrapper for selecting redundancy-reduced sets of numerical scales, introduced in [Breimann24a].

You will learn

• Tool: AAclust

• Input: a numeric matrix X (e.g. df_scales), optional n_clusters

• Output: a redundancy-reduced subset via labels_ and medoids_

• Best used for: selecting redundancy-reduced scale sets (or representative proteins)

• Related protocol: P7: Select & reduce features

• Related API: AAclust, AAclustPlot

We load an example scale dataset to showcase it:

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

# Load the full amino acid scale set (scales as columns; transposed for clustering)
df_scales = aa.load_scales()
X = df_scales.T

AAclust can utilize any clustering model that uses the n_clusters parameter:

from sklearn.cluster import KMeans

# AAclust with KMeans (default)
aac = aa.AAclust(model_class=KMeans)

By fitting AAclust, its three-step algorithm is performed to select an optimized n_clusters (k). The three steps involve (1) an estimation of lower bound of k, (2) refinement of k, and (3) an optional cluster merging. Various results are saved as attributes:

# Fit clustering model (KMeans by default)
aac = aa.AAclust()
aac.fit(X)
# Get output parameters
n_clusters = aac.n_clusters
print("n_clusters: ", n_clusters)

n_clusters:  56

Instead of optimizing the number of clusters, we can pre-define it using the n_clusters parameter:

# Fit clustering model with pre-selected k
labels = aac.fit(X, n_clusters=5).labels_

We can visualize the clustering results and the obtained cluster centers using the respective plotting AAclustPlot class. For scales, pass the scales DataFrame directly via df_scales; it is transposed internally, so you never call .T yourself. All data points are shown in the PCA plot with the cluster centers highlighted by an ‘x’:

import matplotlib.pyplot as plt

aac_plot = aa.AAclustPlot()
aa.plot_settings()
fig, ax = aac_plot.centers(df_scales=df_scales, labels=labels)
plt.tight_layout()
plt.show()

To obtain redundancy-reduced scale sets, AAclust selects one medoid per cluster, which is the scale closest to the center of the respective cluster. These can be highlighted using the medoids() method:

aac_plot = aa.AAclustPlot()
aa.plot_settings()
fig, ax = aac_plot.medoids(df_scales=df_scales, labels=labels)
plt.tight_layout()
plt.show()

A one-call shortcut: select_scales

The fit → medoid → map-back-to-columns sequence above is the canonical scale-selection workflow, so AAclust bundles it into a single call. select_scales clusters the scales in df_scales and returns the columns of the representative (medoid) scales, one per cluster, without the manual transpose and medoid-name bookkeeping. The fitted attributes (labels_, medoid_names_, …) remain available on the instance afterwards.

# One-call redundancy reduction: return the representative (medoid) scale per cluster
df_scales_selected = aac.select_scales(df_scales, n_clusters=100)
aa.display_df(df_scales_selected, n_rows=10, show_shape=True)

DataFrame shape: (20, 100)

	ANDN920101	SIMZ760101	ARGP820103	AURR980114	ROBB760106	ROBB760113	MEIH800101	BIGC670101	BIOV880101	MEEJ810102	BROC820102	COHE430101	ONEK900101	BUNA790103	ROBB760101	FAUJ880101	RACS820112	QIAN880133	QIAN880123	KARS160101	LINS030101	CHAM830107	CHAM830108	LINS030104	CHOC760103	CHOP780101	ROBB760102	QIAN880109	QIAN880115	LEVM780105	WOLS870103	FINA910102	CHOP780214	CHOP780215	CIDH920105	PONP930101	GEIM800101	KARS160121	CEDJ970104	DAYM780201	NADH010102	WOLR790101	FAUJ880111	AURR980106	FASG760105	FINA910101	LINS030106	FAUJ880112	KLEP840101	MITS020101	GEIM800102	PALJ810105	GRAR740101	MIYS990105	MEIH800102	FUKS010102	HOPT810101	ROSG850101	ISOY800102	ISOY800106	MAXF760104	YUTK870104	OOBM850105	LEVM760103	RICJ880103	NAKH900103	NAKH900104	NAKH900109	NAKH920105	OOBM770104	OOBM850104	CORJ870107	PRAM820102	KUMS000103	QIAN880104	QIAN880117	SUYM030101	QIAN880130	QIAN880137	RACS820103	MUNV940102	RICJ880101	RICJ880113	RICJ880114	ROBB760107	SNEP660101	SUEM840102	TANS770108	VASM830101	VASM830102	VELV850101	BASU050102	ZIMJ680101	GEOR030106	AURR980118	KOEH090102	NADH010107	GEOR030102	KARS160110	LINS030117
AA
A	0.494000	0.268000	0.504000	0.746000	0.342000	0.101000	0.556000	0.164000	0.476000	0.221000	0.567000	0.500000	1.000000	0.691000	0.921000	0.305000	0.427000	0.291000	0.133000	0.100000	0.152000	0.000000	0.000000	0.093000	0.627000	0.174000	0.484000	0.754000	0.554000	0.247000	0.466000	0.024000	0.124000	0.044000	0.393000	0.471000	0.725000	0.248000	0.905000	0.707000	0.749000	0.979000	0.000000	0.412000	0.852000	0.154000	0.046000	0.000000	0.500000	0.000000	0.620000	0.276000	0.000000	0.492000	0.420000	0.512000	0.453000	0.147000	0.270000	0.543000	0.034000	0.966000	0.465000	0.999000	0.118000	0.329000	0.224000	0.829000	0.442000	0.899000	0.567000	0.446000	0.823000	1.000000	0.842000	0.455000	0.403000	0.358000	0.116000	0.287000	0.000000	0.250000	0.632000	0.324000	0.333000	0.653000	0.103000	0.097000	0.437000	0.617000	0.295000	0.275000	0.270000	0.516000	0.356000	0.174000	0.282000	0.250000	0.863000	0.186000
C	0.864000	0.258000	0.387000	0.219000	0.747000	0.849000	0.370000	0.323000	0.936000	0.420000	0.000000	0.033000	0.739000	0.819000	0.445000	0.422000	1.000000	0.369000	0.676000	0.200000	0.433000	0.000000	1.000000	0.000000	0.746000	0.661000	0.387000	0.645000	0.297000	0.469000	1.000000	0.024000	0.584000	0.658000	0.647000	0.760000	0.176000	0.776000	0.095000	0.017000	1.000000	0.837000	0.000000	0.000000	0.164000	0.154000	0.000000	0.000000	0.500000	0.000000	1.000000	0.157000	1.000000	0.016000	0.220000	0.000000	0.375000	0.429000	0.986000	0.510000	0.189000	0.929000	0.000000	0.932000	0.147000	0.000000	0.300000	0.000000	0.111000	0.751000	0.543000	0.725000	0.136000	0.000000	0.544000	0.000000	0.888000	1.000000	0.388000	0.000000	0.420000	0.188000	0.526000	0.471000	0.820000	0.496000	0.337000	0.000000	0.521000	0.734000	0.657000	0.866000	0.482000	0.000000	0.000000	0.115000	1.000000	0.246000	0.557000	0.000000
D	1.000000	0.206000	0.000000	0.314000	0.234000	0.790000	0.852000	0.324000	0.191000	0.138000	0.358000	0.000000	0.754000	0.745000	0.555000	0.382000	0.509000	0.735000	0.362000	0.400000	0.451000	1.000000	0.000000	0.588000	0.237000	0.908000	0.813000	0.340000	0.475000	0.037000	0.766000	0.081000	0.933000	0.246000	0.034000	0.021000	0.516000	0.683000	0.581000	0.759000	0.371000	0.459000	0.000000	0.431000	0.098000	1.000000	0.754000	1.000000	0.000000	0.000000	0.240000	0.622000	0.502000	0.825000	0.700000	0.655000	1.000000	0.216000	0.041000	0.642000	0.215000	0.934000	0.317000	0.993000	0.588000	0.075000	0.000000	0.286000	0.012000	0.755000	0.986000	0.000000	0.585000	0.400000	0.684000	0.491000	0.474000	0.274000	0.727000	0.923000	0.270000	0.688000	0.263000	0.118000	0.099000	0.091000	0.909000	1.000000	0.976000	0.225000	1.000000	0.008000	0.208000	0.404000	0.658000	1.000000	0.824000	0.091000	0.293000	0.186000
E	0.420000	0.210000	0.032000	0.805000	0.000000	0.092000	0.889000	0.488000	0.106000	0.227000	0.212000	0.200000	0.826000	0.745000	1.000000	0.372000	0.586000	0.398000	0.286000	0.500000	0.616000	1.000000	0.000000	0.804000	0.288000	0.248000	1.000000	0.754000	0.653000	0.000000	0.445000	0.081000	0.360000	0.096000	0.000000	0.092000	0.945000	0.710000	0.797000	0.724000	0.263000	0.435000	0.000000	1.000000	0.459000	0.423000	0.923000	1.000000	0.000000	0.183000	0.760000	0.228000	0.335000	0.857000	0.780000	1.000000	1.000000	0.315000	0.000000	0.576000	0.056000	0.935000	0.518000	0.969000	0.088000	0.112000	0.059000	0.378000	0.000000	0.665000	0.808000	0.233000	0.551000	0.621000	0.833000	0.227000	0.336000	0.326000	0.512000	0.916000	0.048000	0.438000	0.789000	0.235000	0.613000	0.223000	0.077000	0.000000	0.608000	0.531000	0.046000	0.061000	0.212000	0.610000	0.548000	0.802000	0.540000	0.404000	0.396000	0.349000
F	0.877000	0.887000	0.670000	0.680000	0.703000	0.328000	0.000000	0.783000	1.000000	0.862000	0.903000	0.567000	0.812000	1.000000	0.622000	0.702000	0.539000	0.301000	0.429000	0.700000	0.744000	0.000000	1.000000	0.082000	0.831000	0.119000	0.265000	0.764000	0.446000	0.679000	0.514000	0.024000	0.292000	0.105000	0.844000	0.665000	0.549000	0.842000	0.365000	0.198000	0.915000	0.859000	0.000000	0.314000	0.557000	0.154000	0.000000	0.000000	0.500000	0.000000	0.380000	0.307000	0.000000	0.000000	0.000000	0.128000	0.141000	0.811000	0.676000	0.351000	0.056000	0.934000	0.449000	0.969000	0.029000	0.382000	0.806000	0.536000	0.288000	0.305000	0.413000	0.950000	0.340000	0.350000	0.798000	0.655000	0.689000	0.232000	0.455000	0.000000	0.301000	0.562000	0.053000	0.206000	0.396000	0.612000	0.233000	0.395000	0.269000	0.023000	0.749000	1.000000	0.896000	0.712000	0.521000	0.068000	0.547000	0.536000	0.208000	0.326000
G	0.025000	0.032000	0.170000	0.077000	0.266000	1.000000	0.852000	0.000000	0.388000	0.177000	0.374000	0.133000	0.696000	0.596000	0.000000	0.000000	0.603000	0.828000	0.629000	0.000000	0.000000	0.000000	0.000000	0.144000	0.593000	1.000000	0.000000	0.498000	0.693000	0.284000	0.494000	0.049000	0.994000	0.868000	0.115000	0.275000	0.000000	0.000000	0.797000	0.267000	0.561000	1.000000	0.000000	0.142000	0.869000	0.154000	0.123000	0.000000	0.500000	0.000000	0.020000	1.000000	0.269000	0.492000	0.560000	0.605000	0.531000	0.000000	0.054000	0.000000	1.000000	0.967000	0.866000	0.000000	0.500000	0.323000	0.291000	0.754000	0.363000	1.000000	0.638000	0.352000	1.000000	0.286000	0.000000	1.000000	0.776000	0.737000	0.264000	0.570000	0.639000	0.812000	0.000000	0.059000	0.027000	0.000000	0.000000	0.148000	0.143000	0.455000	0.040000	0.000000	0.033000	0.210000	0.562000	0.328000	0.143000	0.000000	0.863000	0.023000
H	0.840000	0.387000	0.053000	0.562000	0.323000	0.454000	0.407000	0.561000	0.579000	0.044000	0.508000	0.233000	0.739000	0.851000	0.598000	0.714000	0.457000	0.443000	0.638000	0.600000	0.640000	0.000000	1.000000	0.423000	0.271000	0.440000	0.155000	0.847000	0.782000	0.235000	0.601000	0.024000	0.449000	0.009000	0.475000	0.326000	0.769000	0.683000	0.122000	0.414000	0.439000	0.432000	1.000000	0.279000	0.000000	0.154000	0.400000	0.000000	0.500000	0.209000	0.540000	0.032000	0.211000	0.492000	0.380000	0.197000	0.453000	0.575000	0.541000	0.530000	0.114000	0.970000	0.459000	0.969000	0.294000	0.089000	0.424000	0.082000	0.005000	0.464000	0.814000	0.610000	0.483000	0.136000	0.360000	0.545000	0.646000	0.737000	0.198000	0.000000	0.357000	0.562000	0.842000	1.000000	0.405000	0.372000	0.000000	0.356000	0.745000	0.345000	0.191000	0.305000	0.358000	0.612000	1.000000	0.650000	0.134000	0.201000	0.355000	0.419000
I	0.000000	0.990000	0.543000	0.893000	1.000000	0.151000	0.037000	0.663000	0.885000	0.552000	0.788000	1.000000	0.797000	0.745000	0.561000	1.000000	0.349000	0.049000	0.514000	0.400000	0.530000	0.000000	0.000000	0.010000	1.000000	0.000000	0.077000	0.690000	0.297000	1.000000	0.318000	0.024000	0.000000	0.026000	1.000000	1.000000	0.462000	0.604000	0.541000	0.672000	0.909000	0.989000	0.000000	0.392000	0.885000	0.000000	0.000000	0.000000	0.500000	0.000000	0.460000	0.047000	0.000000	0.079000	0.060000	0.191000	0.250000	0.588000	0.878000	0.464000	0.052000	1.000000	0.251000	0.975000	0.029000	0.565000	0.824000	0.547000	0.683000	0.568000	0.257000	0.969000	0.313000	0.500000	0.588000	0.727000	0.516000	0.200000	0.306000	0.811000	0.221000	0.375000	0.105000	0.206000	0.324000	0.934000	0.714000	0.615000	0.570000	0.070000	0.000000	0.917000	1.000000	0.457000	0.644000	0.000000	0.347000	0.161000	0.033000	0.140000
K	0.506000	0.516000	0.004000	0.970000	0.329000	0.613000	1.000000	0.694000	0.000000	0.000000	0.368000	0.733000	0.942000	0.691000	0.665000	0.451000	0.392000	0.456000	0.238000	0.500000	0.768000	0.000000	1.000000	1.000000	0.034000	0.495000	0.071000	0.872000	0.644000	0.136000	0.040000	0.000000	0.331000	0.368000	0.247000	0.000000	0.769000	0.660000	0.743000	0.328000	0.000000	0.467000	1.000000	0.402000	0.557000	0.038000	0.831000	0.000000	1.000000	0.530000	0.520000	0.362000	0.120000	1.000000	1.000000	0.784000	1.000000	0.325000	0.149000	0.616000	0.233000	0.956000	1.000000	1.000000	0.176000	0.151000	0.153000	0.246000	0.004000	0.743000	0.291000	0.027000	0.027000	0.543000	0.421000	0.618000	0.351000	0.558000	0.413000	1.000000	0.142000	0.438000	0.895000	0.588000	0.054000	0.562000	0.012000	0.247000	0.559000	0.688000	0.294000	0.112000	0.521000	0.536000	0.822000	0.809000	0.536000	0.195000	0.433000	1.000000
L	0.272000	0.835000	0.989000	0.923000	0.747000	0.076000	0.259000	0.663000	0.867000	0.696000	0.913000	1.000000	0.942000	0.691000	0.720000	0.618000	0.543000	0.204000	0.438000	0.400000	0.567000	0.000000	0.000000	0.041000	0.746000	0.110000	0.200000	0.936000	0.317000	0.469000	0.325000	0.024000	0.129000	0.149000	0.773000	0.734000	0.747000	0.604000	1.000000	0.190000	0.901000	0.995000	0.000000	0.284000	0.885000	0.154000	0.000000	0.000000	0.500000	0.000000	0.620000	0.000000	0.000000	0.016000	0.180000	0.326000	0.250000	0.626000	0.473000	0.682000	0.042000	0.990000	0.351000	0.968000	0.029000	1.000000	0.935000	1.000000	1.000000	0.562000	0.394000	1.000000	0.340000	0.643000	1.000000	0.064000	0.591000	0.326000	0.215000	0.000000	0.104000	0.375000	0.316000	0.206000	0.000000	1.000000	0.428000	0.123000	0.717000	0.771000	0.000000	0.920000	0.821000	0.690000	0.671000	0.012000	0.364000	0.513000	0.218000	0.186000

Next application: selecting representative proteins

AAclust clusters any numerical feature matrix, not only scales. Because a medoid is a real sample rather than a synthetic centroid, clustering proteins by their feature vectors and keeping one medoid per cluster yields a redundancy-reduced set of representative proteins, each an actual entry from df_seq. select_proteins does this in one call and reports a cluster label, an is_representative flag, and dist_to_rep (the distance of each protein to its cluster representative, i.e. a redundancy score).

Below, the per-protein matrix comes from CPP features. The same call accepts other per-protein representations just as well, such as pooled protein-language-model embeddings or structural (e.g. DSSP) features, which the data-representations tutorial demonstrates.

# Build a per-protein CPP feature matrix (one row per protein)
df_seq = aa.load_dataset(name="DOM_GSEC", n=25)
labels = list(df_seq["label"])
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)
X = sf.feature_matrix(df_parts=df_parts, features=df_feat["feature"])
print("X (n_proteins, n_features):", X.shape)

X (n_proteins, n_features): (50, 100)

# Select one representative protein (cluster medoid) per cluster
df_repr = aac.select_proteins(df_seq=df_seq, X=X, n_clusters=10)
aa.display_df(df_repr, n_rows=10, show_shape=True)

DataFrame shape: (50, 12)

	entry	gene	sequence	label	tmd_start	tmd_stop	jmd_n	tmd	jmd_c	cluster	is_representative	dist_to_rep
1	Q14802	FXYD3	MQKVTLGLLVFLAGF...PGETPPLITPGSAQS	0	37	59	NSPFYYDWHS	LQVGGLICAGVLCAMGIIIVMSA	KCKCKFGQKS	8	0	1.465234
2	Q86UE4	MTDH	MAARSWQDELAQQAE...SPKQIKKKKKARRET	0	50	72	LGLEPKRYPG	WVILVGTGALGLLLLFLLGYGWA	AACAGARKKR	3	0	1.379094
3	Q969W9	PMEPA1	MHRLMGVNSTAAAAA...AIWSKEKDKQKGHPL	0	41	63	FQSMEITELE	FVQIIIIVVVMMVMVVVITCLLS	HYKLSARSFI	7	0	1.704990
4	P53801	PTTG1IP	MAPGVARGPTPYWRL...GLFKEENPYARFENN	0	97	119	RWGVCWVNFE	ALIITMSVVGGTLLLGIAICCCC	CCRRKRSRKP	7	1	0.000000
5	Q8IUW5	RELL1	MAPRALPGSAVLAAA...EVPATPVKRERSGTE	0	59	81	NDTGNGHPEY	IAYALVPVFFIMGLFGVLICHLL	KKKGYRCTTE	7	0	1.294481
6	P01135	TGFA	MVPSAGQLALFALGI...LLKGRTACCHSETVV	0	99	121	AVVAASQKKQ	AITALVVVSIVALAVLIITCVLI	HCCQVRKHCE	8	1	0.000000
7	O43914	TYROBP	MGGLEPCSRLLLLPL...SDVYSDLNTQRPYYK	0	42	64	DCSCSTVSPG	VLAGIVMGDLVLTVLIALAVYFL	GRLVPRGRGA	9	0	1.563711
8	P05556	ITGB1	MNLQPIFWIGLISSV...KSAVTTVVNPKYEGK	0	729	751	ENPECPTGPD	IIPIVAGVVAGIVLIGLALLLIW	KLLMIIHDRR	9	1	0.000000
9	P16234	PDGFRA	MGTSHPAFLVLGCLL...DIGIDSSDLVEDSFL	0	527	549	VAPTLRSELT	VAAAVLVLLVIVIISLIVLVVIW	KQKPRYEIRW	0	0	1.123703
10	P50895	BCAM	MEPPDAPAQARGAPR...SGGARGGSGGFGDEC	0	549	571	TVSPQTSQAG	VAVMAVAVSVGLLLLVVAVFYCV	RRKGGPCCRQ	2	0	1.325820

The medoid of each cluster is a representative protein, so is_representative sums to the number of clusters and dist_to_rep is 0 for the representatives. To obtain a fixed number of representatives, set n_clusters; to let AAclust choose it, leave n_clusters=None and tune min_th. Use return_data="filtered" to get only the representative proteins together with their feature matrix:

# Keep only the representative proteins (and their feature rows)
df_repr_only, X_repr = aac.select_proteins(df_seq=df_seq, X=X, n_clusters=10,
                                           return_data="filtered")
print("representative proteins:", df_repr_only.shape[0])
aa.display_df(df_repr_only, n_rows=10, show_shape=True)

representative proteins: 10
DataFrame shape: (10, 12)

	entry	gene	sequence	label	tmd_start	tmd_stop	jmd_n	tmd	jmd_c	cluster	is_representative	dist_to_rep
1	Q14802	FXYD3	MQKVTLGLLVFLAGF...PGETPPLITPGSAQS	0	37	59	NSPFYYDWHS	LQVGGLICAGVLCAMGIIIVMSA	KCKCKFGQKS	2	1	0.000000
2	P53801	PTTG1IP	MAPGVARGPTPYWRL...GLFKEENPYARFENN	0	97	119	RWGVCWVNFE	ALIITMSVVGGTLLLGIAICCCC	CCRRKRSRKP	8	1	0.000000
3	P05556	ITGB1	MNLQPIFWIGLISSV...KSAVTTVVNPKYEGK	0	729	751	ENPECPTGPD	IIPIVAGVVAGIVLIGLALLLIW	KLLMIIHDRR	5	1	0.000000
4	Q9BZ11	ADAM33	MGWRPRRARGTPLLL...DPQADQVQMPRSCLW	0	13	35	WRPRRARGTP	LLLLLLLLLLWPVPGAGVLQGHI	PGQPVTPHWV	0	1	0.000000
5	Q7Z6A9	BTLA	MKTLPAMLGTGKLFW...VKEAPTEYASICVRS	0	153	175	PSKDEMASRP	WLLYRLLPLGGLPLLITTCFCLF	CCLRRHQGKQ	4	1	0.000000
6	P15509	CSF2RA	MLLLVTSLLLCELPH...GKGYREEVLTVKEIT	0	324	346	EFGSDDGNLG	SVYIYVLLIVGTLVCGIVLGFLF	KRFLRIQRLF	7	1	0.000000
7	Q03157	Aplp1	MGPTSPAARGQGRRW...HGYENPTYRFLEERP	1	585	607	APSGTGVSRE	ALSGLLIMGAGGGSLIVLSLLLL	RKKKPYGTIS	3	1	0.000000
8	P19022	CDH2	MCRIAGALRTLLPLL...PRFKKLADMYGGGDD	1	724	746	RIVGAGLGTG	AIIAILLCIIILLILVLMFVVWM	KRRDKERQAK	9	1	0.000000
9	D3ZZK3	Epha4	MAGIFYFILFSFLFG...MRTQMQQMHGRMVPV	1	548	570	RIIGDGANST	VLLVSVSGSVVLVVILIAAFVIS	RRRSKYSQAK	6	1	0.000000
10	Q61483	Dll1	MGRRSALALAVVSAL...VLSAEKDECVIATEV	1	545	567	HMESQGGPFP	WVAVCAGVVLVLLLLLGCAAVVV	CVRLKLQKHQ	1	1	0.000000

Visualizing the representative proteins

The protein clustering is visualized with the same medoids() method, but here the input is the per-protein feature matrix X (samples × features), passed as-is: proteins need no transpose, unlike the df_scales form used for scales above. The method projects X into PCA space and marks the representative protein (medoid) of each cluster.

# Visualize the protein clusters and their representatives (medoids) in PCA space
aac_plot = aa.AAclustPlot()
aa.plot_settings()
fig, ax = aac_plot.medoids(X, labels=df_repr["cluster"])
plt.tight_layout()
plt.show()

Feature-space vs. sequence-identity redundancy

select_proteins reduces redundancy in CPP feature space (physicochemical similarity). A complementary tool, filter_seq() (CD-HIT; requires the pro extra), reduces redundancy by sequence identity, keeping one representative per cluster of similar sequences. Both annotate the proteins with the same cluster / is_representative columns, so they compose, but they answer different questions.

# Sequence-identity redundancy reduction with CD-HIT (requires aaanalysis[pro])
df_cdhit = aa.filter_seq(df_seq, method="cd-hit", similarity_threshold=0.7)
n_cpp = int(df_repr["is_representative"].sum())
n_seq = int(df_cdhit["is_representative"].sum())
print(f"representatives by select_proteins (CPP feature space): {n_cpp} of {len(df_seq)}")
print(f"representatives by filter_seq    (sequence identity):   {n_seq} of {len(df_seq)}")
aa.display_df(df_cdhit, n_rows=10, show_shape=True)

representatives by select_proteins (CPP feature space): 10 of 50
representatives by filter_seq    (sequence identity):   50 of 50
DataFrame shape: (50, 4)

	entry	cluster	identity_with_rep	is_representative
1	Q9ERC8	0	100.000000	1
2	Q63155	1	100.000000	1
3	Q8VHS2	2	100.000000	1
4	P08069	3	100.000000	1
5	Q15303	4	100.000000	1
6	Q6ZRH7	5	100.000000	1
7	P16234	6	100.000000	1
8	D3ZZK3	7	100.000000	1
9	P54763	8	100.000000	1
10	O94985	9	100.000000	1

At a 70% identity threshold, CD-HIT finds these sequences non-redundant and keeps all of them, while select_proteins still compresses them to 10 physicochemical representatives. The two tools are complementary, not interchangeable: use filter_seq() to drop near-duplicate sequences before training, and select_proteins to pick a diverse, interpretable subset in feature space.

For further details, see our Feature Engineering API, AAontology Usage Principles, and AAclust Usage Principles.