File size: 13,709 Bytes
fae1173 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 | import numpy as np
import pandas as pd
import bokeh.io
import bokeh.plotting
import bokeh.palettes
from bokeh.transform import factor_cmap
import datashader
import holoviews as hv
import holoviews.operation.datashader
hv.extension("bokeh")
import warnings
# Suppress FutureWarning messages
warnings.simplefilter(action='ignore',)
cr9114_dict = {
'ab_name' : 'CR9114',
'files' : {
'ESM-1v Ab-Ag': 'output/ab_mutagenesis_expts/cr9114/cr9114_esm1vAbAg_exp_data_maskMargLabeled.csv',
'ESM-1v Ab only': 'output/ab_mutagenesis_expts/cr9114/cr9114_esm1vbothchains_exp_data_maskMargLabeled.csv',
'ESM-1v Ab VH only': 'output/ab_mutagenesis_expts/cr9114/cr9114_exp_data_maskMargLabeled.csv',
},
'dms_df' : pd.read_csv('data/ab_mutagenesis_expts/cr9114/cr9114_exp_data.csv',
dtype = {'genotype': str},
).rename(columns={'h1_mean': 'H1', 'h3_mean' : 'H3'}),
'ag_columns': ['H1', 'H3'],
'expt_type': 'Combinatorial Mutagenesis for Affinity',
'palette': bokeh.palettes.Spectral6
}
cr6261_dict = {
'ab_name' : 'CR6261',
'files' : {
'ESM-1v Ab-Ag': 'output/ab_mutagenesis_expts/cr6261/cr6261_esm1vAbAg_exp_data_maskMargLabeled.csv',
'ESM-1v Ab only': 'output/ab_mutagenesis_expts/cr6261/cr6261_esm1vbothchains_exp_data_maskMargLabeled.csv',
'ESM-1v Ab VH only': 'output/ab_mutagenesis_expts/cr6261/cr6261_exp_data_maskMargLabeled.csv',
},
'dms_df' : pd.read_csv('data/ab_mutagenesis_expts/cr6261/cr6261_exp_data.csv',
dtype={'genotype': str},
).rename(columns={'h1_mean': 'H1', 'h9_mean' : 'H9'}),
'ag_columns': ['H1', 'H9'],
'expt_type': 'Combinatorial Mutagenesis for Affinity',
'palette': bokeh.palettes.Dark2_6,
}
g6LC_dict = {
'ab_name' : 'g6',
'files' : {
'LM Ab-Ag': 'output/ab_mutagenesis_expts/g6/g6Lc_esm1vAbAg_exp_data_maskMargLabeled.csv',
'LM Ab only': 'output/ab_mutagenesis_expts/g6/g6Lc_esm1vbothchains_exp_data_maskMargLabeled.csv',
'LM Ab VH/VL only': 'output/ab_mutagenesis_expts/g6/g6Lc_exp_data_maskMargLabeled.csv',
},
'dms_df' : pd.read_csv('data/ab_mutagenesis_expts/g6/g6_lc_exp_data.csv'),
'ag_columns': ['norm_binding'],
'expt_type': 'Deep Mutational Scan for Binding',
'palette': bokeh.palettes.Pastel1_6,
'chain': 'VL'
}
g6HC_dict = {
'ab_name' : 'g6',
'files' : {
'LM Ab-Ag': 'output/ab_mutagenesis_expts/g6/g6Hc_esm1vAbAg_exp_data_maskMargLabeled.csv',
'LM Ab only': 'output/ab_mutagenesis_expts/g6/g6Hc_esm1vbothchains_exp_data_maskMargLabeled.csv',
'LM Ab VH/VL only': 'output/ab_mutagenesis_expts/g6/g6Hc_exp_data_maskMargLabeled.csv',
},
'dms_df' : pd.read_csv('data/ab_mutagenesis_expts/g6/g6_hc_exp_data.csv'),
'ag_columns': ['norm_binding'],
'expt_type': 'Deep Mutational Scan for Binding',
'palette': bokeh.palettes.Pastel1_4,
'chain': 'VH'
}
def apply_mask_and_average(scores, genotype):
#scores here should already be a in log space
if '1' in genotype:
masked_list = []
for s, mask_char in zip(scores, genotype):
if mask_char == '1':
masked_list.append(s)
multi_avg = np.mean(masked_list)
else:
#if genotype is all zeros = wt
multi_avg = 0
return multi_avg
def transform_single_to_multi( dms_df, singleMuts_df, condition, sort_col, ascending ):
multi_scores = []
#use input parameter as method bc key is kwarg in pd.sort_values
#method should be either esm1v or abysis or AbLang
#for cr sorts the single mutation dataframe in residue order, ie binary '10000' before '00010'. ascending = False
#for g6 sorts the single mutation dataframe in position order, ie residue index. ascending = True
singleMuts_sorted = singleMuts_df.sort_values(sort_col, key=lambda x: x.astype(int), ascending=ascending)
singles_scores = singleMuts_sorted[condition].to_list()
for g in dms_df['genotype']:
multi_scores.append(apply_mask_and_average(singles_scores, g ))
return multi_scores
#get melted correlations matrix of structure input x target antigen, for a given antibody
def get_corr(ab_name, files, dms_df, ag_columns, dropLOQ = False ):
#retreive correlations from abysis and esm1v scores
for key, filepath in files.items():
singleMuts_df = pd.read_csv(filepath)
# Get the columns for all esm1v models
esm_columns = [col for col in singleMuts_df.columns if col.startswith('esm')]
# Calculate the average for each esm column
esm_avg_values = singleMuts_df[esm_columns].mean(axis=1)
singleMuts_df[key] = esm_avg_values
if ab_name == 'g6':
#g6 data set is only single mutations
dms_df[key] = esm_avg_values
else:
#cr datasets are multiple mutations
dms_df[key] = transform_single_to_multi(dms_df, singleMuts_df, key, 'genotype', ascending = False)
conditions = list(files.keys())
if not dropLOQ:
correlations = dms_df[ag_columns + conditions].corr(method='spearman')
correlations = correlations.drop(conditions, axis= 1)
correlations = correlations.drop(ag_columns, axis= 0)
# Melt the correlations DataFrame and rename the index column
melted_correlations = (
correlations
.reset_index()
.melt(id_vars='index', var_name='Target', value_name='Correlation')
.rename(columns={'index': 'Input'})
)
melted_correlations['Method'] = ['LM' for i in melted_correlations['Input']]
return melted_correlations
#if we wish to drop the samples on the lower limit of quantitation, drop samples from each ag group individually and compute correlation with IF log_like
else:
all_ag_melted = pd.DataFrame({})
for ag in ag_columns:
filt_df = dms_df[dms_df[ag] > min(dms_df[ag])]
correlations = filt_df[[ag] + conditions].corr(method='spearman')
correlations = correlations.drop(conditions, axis= 1)
correlations = correlations.drop(ag, axis= 0)
# Melt the correlations DataFrame and rename the index column
melted_correlations = (
correlations
.reset_index()
.melt(id_vars='index', var_name='Target', value_name='Correlation')
).rename(columns={'index': 'Input'})
all_ag_melted = pd.concat([all_ag_melted, melted_correlations], ignore_index= True)
all_ag_melted['Method'] = ['LM' for i in all_ag_melted['Input']]
return all_ag_melted
#get melted correlations matrix of structure input x target antigen, for a given antibody
def get_g6_corr(g6Hc_dict, g6LC_dict, dropLOQ = False ):
ab_name = 'g6'
ag_columns = g6LC_dict['ag_columns']
vh_and_vl_dms_df = pd.DataFrame({})
for d in [g6HC_dict, g6LC_dict]:
files = d['files']
dms_df = d['dms_df']
#retreive correlations from abysis, ablang, and esm1v scores
for key, filepath in files.items():
singleMuts_df = pd.read_csv(filepath)
# Get the columns for all esm1v models
esm_columns = [col for col in singleMuts_df.columns if col.startswith('esm')]
# Calculate the average for each esm column
esm_avg_values = singleMuts_df[esm_columns].mean(axis=1)
singleMuts_df[key] = esm_avg_values
if ab_name == 'g6':
#g6 data set is only single mutations
dms_df[key] = esm_avg_values
vh_and_vl_dms_df = pd.concat([vh_and_vl_dms_df, dms_df], ignore_index= True)
conditions = list(files.keys())
#
if not dropLOQ:
correlations = vh_and_vl_dms_df[ag_columns + conditions].corr(method='spearman')
correlations = correlations.drop(conditions, axis= 1)
correlations = correlations.drop(ag_columns, axis= 0)
# Melt the correlations DataFrame and rename the index column
melted_correlations = (
correlations
.reset_index()
.melt(id_vars='index', var_name='Target', value_name='Correlation')
.rename(columns={'index': 'Input'})
)
melted_correlations['Method'] = ['LM' for i in melted_correlations['Input']]
melted_correlations['Target'] = ['VEGF-A'] * len(melted_correlations)
return melted_correlations
#if we wish to drop the samples on the lower limit of quantitation, drop samples from each ag group individually and compute correlation with IF log_like
else:
all_ag_melted = pd.DataFrame({})
for ag in ag_columns:
filt_df = vh_and_vl_dms_df[vh_and_vl_dms_df[ag] > min(vh_and_vl_dms_df[ag])]
correlations = filt_df[[ag] + conditions].corr(method='spearman')
correlations = correlations.drop(conditions, axis= 1)
correlations = correlations.drop(ag, axis= 0)
# Melt the correlations DataFrame and rename the index column
melted_correlations = (
correlations
.reset_index()
.melt(id_vars='index', var_name='Target', value_name='Correlation')
).rename(columns={'index': 'Input'})
all_ag_melted = pd.concat([all_ag_melted, melted_correlations], ignore_index= True)
all_ag_melted['Method'] = ['LM' for i in melted_correlations['Input']]
melted_correlations['Target'] = ['VEGF-A'] * len(melted_correlations)
return all_ag_melted
#plot correlation bars for cr antibodies: cr6261 and cr9114
def plot_hbar(title, melted_correlations, palette = bokeh.palettes.Spectral6, compare = 'model'):
if compare == 'model':
melted_correlations.replace('Ab-Ag','InverseFolding', inplace = True)
color_by = 'Input'
elif compare == 'IF':
color_by = 'Method'
else:
color_by = 'Target'
melted_correlations['cats'] = melted_correlations.apply(lambda x: (x["Target"], x["Input"]), axis = 1)
factors = list(melted_correlations.cats)[::-1]
p = bokeh.plotting.figure(
height=340,
width=440,
x_axis_label="Spearman Correlation",
x_range=[0, 1],
y_range=bokeh.models.FactorRange(*factors),
tools="save",
title = title
)
p.hbar(
source=melted_correlations,
y="cats",
right="Correlation",
height=0.6,
line_color = 'black',
legend_field = color_by,
# use the palette to colormap based on the the x[1:2] values
# fill_color=factor_cmap(
# color_by,
# palette= palette,
# factors = list(melted_correlations[color_by].unique()),
# start=1,
# end=3)
fill_color=bokeh.palettes.Dark2_6[1]
)
#labels_df = melted_correlations[melted_correlations['Input'].isin(['Ab-Ag','ESM-1v', 'abYsis', 'InverseFold'])]
labels_df = melted_correlations
labels_df['corr_str'] = labels_df['Correlation'].apply(lambda x: round(x, 2)).astype(str)
labels_source = bokeh.models.ColumnDataSource(labels_df)
labels = bokeh.models.LabelSet(x='Correlation', y='cats', text='corr_str',text_font_size = "10px",
x_offset=12, y_offset=-5, source=labels_source, render_mode='canvas')
p.ygrid.grid_line_color = None
p.y_range.range_padding = 0.1
p.add_layout(labels)
p.legend.visible = False
p.output_backend = "svg"
return p
if __name__ == '__main__':
datasets = [cr9114_dict, cr6261_dict, (g6HC_dict, g6LC_dict) ]
#compute with and without points on lower limit of quantitation ignored
for dropLOQ in [False]:
all_corr_plots = []
for d in datasets:
if type(d) is tuple:
g6Hc, g6Lc = d
title = g6Hc['ab_name'] + ', ' + g6Hc['expt_type']
g6_combined = get_g6_corr(g6Hc, g6Lc, dropLOQ= dropLOQ)
all_corr_plots.append(plot_hbar(title, g6_combined, g6Lc['palette'], 'IF'))
else:
corr_df = get_corr(d['ab_name'], d['files'], d['dms_df'], d['ag_columns'], dropLOQ= dropLOQ)
title = d['ab_name'] + ', ' + d['expt_type']
all_corr_plots.append(plot_hbar(title, corr_df, d['palette'], 'Target'))
all_corr_fname = f"output/ab_mutagenesis_expts/esm1v_benchmarks{'_dropLOQ' if dropLOQ else ''}.html"
bokeh.plotting.output_file(all_corr_fname)
bokeh.io.show(bokeh.layouts.gridplot(all_corr_plots, ncols = len(all_corr_plots)))
|