mkt.schema.conservation_schema

Persisted output of the KLIFS hierarchical-conservation clustering.

Defines KLIFSConservationData, a dependency-light (pure-Python, no scipy/numpy) Pydantic artifact holding the pairwise distances, dendrogram, leaf order, and display-leaf composition produced by mkt.databases.conservation.KLIFSHierarchicalConservation, plus the provenance metadata describing how they were assembled. Serialized via mkt.schema.io_utils.serialize_conservation_data() and shipped as package data so downstream consumers can reuse the tree/distances without recomputing them (or importing scipy).

Module Attributes

DICT_CLADE_ANCHORS

Curated clade names keyed to a minimal set of anchor kinases that uniquely identify each clade by membership (not by display-order number, which is fragile to re-clustering).

Classes

KLIFSConservationData(*, metric, ...)

Persisted distances + dendrogram from the KLIFS conservation clustering.

mkt.schema.conservation_schema.DICT_CLADE_ANCHORS: dict[str, tuple[str, ...]] = {'EPH': ('EPHA3', 'EPHB1'), 'ERBB': ('EGFR', 'ERBB2'), 'FGFR': ('FGFR2', 'FGFR3'), 'INSR/ALK': ('INSR', 'ALK'), 'MET/AXL': ('MET', 'AXL'), 'PDGFR/VEGFR': ('KDR', 'KIT'), 'RET/TIE': ('RET', 'TEK'), 'SRC/ABL': ('SRC', 'ABL1'), 'TEC': ('BTK', 'TEC'), 'TRK/DDR': ('NTRK1', 'DDR2')}: Curated clade names keyed to a minimal set of anchor kinases that uniquely identify each clade by membership (not by display-order number, which is fragile to re-clustering). A display leaf is given a name only if it contains all of that name’s anchors, so a name silently disappears rather than mislabels if a clade splits. The SYK/ZAP70 pair is deliberately omitted – it clusters inside an EphA subclade rather than forming a clean Syk clade, and is excluded from the clade analysis.

class mkt.schema.conservation_schema.KLIFSConservationData(*, metric: str, blosum_name: str, linkage_method: str, conservation_threshold: float, weighting: str, exclude_pseudokinases: bool, gap_chars: str, n_kinases: int, pocket_length: int, names: list[str], position_labels: list[str], distances_condensed: list[float], diagonal_is_similarity: bool = True, linkage_matrix: list[list[float]], leaves_order: list[int], display_leaves: list[dict] = <factory>, display_order: list[int] = <factory>)[source]

Bases: BaseModel

Persisted distances + dendrogram from the KLIFS conservation clustering.

The pairwise distances are stored as the condensed upper-triangle vector (the same row-major order SciPy’s squareform uses), and the dendrogram as the SciPy linkage matrix (N-1 rows of [child_a, child_b, height, count]), both as plain Python numbers so the schema stays free of numpy/scipy. The symmetric matrix and the (kin1, kin2, distance) long form are rebuilt on demand via rebuild_square() and to_long_form(); the full tree is reconstructed from linkage_matrix in mkt.databases (where scipy lives).

blosum_name: str: Substitution matrix name used by the "blosum" metric.

conservation_threshold: float: Minimum consensus-residue frequency for a column to count as conserved.

diagonal_is_similarity: bool

rebuild as a similarity matrix (1 - distance, diagonal 1.0) rather than raw distance (diagonal 0.0).

Type:: Documents rebuild_square()’s default

display_leaf_names(dict_anchors: dict[str, tuple[str, ...]] = {'EPH': ('EPHA3', 'EPHB1'), 'ERBB': ('EGFR', 'ERBB2'), 'FGFR': ('FGFR2', 'FGFR3'), 'INSR/ALK': ('INSR', 'ALK'), 'MET/AXL': ('MET', 'AXL'), 'PDGFR/VEGFR': ('KDR', 'KIT'), 'RET/TIE': ('RET', 'TEK'), 'SRC/ABL': ('SRC', 'ABL1'), 'TEC': ('BTK', 'TEC'), 'TRK/DDR': ('NTRK1', 'DDR2')}) → list[str | None][source]

Return the curated clade name per display leaf (aligned to display_leaves).

Parameters:: dict_anchors (dict[str, tuple[str, …]]) – Clade-name -> anchor-kinase mapping. Defaults to DICT_CLADE_ANCHORS.
Returns:: One name (or None) per entry of display_leaves, in that order.
Return type:: list[str | None]

display_leaves: list[dict]

one entry per leaf as {"members": list[int], "kind": str} (indices into names).

Type:: Display-tree leaf rows from build_display_tree

display_order: list[int]: In-order (top-to-bottom) display order as indices into display_leaves.

distances_condensed: list[float]: Condensed upper-triangle distance vector (N * (N - 1) / 2 entries), in the row-major order used by scipy.spatial.distance.squareform.

exclude_pseudokinases: bool: Whether predicted pseudokinases were dropped before clustering.

gap_chars: str: Characters treated as gap/unknown and excluded from scoring.

leaves_order: list[int]: Leaf order (scipy.cluster.hierarchy.leaves_list), as indices into names, cached so consumers need not import scipy.

linkage_matrix: list[list[float]]: SciPy linkage matrix (N-1 rows of [child_a, child_b, height, count]); fully reconstructs the dendrogram via scipy.cluster.hierarchy.

linkage_method: str: SciPy linkage method ("average" UPGMA or "complete").

metric: str: Pairwise distance metric used ("blosum" or "identity").

n_kinases: int: Number of kinases in the panel (N; matrix dimension).

name_for_members(member_idx: list[int], dict_anchors: dict[str, tuple[str, ...]] = {'EPH': ('EPHA3', 'EPHB1'), 'ERBB': ('EGFR', 'ERBB2'), 'FGFR': ('FGFR2', 'FGFR3'), 'INSR/ALK': ('INSR', 'ALK'), 'MET/AXL': ('MET', 'AXL'), 'PDGFR/VEGFR': ('KDR', 'KIT'), 'RET/TIE': ('RET', 'TEK'), 'SRC/ABL': ('SRC', 'ABL1'), 'TEC': ('BTK', 'TEC'), 'TRK/DDR': ('NTRK1', 'DDR2')}) → str | None[source]

Return the curated clade name for a set of member indices.

A clade name applies only when the members contain all of that name’s anchor kinases (see DICT_CLADE_ANCHORS), so the mapping is robust to display-order renumbering and never mislabels a split clade.

Parameters:

member_idx (list[int]) – Member indices into names (e.g. a display_leaves entry’s "members").
dict_anchors (dict[str, tuple[str, …]]) – Clade-name -> anchor-kinase mapping. Defaults to DICT_CLADE_ANCHORS.

Returns:

The matching clade name, or None if no anchor set is fully contained.

Return type:

str | None

names: list[str]: HGNC kinase names in panel order (distance-matrix row/column order).

pocket_length: int: KLIFS pocket length (columns per kinase; 85 for the standard panel).

position_labels: list[str]: KLIFS region labels (e.g. "a.l:84") for the pocket columns.

rebuild_square(as_similarity: bool = True, reorder_by_leaves: bool = False) → list[list[float]][source]

Reconstruct the symmetric matrix from the condensed distance vector.

Parameters:

as_similarity (bool) – If True (default), return a similarity matrix (1 - distance) with a unit diagonal; if False, return raw distances with a zero diagonal.
reorder_by_leaves (bool) – If True, reorder rows and columns by leaves_order (dendrogram leaf order) rather than the panel order.

Returns:

The symmetric N x N matrix as nested lists.

Return type:

list[list[float]]

to_long_form() → list[tuple[str, str, float]][source]

Expand the condensed distances into (kin1, kin2, distance) triples.

Returns:: One triple per upper-triangle pair, using names for the labels.
Return type:: list[tuple[str, str, float]]

weighting: str: Per-node consensus weighting ("none" or "henikoff").