Debarcoding¶

Runtime: about 20s.

In [1]:

import scyan

import scyan

Global seed set to 0

Model initialization and fitting¶

In [2]:

adata, table = scyan.data.load("debarcoding")

model = scyan.Scyan(adata, table)

adata, table = scyan.data.load("debarcoding") model = scyan.Scyan(adata, table)

[INFO] (scyan.model) Initialized Scyan model with N=100000 cells, P=22 populations and M=6 markers.
   ├── No covariate provided
   ├── No continuum-marker provided
   └── Batch correction mode: False

The table contains the barcodes of each patient (or sample):

In [5]:

table.head()

table.head()

Out[5]:

	BC1	BC2	BC3	BC4	BC5	BC6
Sample ID
Pos	1	1	1	1	1	1
A1	1	1	1	-1	-1	-1
A2	1	1	-1	1	-1	-1
A3	1	1	-1	-1	1	-1
A4	1	1	-1	-1	-1	1

In [3]:

model.fit()

model.fit()

INFO:scyan.model:Training scyan with the following hyperparameters:
"batch_key":       None
"batch_size":      16384
"hidden_size":     16
"lr":              0.001
"max_samples":     200000
"modulo_temp":     2
"n_hidden_layers": 7
"n_layers":        7
"prior_std":       0.25
"temperature":     0.5

GPU available: True (mps), used: False
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
/Users/quentinblampey/Library/Caches/pypoetry/virtualenvs/scyan-5lsXrWE1-py3.9/lib/python3.9/site-packages/pytorch_lightning/trainer/setup.py:200: UserWarning: MPS available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='mps', devices=1)`.
  rank_zero_warn(

  | Name   | Type        | Params
---------------------------------------
0 | module | ScyanModule | 29.7 K
---------------------------------------
29.7 K    Trainable params
0         Non-trainable params
29.7 K    Total params
0.119     Total estimated model params size (MB)
/Users/quentinblampey/Library/Caches/pypoetry/virtualenvs/scyan-5lsXrWE1-py3.9/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py:1595: PossibleUserWarning: The number of training batches (6) is smaller than the logging interval Trainer(log_every_n_steps=10). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.
  rank_zero_warn(

Training: 0it [00:00, ?it/s]

INFO:scyan.model:Successfully ended traning.

Out[3]:

Scyan model with N=100000 cells, P=22 populations and M=6 markers.
   ├── No covariate provided.
   └── Batch correction mode: False

In [4]:

model.predict(log_prob_th=-10);  # Predictions are saved in adata.obs.scyan_pop by default

model.predict(log_prob_th=-10); # Predictions are saved in adata.obs.scyan_pop by default

DataLoader:   0%|          | 0/7 [00:00<?, ?it/s]

Results visualization and interpretability¶

In [5]:

scyan.plot.umap(adata, color="scyan_pop")

scyan.plot.umap(adata, color="scyan_pop")

In [6]:

scyan.plot.scatter(adata, ["A1", "A2", "B2", "C4", "D5"], markers=["BC1", "BC2", "BC3", "BC4"])

scyan.plot.scatter(adata, ["A1", "A2", "B2", "C4", "D5"], markers=["BC1", "BC2", "BC3", "BC4"])

The latent space corresponds to the barcodes!

In [7]:

scyan.plot.pops_expressions(model)

scyan.plot.pops_expressions(model)

DataLoader:   0%|          | 0/5 [00:00<?, ?it/s]

One sample latent expression is its barcode.

In [8]:

scyan.plot.pop_expressions(model, "D4")

scyan.plot.pop_expressions(model, "D4")

DataLoader:   0%|          | 0/1 [00:00<?, ?it/s]