def __init__(self, generative_model, genomic_data):
"""Initialize SequenceGenerationVJ
This intialization computes all of the cumulative probability
distributions that will be needed for efficient Monte Carlo sequence
generation out of a GenerativeModelVJ.
Parameters
----------
generative_model : GenerativeModelVJ
VJ generative model class containing the model parameters.
genomic_data : GenomicDataVJ
VJ genomic data class containing the V and J germline
sequences and info.
"""
self.CPVJ = (generative_model.PVJ /
np.sum(generative_model.PVJ)).flatten().cumsum()
self.CPinsVJ = (generative_model.PinsVJ /
np.sum(generative_model.PinsVJ)).cumsum()
for V in range(generative_model.PdelV_given_V.shape[1]):
if np.sum(generative_model.PdelV_given_V[:, V]) > 0:
generative_model.PdelV_given_V[:,
V] = generative_model.PdelV_given_V[:, V] / np.sum(
generative_model.
PdelV_given_V[:, V])
self.given_V_CPdelV = generative_model.PdelV_given_V.T.cumsum(axis=1)
for J in range(generative_model.PdelJ_given_J.shape[1]):
if np.sum(generative_model.PdelJ_given_J[:, J]) > 0:
generative_model.PdelJ_given_J[:,
J] = generative_model.PdelJ_given_J[:, J] / np.sum(
generative_model.
PdelJ_given_J[:, J])
self.given_J_CPdelJ = generative_model.PdelJ_given_J.T.cumsum(axis=1)
self.C_Rvj = generative_model.Rvj.T.cumsum(axis=1)
if generative_model.first_nt_bias_insVJ == None:
first_nt_bias_insVJ = calc_steady_state_dist(generative_model.Rvj)
else:
first_nt_bias_insVJ = generative_model.first_nt_bias_insVJ
self.C_first_nt_bias_insVJ = first_nt_bias_insVJ.cumsum()
self.num_J_genes = generative_model.PVJ.shape[1]
self.cutV_genomic_CDR3_segs = genomic_data.cutV_genomic_CDR3_segs
self.cutJ_genomic_CDR3_segs = genomic_data.cutJ_genomic_CDR3_segs
def __init__(self, generative_model, genomic_data, alphabet_file = None):
"""Initialize PreprocessedParametersVJ
This intialization computes all of the attributes that will be needed
for Pgen computation of CDR3 sequences generated by VJ recombination.
Parameters
----------
generative_model : GenerativeModelVJ
VJ generative model class containing the model parameters.
genomic_data : GenomicDataVJ
VJ genomic data class containing the V and J germline
sequences and info.
alphabet_file : str, optional
File name (full pathing from current directory) for a custom alphabet
definition. If no file is provided, the default alphabet is used, i.e.
standard amino acids, undetermined amino acids (B, J, X, and Z), and
single codon symbols.
"""
PreprocessedParameters.__init__(self, generative_model, genomic_data, alphabet_file)
#Check that the generative_model and genomic_data are VDJ
if not all([generative_model.__class__.__name__.endswith('VJ'),genomic_data.__class__.__name__.endswith('VJ')]):
raise ValueError #Need VJ model and data
self.PVJ = generative_model.PVJ
#Set the V genomic Pi arrays
self.PVdelV_nt_pos_vec = None
self.PVdelV_2nd_nt_pos_per_aa_vec = None
self.generate_PVdelV_nt_pos_vecs(generative_model, genomic_data) #Computes and sets the above attributes
#Set the J genomic Pi arrays
self.PJdelJ_nt_pos_vec = None
self.PJdelJ_2nd_nt_pos_per_aa_vec = None
self.generate_PJdelJ_nt_pos_vecs(generative_model, genomic_data) #Computes and sets the above attributes
#Trim, then zeropad PinsVJ
self.PinsVJ = np.append(np.trim_zeros(generative_model.PinsVJ, 'b'), [0., 0., 0., 0.])
#insVJ Dinucleotide bias transition matrix
#Note, this used to be called 'RnucleotideVJ_per_nucleotideVJ_5prime'
self.Rvj = generative_model.Rvj
#Check if first insertion nt probability distribution is given.
#Note, this is not normally inferred by IGoR. If the nt bias dist isn't
#present, use the steady-state distributions defined from Rvj.
#When not using the steady-state distributions we need to compute the
#nt bias for the position previous to first_nt_bias (R^{-1}p_0)
if generative_model.first_nt_bias_insVJ == None:
self.first_nt_bias_insVJ = calc_steady_state_dist(self.Rvj)
#In steady-state zero_nt_bias_insVJ = first_nt_bias_insVJ so no need to recalculate
self.zero_nt_bias_insVJ = self.first_nt_bias_insVJ
else:
self.first_nt_bias_insVJ = generative_model.first_nt_bias_insVJ
#Require Rvj*zero_nt_bias_insVJ = first_nt_bias_insVJ --- can use pseudo-inverse
self.zero_nt_bias_insVJ = np.dot(np.linalg.pinv(self.Rvj), generative_model.first_nt_bias_insVJ)
#Compute VJ insertion transfer matrices
self.Tvj = None
self.Svj = None
self.Dvj = None
self.lTvj = None
self.lDvj = None
self.generate_VJ_junction_transfer_matrices() #Computes and sets the transfer matrices
def __init__(self, generative_model, genomic_data, alphabet_file = None):
"""Initialize PreprocessedParametersVDJ
This intialization computes all of the attributes that will be needed
for Pgen computation of CDR3 sequences generated by VDJ recombination.
Parameters
----------
generative_model : GenerativeModelVDJ
VDJ generative model class containing the model parameters.
genomic_data : GenomicDataVDJ
VDJ genomic data class containing the V, D, and J germline
sequences and info.
alphabet_file : str, optional
File name (full pathing from current directory) for a custom alphabet
definition. If no file is provided, the default alphabet is used, i.e.
standard amino acids, undetermined amino acids (B, J, X, and Z), and
single codon symbols.
"""
PreprocessedParameters.__init__(self, generative_model, genomic_data, alphabet_file)
#Check that the generative_model and genomic_data are VDJ
if not all([generative_model.__class__.__name__.endswith('VDJ'),genomic_data.__class__.__name__.endswith('VDJ')]):
raise ValueError #Need VDJ model and data
self.cutD_genomic_CDR3_segs = genomic_data.cutD_genomic_CDR3_segs
self.D_allele_names = [D[0] for D in genomic_data.genD]
#Set the V genomic Pi arrays
self.PVdelV_nt_pos_vec = None
self.PVdelV_2nd_nt_pos_per_aa_vec = None
self.generate_PVdelV_nt_pos_vecs(generative_model, genomic_data) #Computes and sets the above attributes
#Set the D genomic Pi arrays and info
self.PD_nt_pos_vec = None
self.PD_2nd_nt_pos_per_aa_vec = None
self.min_delDl_given_DdelDr = None
self.max_delDl_given_DdelDr = None
self.zeroD_given_D = None
self.preprocess_D_segs(generative_model, genomic_data) #Computes and sets the above attributes
self.PdelDldelDr_given_D = generative_model.PdelDldelDr_given_D
#Set the J genomic Pi arrays and info
self.PJdelJ_nt_pos_vec = None
self.PJdelJ_2nd_nt_pos_per_aa_vec = None
self.generate_PJdelJ_nt_pos_vecs(generative_model, genomic_data) #Computes and sets the above attributes
#allow for 0 prob J genes
self.PD_given_J = np.zeros(generative_model.PDJ.shape)
self.PD_given_J[:, np.sum(generative_model.PDJ, axis = 0) > 0] = np.multiply(generative_model.PDJ[:, np.sum(generative_model.PDJ, axis = 0) > 0], 1/np.sum(generative_model.PDJ, axis = 0)[np.sum(generative_model.PDJ, axis = 0) > 0])
#Trim, then zeropad PinsVD and PinsDJ
self.PinsVD = np.append(np.trim_zeros(generative_model.PinsVD, 'b'), [0., 0., 0., 0.])
self.PinsDJ = np.append(np.trim_zeros(generative_model.PinsDJ, 'b'), [0., 0., 0., 0.])
#insVD and insDJ Dinucleotide bias transition matrices
#Note, these used to be called 'RnucleotideVD_per_nucleotideVD_5prime'
#and 'RnucleotideDJ_per_nucleotideDJ_3prime'
self.Rvd = generative_model.Rvd
self.Rdj = generative_model.Rdj
#Check if first insertion nt probability distributions are given.
#Note, this is not normally inferred by IGoR. If these nt bias dists
#aren't present, use the steady-state distributions defined from Rvd
#and Rdj. When not using the steady-state distributions we need to
#compute the nt bias for the position previous to first_nt_bias
#(R^{-1}p_0)
if generative_model.first_nt_bias_insVD == None:
self.first_nt_bias_insVD = calc_steady_state_dist(self.Rvd)
#In steady-state zero_nt_bias_insVD = first_nt_bias_insVD so no need to recalculate
self.zero_nt_bias_insVD = self.first_nt_bias_insVD
else:
self.first_nt_bias_insVD = generative_model.first_nt_bias_insVD
#Require Rvd*zero_nt_bias_insVD = first_nt_bias_insVD --- can use pseudo-inverse
self.zero_nt_bias_insVD = np.dot(np.linalg.pinv(self.Rvd), generative_model.first_nt_bias_insVD)
if generative_model.first_nt_bias_insDJ == None:
self.first_nt_bias_insDJ = calc_steady_state_dist(self.Rdj)
#In steady-state zero_nt_bias_insDJ = first_nt_bias_insDJ so no need to recalculate
self.zero_nt_bias_insDJ = self.first_nt_bias_insDJ
else:
self.first_nt_bias_insDJ = generative_model.first_nt_bias_insDJ
#Require Rdj*zero_nt_bias_insDJ = first_nt_bias_insDJ --- can use pseudo-inverse
self.zero_nt_bias_insDJ = np.dot(np.linalg.pinv(self.Rdj), generative_model.first_nt_bias_insDJ)
#Compute VD insertion transfer matrices
self.Tvd = None
self.Svd = None
self.Dvd = None
self.lTvd = None
self.lDvd = None
self.generate_VD_junction_transfer_matrices() #Computes and sets the transfer matrices
#Compute DJ insertion transfer matrices
self.Tdj = None
self.Sdj = None
self.Ddj = None
self.rTdj = None
self.rDdj = None
self.generate_DJ_junction_transfer_matrices() #Computes and sets the transfer matrices
def __init__(self, generative_model, genomic_data):
"""Initialize SequenceGenerationVDJ
This intialization computes all of the cumulative probability
distributions that will be needed for efficient Monte Carlo sequence
generation out of a GenerativeModelVDJ.
Parameters
----------
generative_model : GenerativeModelVDJ
VDJ generative model class containing the model parameters.
genomic_data : GenomicDataVDJ
VDJ genomic data class containing the V, D, and J germline
sequences and info.
"""
self.CPV = (generative_model.PV / np.sum(generative_model.PV)).cumsum()
self.CPDJ = (generative_model.PDJ /
np.sum(generative_model.PDJ)).flatten().cumsum()
self.CinsVD = (generative_model.PinsVD /
np.sum(generative_model.PinsVD)).cumsum()
self.CinsDJ = (generative_model.PinsDJ /
np.sum(generative_model.PinsDJ)).cumsum()
for V in range(generative_model.PdelV_given_V.shape[1]):
if np.sum(generative_model.PdelV_given_V[:, V]) > 0:
generative_model.PdelV_given_V[:,
V] = generative_model.PdelV_given_V[:, V] / np.sum(
generative_model.
PdelV_given_V[:, V])
self.given_V_CPdelV = generative_model.PdelV_given_V.T.cumsum(axis=1)
for J in range(generative_model.PdelJ_given_J.shape[1]):
if np.sum(generative_model.PdelJ_given_J[:, J]) > 0:
generative_model.PdelJ_given_J[:,
J] = generative_model.PdelJ_given_J[:, J] / np.sum(
generative_model.
PdelJ_given_J[:, J])
self.given_J_CPdelJ = generative_model.PdelJ_given_J.T.cumsum(axis=1)
for D in range(generative_model.PdelDldelDr_given_D.shape[2]):
if np.sum(generative_model.PdelDldelDr_given_D[:, :, D]) > 0:
generative_model.PdelDldelDr_given_D[:, :,
D] = generative_model.PdelDldelDr_given_D[:, :, D] / np.sum(
generative_model.
PdelDldelDr_given_D[:, :,
D]
)
self.given_D_CPdelDldelDr = np.array([
generative_model.PdelDldelDr_given_D[:, :, i].flatten().cumsum()
for i in range(generative_model.PdelDldelDr_given_D.shape[2])
])
self.C_Rvd = generative_model.Rvd.T.cumsum(axis=1)
self.C_Rdj = generative_model.Rdj.T.cumsum(axis=1)
if generative_model.first_nt_bias_insVD is None:
first_nt_bias_insVD = calc_steady_state_dist(generative_model.Rvd)
else:
first_nt_bias_insVD = generative_model.first_nt_bias_insVD
if generative_model.first_nt_bias_insDJ is None:
first_nt_bias_insDJ = calc_steady_state_dist(generative_model.Rdj)
else:
first_nt_bias_insDJ = generative_model.first_nt_bias_insDJ
self.C_first_nt_bias_insVD = first_nt_bias_insVD.cumsum()
self.C_first_nt_bias_insDJ = first_nt_bias_insDJ.cumsum()
self.num_J_genes = generative_model.PDJ.shape[1]
self.num_delDr_poss = generative_model.PdelDldelDr_given_D.shape[1]
self.cutV_genomic_CDR3_segs = genomic_data.cutV_genomic_CDR3_segs
self.cutD_genomic_CDR3_segs = genomic_data.cutD_genomic_CDR3_segs
self.cutJ_genomic_CDR3_segs = genomic_data.cutJ_genomic_CDR3_segs