Add notebook on predicting statistics

deepcpg/models/utils.py

@@ -573,7 +573,7 @@ def __call__(self, data_files, class_weights=None, *args, **kwargs):
                     outputs[name] = data_raw['outputs/%s' % name]
                     cweights = class_weights[name] if class_weights else None
                     weights[name] = get_sample_weights(outputs[name], cweights)
-                    if name == 'stats/cat_var':
+                    if name.endswith('cat_var'):
                         output = outputs[name]
                         outputs[name] = to_categorical(output, 3)
                         outputs[name][output == dat.CPG_NAN] = 0

docs/source/data.rst

@@ -70,7 +70,7 @@ For debugging, testing, or reducing compute costs, ``--chromos`` can be used the
 Predicting statistics
 ---------------------
 
-For predicting statistics across methylation profiles, ``--stats`` and ``--win_stats`` can be used. These arguments specify a list of statistics that are computed across profiles for either a single CpG site or in a window of size ``--win_stats_wlen`` that is centered on a target CpG site. Following statistics are supported:
+For predicting statistics across methylation profiles, ``--cpg_stats`` and ``--win_stats`` can be used. These arguments specify a list of statistics that are computed across profiles for either a single CpG site or in windows of length ``--win_stats_wlen`` that are centered on a CpG site. Following statistics are supported:
 
 * ``mean``: the mean methylation rate.
 * ``mode``: the mode of methylation rates.
@@ -81,7 +81,7 @@ For predicting statistics across methylation profiles, ``--stats`` and ``--win_s
 * ``diff``: if a CpG site is differentially methylated, i.e. methylated in one profile but zero in others.
 * ``cov``: the CpG coverage, i.e. the number of profiles for which the methylation state of the target CpG site is observed.
 
-Statistics are only computed or CpG sites that are covered by at least ``--stats_cov`` (default 1) cells. Increasing ``--stats_cov`` will lead to more robust estimates.
+Per-CpG statistics specified by ``--cpg_stats`` are computed only for CpG sites that are covered by at least ``--cpg_stats_cov`` (default 3) cells. Increasing ``--cpg_stats_cov`` will lead to more robust estimates.
 
 
 Common issues

examples/README.md

@@ -15,6 +15,7 @@ bash setup.sh
 * [Fine-tuning](./notebooks/fine_tune/index.ipynb): Fine-tuning a pre-trained model to speed-up training.
 * [Motif analysis](./notebooks/motifs/index.ipynb): Visualizing and analyzing learned motifs.
 * [Mutations effects](./notebooks/snp/index.ipynb): Computing and visualizing mutations effects.
+* [Predicting statistics](./notebooks/stats/index.ipynb): Predicting statistics such as cell-to-cell variance.
 
 ## Shell scripts
 `./scripts` contains shell scrips with recommended default parameters. They may help you to easily build a DeepCpG pipeline for creating data, training models, and evaluating models. Set `test_mode` variable in scripts to `1` for testing, and `0` otherwise.