|
1
|
|
|
""" |
|
2
|
|
|
X. |
|
3
|
|
|
""" |
|
4
|
|
|
from __future__ import annotations |
|
5
|
|
|
from dataclasses import dataclass |
|
6
|
|
|
from pathlib import Path |
|
7
|
|
|
from typing import TypeVar, Mapping, Sequence, List |
|
8
|
|
|
|
|
9
|
|
|
import numpy as np |
|
|
|
|
|
|
10
|
|
|
import pandas as pd |
|
|
|
|
|
|
11
|
|
|
from typeddfs import BaseDf |
|
|
|
|
|
|
12
|
|
|
from typeddfs.df_errors import UnsupportedOperationError |
|
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
from mandos import logger |
|
15
|
|
|
from mandos.analysis.io_defns import SimilarityDfLongForm, SimilarityDfShortForm |
|
16
|
|
|
from mandos.entries.searcher import InputFrame |
|
17
|
|
|
from mandos.model.rdkit_utils import RdkitUtils |
|
18
|
|
|
|
|
19
|
|
|
T = TypeVar("T", bound=BaseDf) |
|
|
|
|
|
|
20
|
|
|
|
|
21
|
|
|
|
|
22
|
|
|
@dataclass(frozen=True, repr=True) |
|
|
|
|
|
|
23
|
|
|
class MatrixPrep: |
|
24
|
|
|
kind: str |
|
25
|
|
|
normalize: bool |
|
26
|
|
|
log: bool |
|
27
|
|
|
invert: bool |
|
28
|
|
|
|
|
29
|
|
|
def from_files(self, paths: Sequence[Path]) -> SimilarityDfLongForm: |
|
|
|
|
|
|
30
|
|
|
dct = {} |
|
31
|
|
|
for p in paths: |
|
|
|
|
|
|
32
|
|
|
key = p.with_suffix("").name |
|
33
|
|
|
try: |
|
34
|
|
|
mx = SimilarityDfShortForm.read_file(p) |
|
|
|
|
|
|
35
|
|
|
dct[key] = mx |
|
36
|
|
|
except (OSError, UnsupportedOperationError, ValueError): |
|
37
|
|
|
logger.error(f"Failed to load matrix at {str(p)}") |
|
38
|
|
|
raise |
|
39
|
|
|
return self.create(dct) |
|
40
|
|
|
|
|
41
|
|
|
def create(self, key_to_mx: Mapping[str, SimilarityDfShortForm]) -> SimilarityDfLongForm: |
|
|
|
|
|
|
42
|
|
|
df = SimilarityDfLongForm( |
|
|
|
|
|
|
43
|
|
|
pd.concat([mx.to_long_form(self.kind, key) for key, mx in key_to_mx.items()]) |
|
44
|
|
|
) |
|
45
|
|
|
vals = df["value"] |
|
46
|
|
|
if self.invert: |
|
47
|
|
|
vals = -vals |
|
48
|
|
|
if self.normalize: |
|
49
|
|
|
mn, mx = vals.min(), vals.max() |
|
|
|
|
|
|
50
|
|
|
vals = (vals - mn) / (mn - mx) |
|
51
|
|
|
if self.log: |
|
52
|
|
|
# this is a bit stupid, but calc the log then normalize again |
|
53
|
|
|
# we can't take the log before normalization because we might have negative values |
|
54
|
|
|
vals = vals.map(np.log10) |
|
55
|
|
|
mn, mx = vals.min(), vals.max() |
|
|
|
|
|
|
56
|
|
|
vals = (vals - mn) / (mn - mx) |
|
57
|
|
|
df["value"] = vals |
|
58
|
|
|
return SimilarityDfLongForm.convert(df) |
|
59
|
|
|
|
|
60
|
|
|
@classmethod |
|
61
|
|
|
def ecfp_matrix(cls, df: InputFrame, radius: int, n_bits: int) -> SimilarityDfShortForm: |
|
|
|
|
|
|
62
|
|
|
# TODO: This is inefficient and long |
|
|
|
|
|
|
63
|
|
|
indices = range(len(df)) |
|
64
|
|
|
keys = df["inchikey"] |
|
65
|
|
|
on_bits = [ |
|
66
|
|
|
RdkitUtils.ecfp(c, radius=radius, n_bits=n_bits).list_on for c in df.get_structures() |
|
67
|
|
|
] |
|
68
|
|
|
the_rows: List[List[float]] = [] |
|
69
|
|
|
for i, row_key, row_print in zip(indices, keys, on_bits): |
|
|
|
|
|
|
70
|
|
|
for j, col_key, col_print in zip(indices, keys, on_bits): |
|
|
|
|
|
|
71
|
|
|
the_row = [] |
|
72
|
|
|
if i < j: |
|
73
|
|
|
jaccard = len(row_print.intersection(col_print)) / len( |
|
74
|
|
|
row_print.union(col_print) |
|
75
|
|
|
) |
|
76
|
|
|
the_row.append(jaccard) |
|
77
|
|
|
the_rows.append(the_row) |
|
78
|
|
|
short = SimilarityDfShortForm(the_rows) |
|
79
|
|
|
short["inchikey"] = keys |
|
80
|
|
|
short = short.set_index("inchikey") |
|
81
|
|
|
short.columns = keys |
|
82
|
|
|
return SimilarityDfShortForm.convert(short) |
|
83
|
|
|
|
|
84
|
|
|
|
|
85
|
|
|
__all__ = ["MatrixPrep"] |
|
86
|
|
|
|