PyTFBS

A Python package for predicting transcription factor binding sites.

Installation

pip install torch numpy pandas scipy PyTFBS

Usage

from PyTFBS import motif, predict, clink
import random
random.seed(42)

# download PyTFBS data, only need to run once!!!
motif.download_data()

# list available models
motif.list_models(species='Homo sapiens', accuracy=0.9, sensitivity=0.8, specificity=0.9, precision=0.8, f1=0.8)

# get avaiable motifs
motifs = motif.get_motifs(species='Homo sapiens', accuracy=0.9, sensitivity=0.8, specificity=0.9, precision=0.8, f1=0.8)
print(motifs)

# get models based on motif name
models = motif.get_models('RFX2_HUMAN.H11MO.0.A')
print(models)

# predict one model
predict.script('CEBPB_HUMAN.H11MO.0.A', 'CEBPB_HUMAN.H11MO.0.A', 'input_seq_file.fasta', 'out_file.txt')

# speed up using mutil-threading (for Windows OS only)
predict.win_bin('CEBPB_HUMAN.H11MO.0.A', 'CEBPB_HUMAN.H11MO.0.A', 'input_seq_file.fasta', 64, 'out_file.txt')

# run prediction with user motif data
# the my_motif_dir should be organized as [[motif], [trace], [par]]
predict.script('CEBPB_HUMAN.H11MO.0.A', 'CEBPB_HUMAN.H11MO.0.A', 'input_seq_file.fasta', data_dir='my_motif_dir')
predict.win_bin('CEBPB_HUMAN.H11MO.0.A', 'CEBPB_HUMAN.H11MO.0.A', 'input_seq_file.fasta', 64, data_dir='my_motif_dir')

# create the CLink TF-target set

# rexp_files = get_rexp_files()
# print(rexp_files)

df_rexps = clink.read_rexp_file()
tf_cnames = clink.read_tf_cname()
tf_code = 'CREB1_HUMAN.H11MO.0.A_RC'
tf_tars = clink.read_PyTFBS_output(tf_code + '_output.txt')

for tf_cname in tf_cnames[tf_code]:
	if tf_cname[1] not in df_rexps.index:
		continue
	else:
		tf_exp = df_rexps.loc[tf_cname[1]].values.tolist()
	df_rexps_tars = df_rexps[df_rexps.index.isin(tf_tars)]
	tar_exps = {idx: row.tolist() for idx, row in df_rexps_tars.iterrows()}
	rcis = clink.rci(tar_exps, tf_exp)
	for rci in rcis:
		score = rci[1] + (random.uniform(1E-10, 1E-15) * random.choice([1, -1]))
		print(tf_cname[0] + '\t' + rci[0] + '\t' + str(score))