liveBook · Manning

Multi-hot tensor

As explained in Build AI Drug Discovery Pipelines

A multi-hot vector is a binary vector in which multiple positions can simultaneously indicate active categories or attributes.

A multi-hot vector represents an atom with several simultaneously active features rather than one exclusive category. In molecular graphs, it combines categorical encodings with an aromaticity indicator for graph-neural-network inputs. section 11.2.2

The atomic representation encodes:

  • Element identity across a 44-category vocabulary, including an Unknown category. section 11.2.2
  • Heavy-atom degree across 11 categories. section 11.2.2
  • Attached hydrogen count across 11 categories. section 11.2.2
  • Implicit valence across 11 categories, plus a binary aromaticity flag. section 11.2.2

A 78-dimensional atom feature

The atom_features function concatenates the element, degree, hydrogen-count, and implicit-valence encodings with atom.GetIsAromatic(), returning a NumPy float32 array.

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines the feature layout and implements its 78-dimensional construction, without detailing model training or optimization.
Examples
Uses element symbols, heavy-atom degree, hydrogen count, implicit valence, and aromaticity as simultaneous atom features.
What you can do after
Build it from scratch with the supplied Python function for converting atom properties into a NumPy feature vector.
How it connects
Connects one-hot encoding, molecular graphs, graph neural networks, and drug-target interactions.

Also explained in

Compare how 2 books explain multi-hot tensor →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.