size factors in guide

docs/user_guide/models/multivi.rst

@@ -96,7 +96,7 @@ The latent variables, along with their description are summarized in the followi
      - Low-dimensional representation capturing the state of a cell.
      - N/A
    * - :math:`\rho_n \in \Delta^{G-1}`
-     - Denoised/normalized gene expression.
+     - Denoised/normalized gene expression. This is a vector that sums to 1 within a cell, unless `size_factor_key is not None` in :class:`~scvi.model.MULTVI.setup_anndata`, in which case this is only forced to be non-negative via softplus.
      - ``px_scale``
    * - :math:`\ell_n \in (0, \infty)`
      - Library size for RNA.
@@ -108,7 +108,7 @@ The latent variables, along with their description are summarized in the followi
      - Accessibility probability estimate
      - N/A
    * - :math:`\ell_n \in \left[0,1\right]`
-     - Cell-wise scaling factor
+     - Cell-wise scaling factor. Learned, but can be set manually with `size_factor_key` in :class:`~scvi.model.MULTIVI.setup_anndata`.
      - ``d``
    * - :math:`r_j \in \left[0,1\right]`
      - Region-wise scaling factor

docs/user_guide/models/scanvi.rst

@@ -65,7 +65,7 @@ cellular state as a continuous, low-dimensional random variable, and has the sam
 However, the prior for this variable takes into account the partial cell-type information to better structure the latent space.
 
 The rest of the model closely follows scVI. In particular, it represents the library size as a random variable,
-and gene expression likelihoods as negative binomial distributions parameterized by functions of :math:`z_n, l_n`,
+and gene expression likelihoods as negative binomial distributions parameterized by functions of :math:`z_n, \ell_n`,
 condition to the batch assignments :math:`s_n`.
 
 .. figure:: figures/scanvi_pgm.png
@@ -105,10 +105,10 @@ scANVI assumes the following factorization for the inference model
    :nowrap:
 
    \begin{align}
-      q_\eta(z_n, l_n, u_n, c_n \mid x_n)
+      q_\eta(z_n, \ell_n, u_n, c_n \mid x_n)
       =
       q_\eta(z_n \mid x_n)
-      q_\eta(l_n \mid x_n)
+      q_\eta(\ell_n \mid x_n)
       q_\eta(c_n \mid z_n)
       q_\eta(u_n \mid c_n, z_n)
    \end{align}

docs/user_guide/models/scvi.rst

@@ -90,10 +90,10 @@ The latent variables, along with their description are summarized in the followi
      - Low-dimensional representation capturing the state of a cell.
      - N/A
    * - :math:`\rho_n \in \Delta^{G-1}`
-     - Denoised/normalized gene expression.
+     - Denoised/normalized gene expression. This is a vector that sums to 1 within a cell, unless `size_factor_key is not None` in :class:`~scvi.model.SCVI.setup_anndata`, in which case this is only forced to be non-negative via softplus.
      - ``px_scale``
    * - :math:`\ell_n \in (0, \infty)`
-     - Library size for RNA. Here it is modeled as a latent variable, but the recent default for scVI is to treat library size as observed, equal to the total RNA UMI count of a cell. This can be controlled by passing ``use_observed_lib_size=False`` to :class:`~scvi.model.SCVI`.
+     - Library size for RNA. Here it is modeled as a latent variable, but the recent default for scVI is to treat library size as observed, equal to the total RNA UMI count of a cell. This can be controlled by passing ``use_observed_lib_size=False`` to :class:`~scvi.model.SCVI`. The library size can also be set manually using `size_factor_key` in :class:`~scvi.model.SCVI.setup_anndata`.
      - N/A
    * - :math:`\theta_g \in (0, \infty)`
      - Inverse dispersion for negative binomial. This can be set to be gene/batch specific for example (and would thus be :math:`\theta_{kg}`), by passing ``dispersion="gene-batch"`` during model intialization. Note that ``px_r`` also refers to the underlying real-valued torch parameter that is then exponentiated on every forward pass of the model.

docs/user_guide/models/totalvi.rst

@@ -115,7 +115,7 @@ The latent variables, along with their description are summarized in the followi
      - Low-dimensional representation capturing joint state of a cell
      - N/A
    * - :math:`\rho_n \in \Delta^{G-1}`
-     - Denoised/normalized gene expression,
+     - Denoised/normalized gene expression. This is a vector that sums to 1 within a cell, unless `size_factor_key is not None` in :class:`~scvi.model.TOTALVI.setup_anndata`, in which case this is only forced to be non-negative via softplus.
      - ``px_["scale"]``
    * - :math:`\alpha_n \in [1, \infty)^T`
      - Foreground scaling factor for proteins, identifies the mixture distribution (see below)
@@ -124,7 +124,7 @@ The latent variables, along with their description are summarized in the followi
      - Probability of background for each protein
      - ``py_["mixing"]`` (logits scale).
    * - :math:`l_n \in (0, \infty)`
-     - Library size for RNA. Here it is modeled as a latent variable, but the recent default for totalVI is to treat library size as observed, equal to the total RNA UMI count of a cell. This can be controlled by passing ``use_observed_lib_size=False`` to :class:`~scvi.model.TOTALVI`.
+     - Library size for RNA. Here it is modeled as a latent variable, but the recent default for totalVI is to treat library size as observed, equal to the total RNA UMI count of a cell. This can be controlled by passing ``use_observed_lib_size=False`` to :class:`~scvi.model.TOTALVI`. The library size can also be set manually using `size_factor_key` in :class:`~scvi.model.TOTALVI.setup_anndata`.
      - N/A
    * - :math:`\beta_{nt} \in (0, \infty)`
      - Protein background intensity. Used twice to identify the protein mixture model.