liveBook · Manning

Structure-property relationship

As explained in Build AI Drug Discovery Pipelines

A structure-property relationship describes how molecular structure corresponds to a chemical, biological, or functional property.

A structure–property relationship describes how molecular structure corresponds to a chemical, biological, or functional property. Quantitative structure–property relationship (QSPR) is the general form; quantitative structure–activity relationship (QSAR) focuses specifically on activity. In virtual screening, the relationship connects structural features with measured compound behavior. section 2.1.1

Use the relationship as a practical reasoning and modeling workflow:

  • Represent molecules through structural features, then compare compounds using a similarity metric that measures shared features. section 2.1.3
  • Train a machine-learning model when labeled positive and negative compounds are available, allowing it to learn which substructures correlate with activity. section 2.1.1 section 2.1.3
  • Apply the idea beyond activity to properties such as toxicity, drug-likeness, size, solubility, and permeability. section 2.1.1

Why similarity is not enough

Two molecules may share substantial structural similarity yet differ in desired functionality because some structural features matter, others are irrelevant, and others contradict the property. A model trained on known antimalarial compounds can learn which substructures most influence antimalarial activity. section 2.1.3

Figure 2.3 Structure-property relationships are nuanced. Discovering similar molecules doesn’t guarantee desired, similar function with respect to a query molecule.

Use this figure to inspect why similar molecules do not necessarily produce similar properties or biological functions. section 2.1.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines QSPR, links features to properties, and notes machine-learning extension while deferring detailed similarity methods.
Examples
Uses active and inactive compounds, drug-likeness properties, and antimalarial substructures as examples.
What it assumes
Assumes molecular structures and property labels, while introducing QSPR, QSAR, structural features, and positive and negative classes.
What you can do after
Evaluate molecular similarity cautiously and identify structural features that may predict activity or other properties.
How it connects
Connects QSPR with QSAR, machine learning, similarity searching, virtual screening, and ADMET prediction.

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.