def get_all_percentage_pairs(anno, chrom_hmm_anno, start, end, chrom_hmm_start,
chrom_hmm_end, chrom_hmm_len, shapes, bed, thresh,
signals_path, shape_str):
#Set up percentage matrix.
sum_matrix = np.zeros((2, len(shapes)))
cumulative_vec = np.zeros(len(shapes))
#Loop through bed file to compute percentage for each region.
current_start = -1
current_end = -1
current_clust = "none"
prev_start = -1
prev_end = -1
sigs = open(signals_path, "r")
junk = 0
for j in range(0, 2): #Loop through BED file twice.
next_signal = sigs.readline().split(",")
if j == 1:
next_signal = sigs.readline().split(",")
for i in range(0, bed.shape[0]):
#Get the previous data, if applicable.
if i > 0:
prev_start = int(bed[i - 1, start])
prev_end = int(bed[i - 1, end])
prev_clust = bed[i - 1, anno]
#Get the next element data.
next_line = bed[i, :]
current_start = int(next_line[start])
current_end = int(next_line[end])
current_clust = next_line[anno]
a = next_line[chrom_hmm_anno]
idx = shapes.index(current_clust)
clust_sig = [float(i) for i in shape_str[idx].split(",")]
#Get the signal data.
if (prev_start >= int(next_signal[1]) or current_start > int(
next_signal[1])) and current_start > prev_start:
next_signal = sigs.readline().split(",")
if current_start > int(next_signal[1]):
next_signal = sigs.readline().split(",")
if int(next_signal[1]) == current_start:
#Add to the existing percentages.
region = [float(i) for i in next_signal[3:len(next_signal)]]
count_clust = wsu.count_above(thresh, "", clust_sig, 0,
len(clust_sig) * BIN_SIZE, 0, 0,
BIN_SIZE)
count_a = wsu.count_above(thresh, a, region, current_start,
current_end,
int(next_line[chrom_hmm_start]),
int(next_line[chrom_hmm_end]),
BIN_SIZE)
if a == "AE" or a == "OE":
sum_matrix[0, idx] += int(next_line[chrom_hmm_len]) if (
count_clust == 0) else count_a
elif a != "0" and a != "AP" and a != "OP" and a != "GE" and a != "TS":
sum_matrix[1, idx] += int(next_line[chrom_hmm_len]) if (
count_clust == 0) else count_a
#Add to the total sum if the current start and end are not equal to the previous ones.
#if(prev_start != current_start and prev_start != "-1"):
cumulative_vec[idx] = np.sum(sum_matrix[:, idx])
#Get the set of percentages.
cumulative_matrix = np.tile(cumulative_vec, (2, 1))
return [sum_matrix / cumulative_matrix, np.sum(sum_matrix, 0)]
def get_labels_and_ground_truth(bed_file, sig_file, wig, annotations,
threshold):
#Set up percentage matrix.
vec_pred = list()
vec_gt = list()
final_stack_pred = np.empty((0, 0))
final_stack_gt = np.empty((0, 0))
#Get scores and labels for each bed file.
bed = np.genfromtxt(bed_file, delimiter='\t', dtype=str)
sigf = open(sig_file, "r")
#Loop through bed file to compute percentage for each region.
current_start = -1
current_end = -1
prev_start = -1
prev_end = -1
sig_i = 0
#Do not move forward if the first line is blank in the sig file.
try:
sig = [float(s) for s in sigf.readline().split(",")]
sum_vec = np.zeros(3)
#Keep track of regions with no ChromHMM annotations.
#These regions will not be used in the analysis.
not_annotated_count = 0
count_in_region = 0
for i in range(0, bed.shape[0]):
#Get the next element data.
next_line = bed[i, :]
current_start = int(next_line[1])
current_end = int(next_line[2])
a = next_line[8]
anno_start = int(next_line[6])
anno_end = int(next_line[7])
our_anno = next_line[3]
anno_length = int(next_line[9])
#Get next signals if needed.
#If we are still on the same region, don't get it.
if current_start != prev_start:
if sig_i != 0:
sig_s = sigf.readline()
sig = [float(s) for s in sig_s.split(",")]
sum_vec = np.zeros(3)
sig_i += 1
#Add to the existing percentages.
#If the region has peaks, consider only regions above peak threshold.
#If no peaks exist, consider entire region.
total_peak_size = wsu.count_above(threshold, "", sig,
current_start, current_end,
current_start, current_end,
BIN_SIZE)
if a == "1_TssA" or a == "2_TssAFlnk" or a == "10_TssBiv" or a == "11_BivFlnk":
if total_peak_size > 0:
sum_vec[0] += wsu.count_above(threshold, a, sig,
current_start, current_end,
anno_start, anno_end,
BIN_SIZE)
else:
sum_vec[0] += anno_length
elif a == "6_EnhG" or a == "7_Enh" or a == "12_EnhBiv":
if total_peak_size > 0:
sum_vec[1] += wsu.count_above(threshold, a, sig,
current_start, current_end,
anno_start, anno_end,
BIN_SIZE)
else:
sum_vec[1] += anno_length
elif a == "9_Het" or a == "15_Quies":
if total_peak_size > 0:
sum_vec[2] += wsu.count_above(threshold, a, sig,
current_start, current_end,
anno_start, anno_end,
BIN_SIZE)
else:
sum_vec[2] += anno_length
#This case is when there is no annotation. Do not count it.
else:
not_annotated_count += 1
count_in_region += 1
#Add the ground truth and predicted value based on the region with the maximum count above the threshold.
next_start = current_start + 1
if i + 1 < len(bed):
next_start = int(bed[i + 1, :][1])
if next_start != current_start and not_annotated_count != count_in_region:
#Add another element to the ground truth and prediction vectors.
vec_gt.append(np.zeros(len(sum_vec)))
vec_pred.append(np.zeros(len(sum_vec)))
max = np.argmax(sum_vec)
for sum in range(0, len(sum_vec)):
#Add ground truth.
if sum == max:
vec_gt[len(vec_gt) - 1][sum] = 1
else:
vec_gt[len(vec_gt) - 1][sum] = 0
#Add predictions.
if annotations[sum] == our_anno:
vec_pred[len(vec_pred) - 1][sum] = 1
else:
vec_pred[len(vec_pred) - 1][sum] = 0
#If it is unknown according to our analysis, do not consider it.
#This includes cases where there is no annotation from ChromHMM or where
#There is signal above the threshold but no ChromHMM annotation in the signal.
if vec_pred[len(vec_pred) - 1][0] == 0 and vec_pred[
len(vec_pred) - 1][1] == 0 and vec_pred[len(vec_pred) -
1][2] == 0:
del vec_pred[len(vec_pred) - 1]
del vec_gt[len(vec_pred) - 1]
elif sum_vec[0] == 0 and sum_vec[1] == 0 and sum_vec[2] == 0:
del vec_pred[len(vec_pred) - 1]
del vec_gt[len(vec_pred) - 1]
#Set count and unannotated count to 0. Do the same for summation vec.
not_annotated_count = 0
count_in_region = 0
#Get the previous data, if applicable.
prev_start = current_start
prev_end = current_end
#Stack all values.
final_stack_pred = np.stack(vec_pred)
final_stack_gt = np.stack(vec_gt)
except:
pass
#Return value.
return [final_stack_pred, final_stack_gt]
def get_tss_labels_and_ground_truth(bed_file, sig_file, wig, annotations,
threshold):
#Set up counts of promoter and non-promoter.
vec_pred = list()
vec_gt = list()
bed = np.genfromtxt(bed_file, delimiter='\t', dtype=str)
sigs = np.genfromtxt(sig_file, delimiter=',', dtype=float)
#Loop through bed file to compute percentage for each region.
current_start = -1
current_end = -1
prev_start = -1
prev_end = -1
sig = sigs[0, :]
sig_i = -1
sum_vec = np.zeros(2)
#Keep track of regions with no ChromHMM annotations.
#These regions will not be used in the analysis.
not_annotated_count = 0
count_in_region = 0
for i in range(0, bed.shape[0]):
#Get the next element data.
next_line = bed[i, :]
current_start = int(next_line[1])
current_end = int(next_line[2])
a = next_line[7]
anno_start = int(next_line[5])
anno_end = int(next_line[6])
our_anno = next_line[3]
anno_length = int(next_line[8])
#Get next signals if needed.
#If we are still on the same region, don't get it.
if current_start != prev_start:
sig_i += 1
sig = sigs[sig_i, :]
sum_vec = np.zeros(2)
#Add to the existing percentages.
#If the region has peaks, consider only regions above peak threshold.
#If no peaks exist, consider entire region.
total_peak_size = wsu.count_above(threshold, "", sig, current_start,
current_end, current_start,
current_end, BIN_SIZE)
if a == "1_TssA" or a == "2_TssAFlnk" or a == "10_TssBiv" or a == "11_BivFlnk":
if total_peak_size > 0:
sum_vec[0] += wsu.count_above(threshold, a, sig, current_start,
current_end, anno_start,
anno_end, BIN_SIZE)
else:
sum_vec[0] += anno_length
elif anno_length != 0:
if total_peak_size > 0:
sum_vec[1] += wsu.count_above(threshold, a, sig, current_start,
current_end, anno_start,
anno_end, BIN_SIZE)
else:
sum_vec[1] += anno_length
#This case is when there is no annotation. Do not count it.
else:
not_annotated_count += 1
count_in_region += 1
#Add the ground truth and predicted value based on the region with the maximum count above the threshold.
next_start = current_start + 1
if i + 1 < len(bed):
next_start = int(bed[i + 1, :][1])
if next_start != current_start and not_annotated_count != count_in_region:
vec_gt.append(np.zeros(len(sum_vec)))
vec_pred.append(np.zeros(len(sum_vec)))
max = np.argmax(sum_vec)
for sum in range(0, len(sum_vec)):
#Add ground truth.
if sum == max:
vec_gt[len(vec_gt) - 1][sum] = 1
else:
vec_gt[len(vec_gt) - 1][sum] = 0
#Add predictions.
if annotations[sum] == our_anno:
vec_pred[len(vec_pred) - 1][sum] = 1
else:
vec_pred[len(vec_pred) - 1][sum] = 0
if sum_vec[0] == 0 and sum_vec[1] == 0:
del vec_pred[len(vec_pred) - 1]
del vec_gt[len(vec_pred) - 1]
#Set count and unannotated count to 0.
not_annotated_count = 0
count_in_region = 0
#Get the previous data, if applicable.
prev_start = current_start
prev_end = current_start
#Return value.
return [np.stack(vec_pred), np.stack(vec_gt)]
def get_all_percentage_pairs(anno, chrom_hmm_anno, start, end, chrom_hmm_start,
chrom_hmm_end, chrom_hmm_len, magnitudes, bed,
thresh, signals_path):
#Set up percentage matrix.
sum_matrix = np.zeros((4, len(magnitudes)))
cumulative_vec = np.zeros(len(magnitudes))
#Loop through bed file to compute percentage for each region.
current_start = -1
current_end = -1
current_clust = "none"
prev_start = -1
prev_end = -1
sigs = open(signals_path, "r")
junk = 0
for j in range(0, 2): #Loop through BED file twice.
next_signal = sigs.readline().split(",")
if j == 1:
next_signal = sigs.readline().split(",")
for i in range(0, bed.shape[0]):
if len(next_signal) > 1:
#Get the previous data, if applicable.
if i > 0:
prev_start = int(bed[i - 1, start])
prev_end = int(bed[i - 1, end])
#Get the next element data.
next_line = bed[i, :]
current_start = int(next_line[start])
current_end = int(next_line[end])
current_clust = next_line[anno]
a = next_line[chrom_hmm_anno]
idx = magnitudes.index(current_clust)
#Get the signal data.
if (prev_start >= int(next_signal[1]) or current_start > int(
next_signal[1])) and current_start > prev_start:
next_signal = sigs.readline().split(",")
if len(next_signal) > 1:
if current_start > int(next_signal[1]):
next_signal = sigs.readline().split(",")
if len(next_signal) > 1:
if int(next_signal[1]) == current_start:
#Add to the existing percentages.
region = [
float(i)
for i in next_signal[3:len(next_signal)]
]
count_a = wsu.count_above(
thresh, a, region, current_start, current_end,
int(next_line[chrom_hmm_start]),
int(next_line[chrom_hmm_end]), BIN_SIZE)
if a == "1_TssA" or a == "2_TssAFlnk" or a == "10_TssBiv" or a == "11_BivFlnk":
sum_matrix[0, idx] += int(
next_line[chrom_hmm_len]
) if (int(current_clust) < thresh) else count_a
elif a == "6_EnhG" or a == "7_Enh" or a == "12_EnhBiv":
sum_matrix[1, idx] += int(
next_line[chrom_hmm_len]
) if (int(current_clust) < thresh) else count_a
elif a == "13_ReprPC" or a == "ReprPCWk":
sum_matrix[2, idx] += int(
next_line[chrom_hmm_len]
) if (int(current_clust) < thresh) else count_a
elif a == "9_Het" or a == "15_Quies":
sum_matrix[3, idx] += int(
next_line[chrom_hmm_len]
) if (int(current_clust) < thresh) else count_a
#Add to the total sum if the current start and end are not equal to the previous ones.
#if(prev_start != current_start and prev_start != "-1"):
cumulative_vec[idx] = np.sum(sum_matrix[:, idx])
#Get the set of percentages.
cumulative_matrix = np.tile(cumulative_vec, (4, 1))
return [sum_matrix / cumulative_matrix, np.sum(sum_matrix, 0)]