When (Q)SAR model predictions conflict, are inconclusive, or contain structural features requiring further interpretation, additional evidence can help support a scientifically defensible expert review.
A standard ICH M7 evaluation involves two complementary methodologies using a statistical and an expert rule-based model. Each model evaluates structural features, or fragments, of a query compound against information represented in the model. When a model detects one or more fragments, a probability is computed. If the probability falls above the model’s decision threshold, the compound is predicted positive; below it, negative. A probability calculated inside the model’s grey zone gives an inconclusive result. Depending on the strength and coverage of the available evidence, the resulting predictions can contain any combination of positive, negative, inconclusive, or out-of-domain output from each model.
Interpreting these outcomes, particularly when the models disagree or evidence is insufficient, requires expert judgement. While reviewing individual model descriptors and their supporting evidence can provide the context needed to make the final determination, the process can become tedious and time-consuming. This is especially true when predictions conflict or contain unfamiliar structural features.
CASE Ultra’s Konsolidator tool was developed to streamline the expert review process by organizing relevant evidence together, reducing the time it takes to perform expert review.
Konsolidator uses the results from CASE Ultra’s statistical (GT1_BMUT) and expert rule-based (GT_EXPERT) models as the starting point for a broader evaluation. The tool examines structural alerts, relevant analogs, unknown structural features, experimental data, and other supporting information. These findings are presented alongside the original model predictions in a structured report, beginning with high-level results and progressing to increasingly detailed evidence. Altogether, this information contributes to a suggested outcome to help inform the assessor’s final determination.
– Combines statistical and expert rule-based model results in a single report.
– Produces both whole structure and alert-specific analogs of the query chemical.
– Compares the query chemical and identified alerts with experimentally tested APIs and other reference substances.
– Provides additional evidence for evaluating the relevance of identified alerts.
– Organizes supporting experimental data and literature references for review and documentation.
Konsolidator does not replace expert judgment, nor does its suggested outcome automatically determine the final ICH M7 classification.
The Konsolidator tool by itself does not override a CASE Ultra prediction. Instead, Konsolidator can provide evidence that allows an expert to determine whether the original model prediction should be accepted, qualified, or overturned as part of the final assessment.
For example, a statistical model may identify a positive alert, while a refined surrogate search reveals experimentally negative analogs containing the same alert within highly relevant structural environments. That evidence may warrant further investigation of the alert’s relevance to the query.
The resulting expert conclusion should therefore be understood as an expert-reviewed interpretation of the model output, rather than a new prediction generated by Konsolidator.
The Konsolidator report provides a structured record of the evidence generated during the review, including model outcomes, alerts, analogs, experimental information, and supporting literature.
However, the report itself should not be viewed as a substitute for the assessor’s scientific rationale, particularly when expert review changes or qualifies the interpretation of an original model result.
A regulatory reviewer should be able to understand not only what evidence was identified, but also why that evidence supports the final conclusion.
For a straightforward concordant result, little additional discussion may be necessary. For conflicting, inconclusive, out-of-domain, or expert-overridden predictions, additional explanation becomes increasingly important.
The objective is to make the scientific reasoning behind the final call transparent and defensible.
If you have further questions regarding the Konsolidator Tool methodology, or ICH M7 evaluations in general, please contact info@multicase.com.