asistradition
commented
1 year ago The amusr regression method is explicitly a multi-task method (it doesn't work without breaking your training data into tasks). inferelator_workflow(regression='amusr', workflow='tfa') won't work and needs to be inferelator_workflow(regression='amusr', workflow='multitask'), with tasks defined.
Alternatively you can try using inferelator_workflow(regression='stars', workflow='tfa') or inferelator_workflow(regression='bbsr', workflow='tfa') with your existing workflow.
I will fix this to provide a useful error message (instead of this cryptic message).
I am trying to run Inferelator on a h5ad file, but I get the following error when trying to run the cv wrap function. The h5ad file originates from a loom file that I made by subsetting a Seurat R object on an expression threshold 0.5<. I am using a TF name list with 3 TFs. The problem is in getting the betas usig a list comprehension. Paths modified for clarity:
This is the log of stdout:
2 Inferelator parallelized using engine joblib
1 Loading AnnData file FILENAME.h5ad (Layer: X)
1 Data loaded: InferelatorData [float32 (427, 28570), Metadata (427, 22)] Memory: 12.20 MB
2 Loading TF feature names from file 3_tf.tsv
2 Loading prior data from file updated_matrix.tsv
2 Loading gold_standard data from file updated_matrix.tsv
1 Setting crossvalidation param random_seed to 42
2 Writing results to /path/to/amu_sr/random_seed_42
1 Gold standard (62611, 1021) split on axis 0. Prior knowledge network (50088, 1021) [16254778 edges] used for activity and gold standard network (12523, 1021) [4041369 edges] used for testing.
2 Filtered prior to 3 TFs from the TF name list
2 Trimming Expression matrix (427, 28570) to 18112 columns
1 Skipping metadata because parsing requires missing columns: Index(['del.t', 'isTs', 'prevCol'], dtype='object')
2 Constructed design (427, 18112) and response (427, 18112) matrices
2 Rebuilt design matrix (427, 3) with TF activity
This is the traceback:
Packages in my conda env: