diff --git a/docker/openproblems-python-tf2.4/Dockerfile b/docker/openproblems-python-tf2.4/Dockerfile
new file mode 100644
index 0000000000..688aee580d
--- /dev/null
+++ b/docker/openproblems-python-tf2.4/Dockerfile
@@ -0,0 +1,15 @@
+FROM singlecellopenproblems/openproblems:latest
+
+ARG NB_USER="sagemaker-user"
+ARG NB_UID="1000"
+ARG NB_GID="100"
+
+USER root
+WORKDIR /
+
+# Install Python packages
+COPY ./docker/openproblems-python-tf2.4/requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+USER $NB_UID
+WORKDIR /home/$NB_USER
diff --git a/docker/openproblems-python-tf2.4/README.md b/docker/openproblems-python-tf2.4/README.md
new file mode 100644
index 0000000000..2704835345
--- /dev/null
+++ b/docker/openproblems-python-tf2.4/README.md
@@ -0,0 +1,14 @@
+# openproblems-python-tf2.4 Docker image
+
+Base image: singlecellopenproblems/openproblems
+
+OS: Debian Stretch
+
+Python: 3.8
+
+Python packages:
+
+
+* keras >=2.4,<2.6
+* tensorflow >=2.4,<2.5
+* dca
diff --git a/docker/openproblems-python-tf2.4/requirements.txt b/docker/openproblems-python-tf2.4/requirements.txt
new file mode 100644
index 0000000000..2948af862e
--- /dev/null
+++ b/docker/openproblems-python-tf2.4/requirements.txt
@@ -0,0 +1,4 @@
+dca==0.3.*
+keras>=2.4,<2.6  # pinned in dca
+pyyaml==5.4.1  # pinned in #431
+tensorflow==2.4.*  # pinned in dca
diff --git a/openproblems/tasks/denoising/README.md b/openproblems/tasks/denoising/README.md
index 41fffaf193..7584024ec5 100644
--- a/openproblems/tasks/denoising/README.md
+++ b/openproblems/tasks/denoising/README.md
@@ -8,10 +8,10 @@ A key challenge in evaluating denoising methods is the general lack of a ground
 
 # The metrics
 
-Metrics for data denoising aim to 
+Metrics for data denoising aim to assess denoising accuracy by comparing the denoised *training* set to the randomly sampled *test* set. 
 
-* **TODO**: TODO
-* **TODO**: TODO
+* **MSE**: The mean squared error between the denoised counts of the training dataset and the true counts of the test dataset after reweighting by the train/test ratio.
+* **Poisson**: The Poisson log likelihood of observing the true counts of the test dataset given the distribution given in the denoised dataset.
 
 ## API
 
diff --git a/openproblems/tasks/denoising/datasets/utils.py b/openproblems/tasks/denoising/datasets/utils.py
index b902fe42b7..ec31002e72 100644
--- a/openproblems/tasks/denoising/datasets/utils.py
+++ b/openproblems/tasks/denoising/datasets/utils.py
@@ -26,6 +26,12 @@ def split_data(
     X_train, X_test = molecular_cross_validation.util.split_molecules(
         X, 0.9, 0.0, random_state
     )
+    # remove zero entries
+    is_missing = X_train.sum(axis=0) == 0
+    X_train, X_test = X_train[:, ~is_missing], X_test[:, ~is_missing]
+
+    adata = adata[:, ~is_missing].copy()
     adata.obsm["train"] = scipy.sparse.csr_matrix(X_train).astype(float)
     adata.obsm["test"] = scipy.sparse.csr_matrix(X_test).astype(float)
+
     return adata
diff --git a/openproblems/tasks/denoising/methods/__init__.py b/openproblems/tasks/denoising/methods/__init__.py
index 1452b21cc2..a2596a2a70 100644
--- a/openproblems/tasks/denoising/methods/__init__.py
+++ b/openproblems/tasks/denoising/methods/__init__.py
@@ -1,4 +1,5 @@
 from .alra import alra
+from .dca import dca
 from .magic import magic
 from .magic import magic_approx
 from .no_denoising import no_denoising
diff --git a/openproblems/tasks/denoising/methods/dca.py b/openproblems/tasks/denoising/methods/dca.py
new file mode 100644
index 0000000000..eee6c78354
--- /dev/null
+++ b/openproblems/tasks/denoising/methods/dca.py
@@ -0,0 +1,35 @@
+from ....tools.decorators import method
+from ....tools.utils import check_version
+
+import scanpy as sc
+
+
+def _dca(adata, test=False, epochs=None):
+    if test:
+        epochs = epochs or 30
+    else:  # pragma: nocover
+        epochs = epochs or 300
+    from dca.api import dca
+
+    # make adata object with train counts
+    adata_train = sc.AnnData(adata.obsm["train"])
+    # run DCA
+    dca(adata_train, epochs=epochs)
+
+    # set denoised to Xmat
+    adata.obsm["denoised"] = adata_train.X
+    # check version of dca
+    adata.uns["method_code_version"] = check_version("dca")
+    return adata
+
+
+@method(
+    method_name="DCA",
+    paper_name="Single-cell RNA-seq denoising using a deep count autoencoder",
+    paper_url="https://www.nature.com/articles/s41467-018-07931-2",
+    paper_year=2019,
+    code_url="https://github.com/theislab/dca",
+    image="openproblems-python-tf2.4",
+)
+def dca(adata, test=False, epochs=None):
+    return _dca(adata, test=test, epochs=epochs)