omicverse.micro.MMvec

omicverse.micro.MMvec#

class omicverse.micro.MMvec(n_latent=3, lr=0.05, epochs=1000, val_frac=0.1, patience=100, l2=0.001, seed=0, device=None)[source]#

MMvec (Morton et al. 2019) in ~80 lines of PyTorch.

The objective is the exact expected multinomial log-likelihood

\[\ell \;=\; \sum_{i,j} W_{ij} \,\log \mathrm{softmax}(u_i \cdot V^\top + \beta)_j\]

where \(W_{ij} = \sum_s c_{s,i} \cdot m_{s,j} / M_s\) is the co-occurrence weight matrix (total microbe-i count × expected metabolite-j fraction over the cohort). For the tutorial-scale data we use the full softmax; the upstream mmvec package uses negative sampling to scale to thousands of features.

Parameters:

  • n_latent (int (default: 3)) – Embedding dimensionality K.

  • lr (float (default: 0.05)) – Adam learning rate.

  • epochs (int (default: 1000)) – Maximum training epochs.

  • val_frac (float (default: 0.1)) – Fraction of samples held out for the validation loss curve / early stopping. Set to 0 to skip validation.

  • patience (int (default: 100)) – Early-stopping patience on validation loss (epochs without improvement before training halts).

  • l2 (float (default: 0.001)) – Weight-decay on U / V / beta.

  • seed (int (default: 0)) – Torch RNG seed.

  • device (Optional[str] (default: None)) – 'cpu' / 'cuda' / None (auto-pick based on availability).

__init__(n_latent=3, lr=0.05, epochs=1000, val_frac=0.1, patience=100, l2=0.001, seed=0, device=None)[source]#
Parameters:

  • n_latent (int (default: 3))

  • lr (float (default: 0.05))

  • epochs (int (default: 1000))

  • val_frac (float (default: 0.1))

  • patience (int (default: 100))

  • l2 (float (default: 0.001))

  • seed (int (default: 0))

  • device (Optional[str] (default: None))

Methods

__init__([n_latent, lr, epochs, val_frac, ...])

conditional_probabilities()

Per-microbe P(metabolite | microbe) — softmax of U @ V.T + β.

cooccurrence()

Raw log-odds co-occurrence matrix U · Vᵀ (microbes × metabolites).

fit(adata_microbe, adata_metabolite[, verbose])

Fit MMvec on paired microbe / metabolite count tables and return self.

top_pairs([n])

Top-n (microbe, metabolite) pairs ranked by |log-odds|.

Attributes

metabolite_embeddings_

microbe_embeddings_

Contents