The (in)dependence of single-cell data inferences on model constructs.

Area of Science:

• Forensic Science
• Genetics
• Biostatistics

Background:

• Single-cell analysis offers advanced forensic insights, but interpretation models can yield varied evidence weights (WoEs).
• Understanding model discrepancies is crucial for consistent forensic evidence review.

Purpose of the Study:

• To evaluate and compare the Weights of Evidence (WoEs) generated by three distinct probabilistic models for single-cell electropherograms (scEPGs).
• To identify sources of inconsistency in forensic single-cell data interpretation across different models.

Main Methods:

• Performance testing of three models (EESCIt, TD, DCM) on 996 scEPGs with true and false contributor tests.
• Analysis of 201,192 outcomes per model to assess hypothesis resolution and WoE calibration.
• Paired analyses using intraclass correlations to evaluate inter-model agreement for scEPG WoEs.

Main Results:

• All three models effectively resolved hypotheses for scEPGs.
• Weights of Evidence (WoEs) showed strong agreement across models, with intraclass correlations ≥ 0.99997.
• Model calibration was generally accurate, with rare discrepancies linked to extreme stutter signals in specific scEPGs.

Conclusions:

• The evaluated models (EESCIt, TD, DCM) appropriately state WoEs for scEPGs within their relevant ranges.
• Findings support the legitimacy of single-cell data interpretation in forensics across these models.
• Proposed interpretive adaptations can ameliorate differences in predicting rare events.