Skip to content

Commit

Permalink
depecrated check_fasta
Browse files Browse the repository at this point in the history
  • Loading branch information
joaomcteixeira committed Apr 5, 2018
1 parent c4fdf5f commit 80a46d4
Showing 1 changed file with 1 addition and 38 deletions.
39 changes: 1 addition & 38 deletions core/fslibs/FarseerCube.py
Original file line number Diff line number Diff line change
Expand Up @@ -369,13 +369,6 @@ def load_experiments(self, filetype='.csv', resonance_type='Backbone'):
format(filetype)
self.log_r(fsw.gen_wet('ERROR', msg, 14))
self.abort()

# if filetype == '.fasta' and resonance_type == 'Backbone':
# self.check_fasta(
# target[parts[1]][parts[2]][lessparts],
# parts[2],
# p
# )

self.checks_xy_datapoints_coherency(target, filetype)

Expand Down Expand Up @@ -463,7 +456,6 @@ def read_FASTA(self, FASTApath):
)
logs = ' * {}-{}-{}'.format(self.FASTAstart, FASTA, dd['ResNo'][-1])
self.log_r(logs)
#self.check_fasta(df, FASTApath)

return df

Expand Down Expand Up @@ -1588,36 +1580,7 @@ def check_ref_res(self, series, ref_res):
self.abort()

return

def check_fasta(self, df, yy, fasta_path):
"""
Checks if loaded FASTA file has more residues than the reference
experiment.
FASTA cannot has less rows than the reference experiment.
WET#18
Parameters:
df (pd.DataFrame): contains the FASTA loaded data in
DataFrame format as prepared by .read_FASTA().
yy (str): the current YY data point name.
fasta_path (srt): the .fasta file path.
"""

if df.shape[0] \
< self.allpeaklists[self.zzref][yy][self.xxref].\
shape[0]:
msg = \
'The .fasta file in {} has less residue entries than the protein sequence \
of the reference experiment [{}][{}][{}]'.\
format(fasta_path, self.zzref, yy, self.xxref)
self.log_r(fsw.gen_wet('ERROR', msg, 18))
self.abort()

return


def compare_fastas(self):
"""
Compares all .fasta files to confirm they have the same size.
Expand Down

0 comments on commit 80a46d4

Please sign in to comment.