Residue representations¶
BLOSUM100 remains the default. Kidera factors provide ten residue-level
physicochemical channels, KF1–KF10, in published order, covering the 20
canonical amino acids. No ambiguous-residue imputation is performed.
Kidera factor descriptions¶
The atlas uses these property descriptions in hover text and tables; stable
channel IDs remain KF1–KF10. These are conventional labels, not independent
measurements of each named property. In particular, KF5–KF10 combine several
properties and are named for their most heavily weighted component. See the
Peptides reference documentation.
Factor |
Property description |
|---|---|
KF1 |
Helix/bend preference |
KF2 |
Side-chain size |
KF3 |
Extended structure preference |
KF4 |
Hydrophobicity |
KF5 |
Double-bend preference |
KF6 |
Partial specific volume |
KF7 |
Flat extended preference |
KF8 |
Occurrence in alpha region |
KF9 |
Carboxyl-group pK (pK-C) |
KF10 |
Surrounding hydrophobicity |
from protspect import SequenceSet
from protspect.embeddings import kidera_factors
features = SequenceSet(embedding=kidera_factors(), compute_cross_spectra=False)
features.from_sequence_list(["ACDE", "ACDF"], spectrum_length=8)
table = features.export_datatable()
The same mapping works with ComplexSet and the optional ML transformer:
from protspect.sklearn import SpectralTransformer
transformer = SpectralTransformer(embedding=kidera_factors(), spectrum_length=8)
X = transformer.fit_transform(["ACDE", "ACDF"])
kidera_factors() returns fresh ordinary dictionaries for customization and
scikit-learn cloning/pickling. protspect.embeddings.kidera exposes the same
numbers as a read-only nested mapping for direct core use. Use the factory
when passing an explicit embedding to scikit-learn estimators.
These are per-residue factor signals, not ten whole-sequence averages. The existing Fourier transform retains sequence ordering. Published numbers are used directly at two-decimal precision; normalization, centering, and fitted standardization remain explicit options. Cross spectra work as for any other embedding. Unknown residues raise the existing validation error.
Reference and provenance¶
Kidera, A., Konishi, Y., Oka, M., Ooi, T., & Scheraga, H. A. (1985). Statistical analysis of the physical properties of the 20 naturally occurring amino acids. Journal of Protein Chemistry, 4(1), 23–55. doi:10.1007/BF01025492.
The ten factors summarize multiple correlated physical properties; interpret them as statistical factors rather than ten pure experimental measurements. Their signs and order are retained exactly.
All 200 numerical values were cross-checked between these pinned reference implementations, without copying their program logic:
CRAN Peptides AAdata,
AAdata$kideraFactors, KF1–KF10.peptides.py Kidera tables,
KF1.csv–KF10.csv.
The original article verifies the citation and method description; its full numerical table was checked against the implementations above, rather than a copy of the paywalled article. Tests pin a digest of the CRAN factor table, selected vectors, and an independently computed direct DFT.
AAindex1¶
Load additional scalar residue properties from an official local AAindex1 file. See AAindex1 usage and provenance.
AAindex2 and AAindex3¶
Substitution and contact-potential matrices can supply canonical row/column profile channels. Their axes and direction are explicit; full matrices are never forced to symmetry. See matrix representations.