However, it is still not clear what are minimum ingredients of a deep functional map pipeline and whether such ingredients unify or generalize all recent work on deep functional maps. We show empirically the minimum components for obtaining state-of-the-art results with different loss functions, supervised aswell asunsupervised.

Figure S2 shows the inferred bipartite network with the top 200 interactions by BayReL. BayReLinfacthasalsoidentified corresponding drugs that have been proposed to target this gene. Themostsignificantlyenriched GO terms include the MHC class II receptor activity (p-value = 0.00099), chemokine activity (p-value = 0.00175), MHC class II protein complex (p-value = 0.00273), chemokine receptor binding (p-value=0.00404),andregulation Immune escape ofrelapsed amlcells after allogeneic transplantation.

Studying molecular interactions across different data types helps reveal signal transduction mechanisms across different classes of molecules.

Random walks are key components of many machine learning algorithms with applications in computing graph partitioning [ST04, GKL+21], spectral embeddings [CPS15, CKK+18], or network inference [HMMT18], as well as learning image segmentation [MS00], ranking nodes in a graph [AC07] and many other applications.

Each step replaces the two mostsimilar clusters by its union.