Salience, legitimacy and credibility of single-cell sperm data and their evaluation

It is difficult to determine how many sperm donors contributed to a sample, who they were, or what number of sperm they donated. A model analyzing single-cell electropherograms (scEPG) born of diploid cells was previously introduced in [1]. In this paper we extend the model to handle haploid scEPGs, specifically sperm cells, and consider if single-cell forensics is ready to traverse the research-operations boundary. To do this we use the Salience-Legitimacy-Credibility (SLC) framework. In Salience we identify relevant forensic questions. Legitimacy then has us first describe the logic underpinning our haploid-aware model, and verify it was correctly implemented into EESCIt™. Using 121 individuated scEPGs - i.e., singlets - we show that the mean square error between the prediction and the empirical outcomes given that prediction is no more than 3.1 × 10−5 at any locus, demonstrating excellent model fidelity. LR-discrepancy analysis confirms the relative (empirical) frequency and magnitude of H1- and H2- LRs align with declared values, with the largest discrepancy being 0.3 in WoE - i.e., logLR - interval [1, 2]. Credibility analysis showed that no singlet produced a misdirected WoE and we receive, on average, one WoE for every ∼1200 RFU carried by a haploid. We then evaluate EESCIt™ on 33 sperm mixtures comprising 2-5 donors and find that estimates about the number of donors, the number of scEPGs by a donor, and the WoE remain legitimate. The system continues to credibly discriminate source hypotheses, often producing WoE of 10 when the PoI-true contributes as few as two haploids and values near 25 when 10 haploids are present. WoEs remain stable across mixture complexity with no dependence on the true number of contributors, n, showing a robustness unique to single cell data. Additionally, computational time is only mildly impacted by n, with the main determiner being the number of cells isolated. With a legitimate, credible haploid model able to address multifarious salient questions, we integrate it into EESCIt™, establishing a single-cell assessment system that is now capable of evaluating haploid single-cell data.