def process_mousebrain_multiinds_data(data_dir, result_dir, sample_ind, pred_ind,
region="FC", sample=False, sample_seed=0, celltype_gran=0):
'''Process mousebrain data from multiple individuals to another
@sample_ind: downsample to a certain individual number
@pred_ind: the predicted individual
'''
## individuals for brain regions
FC_inds = ["P60FCAldh1l1Rep1", "P60FCCx3cr1Rep1", "P60FCRep1", "P60FCRep2",
"P60FCRep3", "P60FCRep4", "P60FCRep6"]
HC_inds = ["P60HippoRep1", "P60HippoRep2", "P60HippoRep3",
"P60HippoRep4", "P60HippoRep5", "P60HippoRep6"]
if region == "FC":
exclude_list = FC_inds
elif region == "HC":
exclude_list = HC_inds
exclude_list.remove(pred_ind)
exclude_inds = '_'.join(exclude_list)
train_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region,
ind=exclude_inds, celltype_gran=celltype_gran)
## sample to number of cells when compared to the sample_id
if sample:
avg_number = train_adata.shape[0] // len(exclude_list)
print("=== Downsample to number:", avg_number)
random.seed(sample_seed)
sampled_cells = random.sample(list(train_adata.obs_names), k=avg_number)
train_adata = train_adata[sampled_cells]
test_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region,
ind=pred_ind, celltype_gran=celltype_gran)
return train_adata, test_adata
def process_mousebrain_regions(data_dir, result_dir, region1, region2,
celltype_gran=0):
''' Use one region to predict another
can allow region_individual input
@celltype_gran: with sub-cell types and check ARI
'''
## get region information and individuals
regions1 = region1.split('_')
regions2 = region2.split('_')
region1 = regions1[0]
if len(regions1) > 1:
regions1_inds = '_'.join(regions1[1:])
else:
regions1_inds = None
region2 = regions2[0]
if len(regions2) > 1:
regions2_inds = '_'.join(regions2[1:])
else:
regions2_inds = None
## get train/test anndata
train_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region1,
ind=regions1_inds, celltype_gran=celltype_gran, curate=True)
test_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region2,
ind=regions2_inds, celltype_gran=celltype_gran, curate=True)
return train_adata, test_adata
def process_mousecortex_protocols_mouseFC_data(data_dir, result_dir,
input1, input2, curate=False):
'''process both mousecortex protocols and mouse FC data to predict
Note here we can only do major cell types because Cortex data does not
provide sub-cell types
@input1/input2: MouseProtocol+exp+protocols_inds, mouseFC_inds
@curate: whether to curate the datasets
'''
## load train and test adata
input1_list = input1.split('_')
input1_dataset = input1_list[0]
if len(input1_list) > 1:
input1_inds = '_'.join(input1_list[1:])
else:
input1_inds = None
input2_list = input2.split('_')
input2_dataset = input2_list[0]
if len(input2_list) > 1:
input2_inds = '_'.join(input2_list[1:])
else:
input2_inds = None
## extract train and test adata accordingly
if "mouseFC" in input1_dataset:
train_adata = load_mousebrain_data.load_brain_adata(data_dir, region="FC",
ind=input1_inds, celltype_gran=0, curate=True)
elif "MouseProtocol" in input1_dataset:
info = input1_dataset.split('+')
exp = info[1]
protocol = info[2]
if exp == "Both":
train_adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dir,
exp=None, protocol=protocol, curate=True)
else:
train_adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dir,
exp=exp, protocol=protocol, curate=True)
if "mouseFC" in input2_dataset:
test_adata = load_mousebrain_data.load_brain_adata(data_dir, region="FC",
ind=input2_inds, celltype_gran=0, curate=True)
elif "MouseProtocol" in input2_dataset:
info = input2_dataset.split('+')
exp = info[1]
protocol = info[2]
if exp == "Both":
test_adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dir,
exp=None, protocol=protocol, curate=True)
else:
test_adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dir,
exp=exp, protocol=protocol, curate=True)
return train_adata, test_adata
def process_mousebrain_data(data_dir, result_dir, ind1, ind2, region="FC",
celltype_gran=0, curate=False):
'''Process Frontal Cortex from one ind to another
@ind: P60FCAldh1l1Rep1/P60FCCx3cr1Rep1/P60FCRep1/P60FCRep2/P60FCRep3/P60FCRep4/P60FCRep6
@celltype_gran: granularity of cell types. 0: major cell types, 1: sub-celltypes
@purify: whether to purify the training dataset
@purify_method: purify using distance or SVM
@curate: whether to curate mouse brain data by combining different neurons/interneurons
'''
train_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region,
ind=ind1, celltype_gran=celltype_gran, curate=curate)
test_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region,
ind=ind2, celltype_gran=celltype_gran, curate=curate)
return train_adata, test_adata
def process_mousebrain_crossdataset_inds_data(data_dir, result_dir, ref_inds, tgt_inds='mFC_P60FCCx3cr1Rep1',
celltype_gran=0, curate=True):
'''Predict mouse FC target individual using pFC individuals and allenbrain
@ ref_inds: pFC_inds+allen_inds
@ target_ind: FC_P60FCCx3cr1Rep1
'''
adata_list = []
dataset_inds_list = ref_inds.split('+')
for dataset_ind in dataset_inds_list:
if 'mFC' in dataset_ind:
FC_inds = dataset_ind.replace('mFC_', '')
FC_adata = load_mousebrain_data.load_brain_adata(data_dir, region="FC",
ind=FC_inds, curate=curate)
adata_list.append(FC_adata)
if 'pFC' in dataset_ind:
pFC_inds = dataset_ind.replace('pFC_', '')
pFC_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage='Adult', ind=pFC_inds, curate=curate)
adata_list.append(pFC_adata)
if 'allen' in dataset_ind:
allen_inds = dataset_ind.replace('allen_', '')
allen_adata = load_mouseAllen_data.load_mouseAllen_10X_data(data_dir, ind=allen_inds, curate=curate)
adata_list.append(allen_adata)
train_adata = anndata.AnnData.concatenate(*adata_list, join='inner')
del adata_list ## release memory
## same as reference, for simplification, I just copied and pasted here
adata_list = []
dataset_inds_list = tgt_inds.split('+')
for dataset_ind in dataset_inds_list:
if 'mFC' in dataset_ind:
FC_inds = dataset_ind.replace('mFC_', '')
FC_adata = load_mousebrain_data.load_brain_adata(data_dir, region="FC",
ind=FC_inds, curate=curate)
adata_list.append(FC_adata)
if 'pFC' in dataset_ind:
pFC_inds = dataset_ind.replace('pFC_', '')
pFC_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage='Adult', ind=pFC_inds, curate=curate)
adata_list.append(pFC_adata)
if 'allen' in dataset_ind:
allen_inds = dataset_ind.replace('allen_', '')
allen_adata = load_mouseAllen_data.load_mouseAllen_10X_data(data_dir, ind=allen_inds, curate=curate)
adata_list.append(allen_adata)
test_adata = anndata.AnnData.concatenate(*adata_list, join='inner')
del adata_list ## release memory
return train_adata, test_adata
def process_FCdatasets_mouseFC_data(data_dir, result_dir, datasets_ref, mousebrain_tgt,
celltype_gran=0, curate=False):
'''Combine mouse brain protocol + pFC datasets together to predict the
mouse FC target individual
@datasets_ref: MouseProtocol+Both+DroNc-seq_pFC+Adult+Saline
@mousebrain_tgt: mousebrain dataset with certain targets
'''
datasets_list = datasets_ref.split('_')
adata_list = []
for dataset in datasets_list:
infos = dataset.split('+')
dataset_name = infos[0]
if dataset_name == "MouseProtocol":
exp = infos[1]
protocol = infos[2]
if exp == "Both":
adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dir,
exp=None, protocol=protocol, curate=curate)
else:
adata = load_mouseprotocol_data.load_mouseprotocol_adata(data_dur,
exp=exp, protocol=protocol, curate=curate)
elif dataset_name == "pFC":
devstage = infos[1]
treatment = infos[2]
adata = load_mouseFC_data.load_FC_adata(data_dir, devstage=devstage,
treatment=treatment, celltype_gran=celltype_gran, curate=curate)
adata_list.append(adata)
## find common genes and concatenate adata
common_columns = ['barcode', 'nGene', 'nUMI', 'percent.mito', 'cell.type']
for adata in adata_list:
adata_obs = adata.obs[common_columns]
adata.obs = adata_obs
train_adata = anndata.AnnData.concatenate(*adata_list, join='inner')
del adata_list ## release space
## get target individuals list
tgt_list = mousebrain_tgt.split('_')
region = tgt_list[0]
if len(tgt_list) > 1:
tgt_inds = '_'.join(tgt_list[1:])
else:
tgt_inds = None
test_adata = load_mousebrain_data.load_brain_adata(data_dir, region=region,
ind=tgt_inds, celltype_gran=celltype_gran, curate=curate)
return train_adata, test_adata
def process_allenbrain_cross_data(data_dir, result_dir, dataset1, dataset2="Allen",
sample_seed=0, celltype_gran=0, curate=True):
'''Predict allenbrain FACS-sorted SSv4 data 20% test cells
@ dataset1: individual FC_P60FCRep1 or pFC_PFCSample1
@ dataset2: Allen
@ curate: whether to curate mouse brain cell types
'''
## 20% test cells from allen brain dataset
adata = load_mouseAllen_data.load_mouseAllen_SS_data(data_dir, curate=curate)
random.seed(sample_seed)
train_cells = random.sample(adata.obs_names.tolist(), round(0.8*adata.shape[0]))
test_cells = list(set(adata.obs_names) - set(train_cells))
test_adata = adata[test_cells]
## construct training dataset
if "FC" == dataset1:
train_adata = load_mousebrain_data.load_brain_adata(data_dir, region="FC",
ind="P60FCRep1", celltype_gran=celltype_gran, curate=curate)
if "pFC" == dataset1:
train_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage="Adult",
ind="PFCSample1", curate=curate)
return train_adata, test_adata
def process_mousebrain_FC_datasets(data_dir, result_dir, dataset1_input, dataset2_input,
sample=False, sample_seed=0, celltype_gran=0):
'''Use FC from one dataset to predict another
@dataset1_input: "FC_inds", if sample == True,
the inds indicate the number of cells to downsample
@dataset2_input: "Adult_inds"
the above two parameters can be interchangable
'''
dataset1_split = dataset1_input.split('_')
dataset2_split = dataset2_input.split('_')
## identify where the dataset points to
regions = ["FC", "HC", "CB", "STR", "SN", "PC"]
stages = ["Adult", "P21"]
dataset1_input = dataset1_split[0]
if len(dataset1_split) > 1:
dataset1_inds = dataset1_split[1:]
dataset1_inds = '_'.join(dataset1_inds)
else:
dataset1_inds = None
dataset2_input = dataset2_split[0]
if len(dataset2_split) > 1:
dataset2_inds = dataset2_split[1:]
dataset2_inds = '_'.join(dataset2_inds)
else:
dataset2_inds = None
if dataset1_input in regions:
train_adata = load_mousebrain_data.load_brain_adata(data_dir,
region=dataset1_input, ind=dataset1_inds, curate=True)
else:
info = dataset1_input.split('-')
if info[0] in stages:
if len(info) > 1:
train_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage=info[0],
treatment=info[1], curate=True)
else:
train_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage=info[0],
ind=dataset1_inds, curate=True)
if dataset2_input in regions:
test_adata = load_mousebrain_data.load_brain_adata(data_dir,
region=dataset2_input, ind=dataset2_inds, curate=True)
else:
info = dataset2_input.split('-')
if info[0] in stages:
if len(info) > 1:
test_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage=info[0],
treatment=info[1], curate=True)
else:
test_adata = load_mouseFC_data.load_FC_adata(data_dir, devstage=info[0],
ind=dataset2_inds, curate=True)
if sample:
## sample number to the given individual cell numbers
avg_number = train_adata.shape[0]
if dataset1_input == "FC":
train_adata = load_mousebrain_data.load_brain_adata(data_dir,
region=dataset1_input, ind=None, curate=True)
if dataset1_input == "Adult":
train_adata = load_mouseFC_data.load_FC_adata(data_dir, dataset1_input,
ind=None, curate=True)
random.seed(sample_seed)
sampled_cells = random.sample(list(train_adata.obs_names), k=avg_number)
train_adata = train_adata[sampled_cells]
return train_adata, test_adata