def cp_colocalization(r, first, second, threshold=0.15, costes=False):
"""Measures overlap, k1/k2, manders, and rank weighted colocalization coefficients.
References:
http://www.scian.cl/archivos/uploads/1417893511.1674 starting at slide 35
Singan et al. (2011) "Dual channel rank-based intensity weighting for quantitative
co-localization of microscopy images", BMC Bioinformatics, 12:407.
"""
results = []
A = masked(r, first).astype(float)
B = masked(r, second).astype(float)
filt = A > 0
if filt.sum() == 0:
return np.nan
A = A[filt]
B = B[filt]
A_thresh, B_thresh = (threshold * A.max(), threshold * B.max())
A_total, B_total = A[A > A_thresh].sum(), B[B > B_thresh].sum()
mask = (A > A_thresh) & (B > B_thresh)
overlap = (A[mask] * B[mask]).sum() / np.sqrt(
(A[mask]**2).sum() * (B[mask]**2).sum())
results.append(overlap)
K1 = (A[mask] * B[mask]).sum() / (A[mask]**2).sum()
K2 = (A[mask] * B[mask]).sum() / (B[mask]**2).sum()
results.extend([K1, K2])
M1 = A[mask].sum() / A_total
M2 = B[mask].sum() / B_total
results.extend([M1, M2])
A_ranks = rankdata(A, method='dense')
B_ranks = rankdata(B, method='dense')
R = max([A_ranks.max(), B_ranks.max()])
weight = ((R - abs(A_ranks - B_ranks)) / R)[mask]
RWC1 = (A[mask] * weight).sum() / A_total
RWC2 = (B[mask] * weight).sum() / B_total
results.extend([RWC1, RWC2])
if costes:
A_costes, B_costes = costes_threshold(A, B)
mask_costes = (A > A_costes) & (B > B_costes)
C1 = A[mask_costes].sum() / A[A > A_costes].sum()
C2 = B[mask_costes].sum() / B[B > B_costes].sum()
results.extend([C1, C2])
return results
def lstsq_slope(r,first,second): A = masked(r,first) B = masked(r,second) filt = A > 0 if filt.sum() == 0: return np.nan A = A[filt] B = B[filt] slope = np.linalg.lstsq(np.vstack([A,np.ones(len(A))]).T,B,rcond=-1)[0][0] return slope
return np.multiply(image,mask)
def mahotas_zernike(r,channel):
image = masked_rect(r,channel)
mfeat = mahotas.features.zernike_moments(image.astype(np.uint32), radius = 9, degree=9)
return mfeat
def mahotas_pftas(r,channel):
image = masked_rect(r,channel)
mfeat = mahotas.features.pftas(image.astype(np.uint32))
### according to this, at least as good as haralick/zernike and much faster:
### https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-8-110
return mfeat
features_nuclear = {
'dapi_nuclear_min': lambda r: np.min(masked(r,0)),
'dapi_nuclear_25': lambda r: np.percentile(masked(r, 0),25),
'dapi_nuclear_mean': lambda r: masked(r, 0).mean(),
'dapi_nuclear_median': lambda r: np.median(masked(r, 0)),
'dapi_nuclear_75': lambda r: np.percentile(masked(r, 0),75),
'dapi_nuclear_max' : lambda r: masked(r, 0).max(),
'dapi_nuclear_int' : lambda r: masked(r, 0).sum(),
'dapi_nuclear_sd': lambda r: np.std(masked(r,0)),
'dapi_nuclear_mad': lambda r: median_absolute_deviation(masked(r,0),scale=1),
# 'dapi_zernike_nuclear': lambda r: mahotas_zernike(r,0),
# 'dapi_pftas_nuclear': lambda r: mahotas_pftas(r,0),
'dm1a_nuclear_min': lambda r: np.min(masked(r,1)),
'dm1a_nuclear_25': lambda r: np.percentile(masked(r, 1),25),
'dm1a_nuclear_mean' : lambda r: masked(r, 1).mean(),
'dm1a_nuclear_median' : lambda r: np.median(masked(r, 1)),
'dm1a_nuclear_75': lambda r: np.percentile(masked(r, 1),75),