. Author manuscript; available in PMC: 2018 Nov 30.

Published in final edited form as: Cell. 2017 Nov 16;171(6):1424–1436.e18. doi: 10.1016/j.cell.2017.10.023

Algorithm.

Sparse Module Activity Factorization.

1. SMAF(X, d, λ, k)

2. Initialize U ∈ ℝ^{g × d} and W ∈ ℝ^{d × n} randomly.

3. For 10 iterations:

a. Update the module dictionary as U = Lasso Nonnegative (X, W,λ).

b. Normalize each module so that ||u_i||₂=1.

c. Update the activity levels as W = OMP (X, U, k).

4. Return U, W.