To fit the LMM, we used a python-based implementation, which follows the derivation of fastLMM 2. The standard LM is implemented in the same framework, setting Σ = I.