Skip to main content
. 2022 Nov 30;13:7374. doi: 10.1038/s41467-022-35032-8

Fig. 3. Rank metrics for efficient genetic forensics.

Fig. 3

a For any given lab-of-origin predictor, the X99 score is the smallest positive integer N such that the top-N accuracy of the predictor is at least 99%. Analogous metrics can be defined for other thresholds; for example, the X95 score is the smallest N such that top-N accuracy exceeds 95%. b X99 & X95 scores achieved by each of the top 100 Prediction Track teams, on a logarithmic scale. c X99 & X95 scores achieved by past ML-based approaches to GEA on the Addgene plasmid database, compared to BLAST (left) and the results of the Genetic Engineering Attribution Challenge (GEAC, right). X99 results for the GEAC 1st place and ensemble models are annotated in orange. Dashed grey horizontal line in (b–c) indicates the total number of labs in the dataset, which represents the largest possible value of any X-metric on this dataset.