Read-Across Analog

Estimate an analog endpoint by caller-scaled descriptor distances and inverse-distance weights; exact descriptor matches are averaged, and applicability/mechanistic similarity are not inferred.

Description

Predict a supplied endpoint using scaled descriptor distances and normalized inverse-distance analogue weights.

Read-across uses evidence from source substances to inform an endpoint for a target substance. Its scientific justification depends on substance identity, relevant similarity, and the reliability of the source studies. Numerical closeness alone does not establish a valid read-across case.^[1]

This calculator receives descriptors and analogue endpoint measurements that you have already selected. A descriptor is a numerical property; its scale controls how much a difference in that property contributes to the overall distance. The tool returns a weighted endpoint estimate in the same units as the supplied endpoints.

When to use Read-Across Analog

Use it to inspect a transparent distance-weighted analogue estimate or compare descriptor scales after deciding which substances and studies are suitable.

How to use Read-Across Analog and understand its results

Enter a nonempty targetDescriptors vector and equally long positive descriptorScales. Each analogDescriptors row must have the same length.
Pair each analogue row with one analogEndpoints value and enter a positive distancePower. Use consistent descriptor definitions and endpoint units.
Run the tool and read standardizedDistances, normalizedAnalogWeights, predictedEndpoint, and exactMatchFound together.

d_i = sqrt[Σ_j ((x_ij − target_j)/scale_j)²]; w_i = (d_min/d_i)^power / Σ_k (d_min/d_k)^power

When no distance is zero, this stable relative-distance expression is equivalent to normalized inverse-distance weighting.

The prediction is Σ(weight × analogue endpoint). If any distance is exactly zero, the implementation gives equal weights only to those matching descriptor vectors and gives zero weight to every other analogue.

json

{
  "input": {
    "targetDescriptors": [
      0,
      0
    ],
    "analogDescriptors": [
      [
        1,
        0
      ],
      [
        0,
        2
      ]
    ],
    "descriptorScales": [
      1,
      1
    ],
    "analogEndpoints": [
      10,
      20
    ],
    "distancePower": 2
  },
  "output": {
    "standardizedDistances": [
      1,
      2
    ],
    "normalizedAnalogWeights": [
      0.8,
      0.2
    ],
    "predictedEndpoint": 12,
    "exactMatchFound": false
  }
}

Two exact descriptor matches with endpoints 10 and 20 predict 15. Exact matching concerns the entered vectors, not chemical identity. Negative finite descriptor and endpoint values are allowed; zero scales and empty or mismatched arrays are rejected.

Limitations and assumptions

The tool does not calculate molecular descriptors, choose analogues, fit scales or power, identify an applicability domain, or estimate prediction uncertainty.

Repeated analogue rows are separate observations and can change weights. Endpoint averaging may be inappropriate for logarithmic values, censored observations, or incompatible assays unless prepared accordingly.

Alternative or Complementary methods

Use hypothesis-based read-across assessment alongside the numeric result. A calibrated property model may be preferable for a supported endpoint; a margin calculation can compare an independently justified point of departure with exposure.

References

Read-Across Assessment Framework (RAAF) — European Chemicals Agency

Similar or alternative tools

Skin Permeability Coefficient
Evaluate a supplied descriptor-based skin-permeability equation.
Margin of Exposure
Compare a justified point of departure with a compatible exposure estimate.