LDVAE#

LDVAE 1 (Linearly-decoded Variational Auto-encoder, also called Linear scVI; Python class LinearSCVI) is a flavor of scVI with a linear decoder.

The advantages of LDVAE are:

  • Can be used to interpret latent dimensions with factor loading matrix.

  • Scalable to very large datasets (>1 million cells).

The limitations of LDVAE include:

  • Less capacity than scVI, which uses a neural network decoder.

  • Less capable of integrating data with complex batch effects.

Tutorials:

  • /tutorials/notebooks/linear_decoder

Contrasting with scVI#

Here we discuss the differences between LDVAE and scVI.

  • In LDVAE, \(f_w(z_n, s_n)\) is a linear function, and thus can be represented by a matrix \(W\) of dimensions \(G\) (genes) by \((d + k)\) (latent space dim plus covariate categories).

  • This matrix \(W\) can be accessed using get_loadings()

  • LDVAE does not offer transfer learning capabilities currently.

1

Valentine Svensson, Adam Gayoso, Nir Yosef, Lior Pachter (2020), Interpretable factor models of single-cell RNA-seq via variational autoencoders, Bioinformatics.