|
1
|
|
|
#! /usr/bin/env python |
|
|
|
|
|
|
2
|
|
|
# |
|
3
|
|
|
# Copyright (C) 2016 Rich Lewis <[email protected]> |
|
4
|
|
|
# License: 3-clause BSD |
|
5
|
|
|
|
|
6
|
1 |
|
import os |
|
7
|
1 |
|
import zipfile |
|
8
|
1 |
|
import logging |
|
9
|
1 |
|
LOGGER = logging.getLogger(__name__) |
|
10
|
|
|
|
|
11
|
1 |
|
import pandas as pd |
|
|
|
|
|
|
12
|
1 |
|
import numpy as np |
|
|
|
|
|
|
13
|
|
|
|
|
14
|
1 |
|
from ... import io |
|
15
|
1 |
|
from .base import Converter, contiguous_order |
|
16
|
|
|
|
|
17
|
1 |
|
from ...cross_validation import SimThresholdSplit |
|
18
|
|
|
|
|
19
|
1 |
|
TXT_COLUMNS = [l.lower() for l in """CAS |
|
20
|
|
|
Formula |
|
21
|
|
|
Mol_Weight |
|
22
|
|
|
Chemical_Name |
|
23
|
|
|
WS |
|
24
|
|
|
WS_temp |
|
25
|
|
|
WS_type |
|
26
|
|
|
WS_reference |
|
27
|
|
|
LogP |
|
28
|
|
|
LogP_temp |
|
29
|
|
|
LogP_type |
|
30
|
|
|
LogP_reference |
|
31
|
|
|
VP |
|
32
|
|
|
VP_temp |
|
33
|
|
|
VP_type |
|
34
|
|
|
VP_reference |
|
35
|
|
|
DC_pKa |
|
36
|
|
|
DC_temp |
|
37
|
|
|
DC_type |
|
38
|
|
|
DC_reference |
|
39
|
|
|
henry_law Constant |
|
40
|
|
|
HL_temp |
|
41
|
|
|
HL_type |
|
42
|
|
|
HL_reference |
|
43
|
|
|
OH |
|
44
|
|
|
OH_temp |
|
45
|
|
|
OH_type |
|
46
|
|
|
OH_reference |
|
47
|
|
|
BP_pressure |
|
48
|
|
|
MP |
|
49
|
|
|
BP |
|
50
|
|
|
FP""".split('\n')] |
|
51
|
|
|
|
|
52
|
1 |
|
class PhysPropConverter(Converter): |
|
|
|
|
|
|
53
|
|
|
|
|
54
|
1 |
|
def __init__(self, directory, output_directory, output_filename='physprop.h5'): |
|
|
|
|
|
|
55
|
|
|
|
|
56
|
|
|
output_path = os.path.join(output_directory, output_filename) |
|
57
|
|
|
|
|
58
|
|
|
sdf, txt = self.extract(directory) |
|
59
|
|
|
mols, data = self.process_sdf(sdf), self.process_txt(txt) |
|
60
|
|
|
|
|
61
|
|
|
LOGGER.debug('Compounds with data extracted: %s', len(data)) |
|
62
|
|
|
|
|
63
|
|
|
data = mols.to_frame().join(data) |
|
64
|
|
|
data = self.drop_inconsistencies(data) |
|
65
|
|
|
|
|
66
|
|
|
y = self.process_targets(data) |
|
|
|
|
|
|
67
|
|
|
LOGGER.debug('Compounds with experimental: %s', len(y)) |
|
68
|
|
|
|
|
69
|
|
|
data = data.ix[y.index] |
|
70
|
|
|
data.columns.name = 'targets' |
|
71
|
|
|
ms, y = data.structure, data.drop('structure', axis=1) |
|
|
|
|
|
|
72
|
|
|
|
|
73
|
|
|
cv = SimThresholdSplit(min_threshold=0.6, block_width=4000, n_jobs=-1).fit(ms) |
|
|
|
|
|
|
74
|
|
|
train, valid, test = cv.split((70, 15, 15)) |
|
75
|
|
|
|
|
76
|
|
|
(ms, y, train, valid, test) = contiguous_order((ms, y, train, valid, test), (train, valid, test)) |
|
|
|
|
|
|
77
|
|
|
splits = (('train', train), ('valid', valid), ('test', test)) |
|
78
|
|
|
|
|
79
|
|
|
self.run(ms, y, output_path=output_path, splits=splits) |
|
80
|
|
|
|
|
81
|
1 |
|
def extract(self, directory): |
|
|
|
|
|
|
82
|
|
|
LOGGER.info('Extracting from %s', directory) |
|
83
|
|
|
with zipfile.ZipFile(os.path.join(directory, 'phys_sdf.zip')) as f: |
|
|
|
|
|
|
84
|
|
|
sdf = f.extract('PhysProp.sdf') |
|
85
|
|
|
with zipfile.ZipFile(os.path.join(directory, 'phys_txt.zip')) as f: |
|
|
|
|
|
|
86
|
|
|
txt = f.extract('PhysProp.txt') |
|
87
|
|
|
return sdf, txt |
|
88
|
|
|
|
|
89
|
1 |
|
def process_sdf(self, path): |
|
|
|
|
|
|
90
|
|
|
LOGGER.info('Processing sdf at %s', path) |
|
91
|
|
|
mols = io.read_sdf(path, read_props=False).structure |
|
92
|
|
|
mols.index = mols.apply(lambda m: m.GetProp('CAS')) |
|
93
|
|
|
mols.index.name = 'cas' |
|
94
|
|
|
LOGGER.debug('Structures extracted: %s', len(mols)) |
|
95
|
|
|
return mols |
|
96
|
|
|
|
|
97
|
1 |
|
def process_txt(self, path): |
|
|
|
|
|
|
98
|
|
|
LOGGER.info('Processing txt at %s', path) |
|
99
|
|
|
data = pd.read_table(path, header=None, engine='python').iloc[:, :32] |
|
100
|
|
|
data.columns = TXT_COLUMNS |
|
101
|
|
|
data_types = data.columns[[s.endswith('_type') for s in data.columns]] |
|
102
|
|
|
data[data_types] = data[data_types].fillna('NAN') |
|
103
|
|
|
data = data.set_index('cas') |
|
104
|
|
|
return data |
|
105
|
|
|
|
|
106
|
1 |
|
def drop_inconsistencies(self, data): |
|
|
|
|
|
|
107
|
|
|
LOGGER.info('Dropping inconsistent data...') |
|
108
|
|
|
formula = data.structure.apply(lambda m: m.to_formula()) |
|
109
|
|
|
LOGGER.info('Inconsistent compounds: %s', (formula != data.formula).sum()) |
|
110
|
|
|
data = data[formula == data.formula] |
|
111
|
|
|
return data |
|
112
|
|
|
|
|
113
|
1 |
|
def process_targets(self, data): |
|
|
|
|
|
|
114
|
|
|
LOGGER.info('Dropping estimated data...') |
|
115
|
|
|
data = pd.concat([self.process_logS(data), |
|
116
|
|
|
self.process_logP(data), |
|
117
|
|
|
self.process_mp(data), |
|
118
|
|
|
self.process_bp(data)], axis=1) |
|
119
|
|
|
LOGGER.info('Dropped compounds: %s', data.isnull().all(axis=1).sum()) |
|
120
|
|
|
data = data[data.notnull().any(axis=1)] |
|
121
|
|
|
LOGGER.debug('Compounds with experimental activities: %s', len(data)) |
|
122
|
|
|
return data |
|
123
|
|
|
|
|
124
|
1 |
|
def process_logS(self, data): |
|
|
|
|
|
|
125
|
|
|
cleaned = pd.DataFrame(index=data.index) |
|
|
|
|
|
|
126
|
|
|
S = 0.001 * data.ws / data.mol_weight |
|
|
|
|
|
|
127
|
|
|
logS = np.log10(S) |
|
|
|
|
|
|
128
|
|
|
return logS[data.ws_type == 'EXP'] |
|
129
|
|
|
|
|
130
|
1 |
|
def process_logP(self, data): |
|
|
|
|
|
|
131
|
|
|
logP = data.logp[data.logp_type == 'EXP'] |
|
|
|
|
|
|
132
|
|
|
return logP[logP > -10] |
|
133
|
|
|
|
|
134
|
1 |
|
def process_mp(self, data): |
|
|
|
|
|
|
135
|
|
|
return data.mp.apply(self.fix_temp) |
|
136
|
|
|
|
|
137
|
1 |
|
def process_bp(self, data): |
|
|
|
|
|
|
138
|
|
|
return data.bp.apply(self.fix_temp) |
|
139
|
|
|
|
|
140
|
1 |
|
@staticmethod |
|
141
|
1 |
|
def fix_temp(s, mean_range=5): |
|
|
|
|
|
|
142
|
|
|
try: |
|
143
|
|
|
return float(s) |
|
144
|
|
|
except ValueError: |
|
145
|
|
|
if '<' in s or '>' in s: |
|
146
|
|
|
return np.nan |
|
147
|
|
|
s = s.strip(' dec') |
|
148
|
|
|
s = s.strip(' sub') |
|
149
|
|
|
if '-' in s and mean_range: |
|
150
|
|
|
rng = [float(n) for n in s.split('-')] |
|
151
|
|
|
if len(rng) > 2: |
|
152
|
|
|
return np.nan |
|
153
|
|
|
if np.abs(rng[1] - rng[0]) < mean_range: |
|
154
|
|
|
return (rng[0] + rng[1])/2 |
|
155
|
|
|
try: |
|
156
|
|
|
return float(s) |
|
157
|
|
|
except ValueError: |
|
158
|
|
|
return np.nan |
|
159
|
|
|
|
|
160
|
|
|
|
|
161
|
|
|
|
|
162
|
1 |
|
if __name__ == '__main__': |
|
163
|
|
|
logging.basicConfig(level=logging.INFO) |
|
164
|
|
|
LOGGER.info('Converting PhysProp Dataset...') |
|
165
|
|
|
PhysPropConverter.convert() |
|
166
|
|
|
|
The coding style of this project requires that you add a docstring to this code element. Below, you find an example for methods:
If you would like to know more about docstrings, we recommend to read PEP-257: Docstring Conventions.