Aware: Graph orientation mismatch causing ValueError in GNNEmbedding

[ramosv]

A couple of users mentioned that they kept getting a ValueError: Number of omics features must match adjacency matrix size. when using the GNNEmbedding class.

Our graph generation functions are orientation agnostic, but GNNEmbedding expects a feature–feature adjacency.

Quick fix (until v1.1.1), transpose your matrix before calling a graph generation:

from bioneuralnet.utils.graphs import gen_similarity_graph

X = your_omics_dataset_as_dataframe.T
A = gen_similarity_graph(X, k=15, metric='cosine', mutual=True)

Plan for v1.1.1

• Expect v1.1.1 release by end of day this Wednesday (08/04/2025)
• Standardize user expectation: rows = samples, cols = features.
• Default all graph generation function to feature–feature (add an axis parameter to let users decide).
• Clearer error messages + docs update.
• Update all tests and examples regarding graph generation.