liveBook · Manning

Anomaly

As explained in Build AI Drug Discovery Pipelines

An outlier is a data point whose value is unusually distant from the pattern or distribution of the other observations.

An outlier is an observation unusually distant from the others. In aqueous-solubility data, extreme logS values can indicate compounds with unusual physicochemical properties that may not occur in production and can hinder model learning. section 4.1.2

Ways to identify or account for unusual observations include:

  • Use the interquartile range (IQR): flag values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR. section 4.1.2
  • Inspect extreme compounds using chemical knowledge; a wide solubility range may signal unusual structures. section 4.1.2
  • For robust model fitting in the presence of outliers, RANSAC is one available method. section 4.5
Figure 4.2 The least soluble molecules (top row) and the most soluble molecules (bottom row) in the training data set.

Compare the least and most soluble training-set molecules when inspecting chemical extremes. section 4.1.2

Anomaly detection can also use density estimation: observations mapping to low-probability regions are possible outliers. section 7.0

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines outliers and gives IQR thresholds, but does not develop a complete anomaly-detection workflow.
Examples
Uses aqueous-solubility values from −12 to 1.97 logS and contrasts them with a typical drug-like range near −6 to −3.
What you can do after
Diagnose potential outliers by applying IQR thresholds, inspecting chemical extremes, or identifying low-probability regions.
How it connects
Connects outliers to solubility modeling, interquartile range, chemical domain knowledge, density estimation, and RANSAC.

Also explained in

Compare how 20 books explain anomaly →

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.