ma-compbio / fast-higashi Goto Github PK

Hi Ruochi,

Thank you for your excellent work in both Higashi and fast-Higashi. I'm currently working with the 4 datasets mentioned in your paper (Lee, Liu, Tan, and Ramani). I noticed that you have created the "label_info.pickle" file, which includes cell subtype information, batch ID, and other annotations for each cell.

However, I couldn't find these annotations in the original GEO accession files. For example, in the Lee et al. dataset, there are cell subtypes like 'L2/3', 'ODC', 'Sst', 'L5', 'Astro', etc. listed in the manuscript, but I can't locate the raw cell subtype information in the raw contacts.txt file or the indexed.tsv file (under each cell sample ID).

Maybe I'm missing something. Your advise will be so helpful. Thank you!

Hi, Ruochi
There is no 'setup.py' provided actually. Thanks.

Hi, I'm interested in pulling out the chrom-specific meta-interactions (as in Figure 3B). Is this derived from the matrices in the B_list of the final model? I looked at these for my dataset but they don't look like the example ones, so I'm wondering if I'm missing something. Thanks!

I'm trying to run tutorial with Fast-Higashi on Lee et al. dataset (sn-m3c-seq on PFC).
I downloaded all of file in goole drive (https://drive.google.com/drive/folders/1SuzqQ_9dliAmTb-fGprFnN3aZrfWS-Fg?usp=sharing)

The code that I used is below

from fasthigashi.FastHigashi_Wrapper import *
config = "/work/magroup/ruochiz/fast_higashi_git/config_pfc.JSON"
model = FastHigashi(config_path=config,
	             path2input_cache="/work/magroup/ruochiz/fast_higashi_git/pfc_500k",
	             path2result_dir="/work/magroup/ruochiz/fast_higashi_git/pfc_500k",
	             off_diag=100,
	             filter=False,
	             do_conv=False,
	             do_rwr=False,
	             do_col=False,
	             no_col=False)
model.prep_dataset(batch_norm=True)

And the error was occurred

total number of cells that pass qc check 4145 bad 93 total: 4238
cache file = /das2/younso/Hic/schic/F_higashi/pfc_500k/cache_intra_500000_offdiag_100_.pkl
loading cached input from /das2/younso/Hic/schic/F_higashi/pfc_500k/cache_intra_500000_offdiag_100_.pkl
---------------------------------------------------------------------------
EOFError                                  Traceback (most recent call last)
Cell In[4], line 1
----> 1 model.prep_dataset(batch_norm=True)

File ~/bin/Fast-Higashi/fasthigashi/FastHigashi_Wrapper.py:484, in FastHigashi.prep_dataset(self, meta_only, batch_norm)
    481         cache_extra = ""
    482 path2input_cache_intra = os.path.join(self.path2input_cache, 'cache_intra_%d_offdiag_%d_%s.pkl' % (
    483         res, self.off_diag, cache_extra))
--> 484 all_matrix += self.preprocess_contact_map(
    485         self.config, reorder=reorder, path2input_cache=path2input_cache_intra,
    486         batch_norm=batch_norm,
    487         is_sym=True,
    488         off_diag=self.off_diag,
    489         fac_size=1,
    490         merge_fac_row=int(res / self.config['resolution']), merge_fac_col=int(res / self.config['resolution']),
    491         filename_pattern='%s_sparse_adj.npy',
    492         force_shift=False,
    493 )
    495 size_list = [m.shape[0] for m in all_matrix]
    496 num_cell = all_matrix[-1].shape[-1]

File ~/bin/Fast-Higashi/fasthigashi/FastHigashi_Wrapper.py:377, in FastHigashi.preprocess_contact_map(self, config, reorder, path2input_cache, batch_norm, key_fn, **kwargs)
    375 with open(path2input_cache, 'rb') as f:
    376         for chrom in self.chrom_list:
--> 377                 all_matrix.append(pickle.load(f))
    378 sys.stdout.flush()
    379 return all_matrix

EOFError: Ran out of input

In the tutorial description, below sentence is located
"To run Fast-Higashi for a new dataset, please prepare the same input files for the Higashi software. Use the Higashi software higashi.process_data() to transform contact pair files to sparse contact maps."

Is this mean that I should use Higashi at first?
But, then, I can't use FastHigashi(config_path=config, ... function because it will be substituted Higashi(config).

Why this error is occurred?

Hello, I was wondering if Fast-Higashi can be used with 10kb scHi-C data for imputation?

Hi Zhang，
Nice work!

I'm having a problem with out-of-order images. But my filelist is in the same order as the cell types in the label_info.pickle file. I would like to ask you why this situation is caused?

Could you please help me with it? Thank you so much!

Hello! Thank you for sharing your code with us and the very detailed Wiki to walk us through the setup process.

I wanted to ask, how do I acquire the imputed matrices from the Fast-Higashi? I know you run a partial_rwr algorithm to generate imputed scHi-C matrices. Is this function adaptable to any tensor of scHi-C matrices? Does the input "x" to this function require the tensor to be arranged in a certain way, for example the scHi-C matrices of the same cell type are closer to each other in the tensor (think of axis 0)? Or can I just borrow this function and run it on my own scHi-C data tensors without any order?

Thanks again!