Skip to main content
. Author manuscript; available in PMC: 2024 Jan 11.
Published in final edited form as: Artif Intell Med. 2023 Feb 28;139:102523. doi: 10.1016/j.artmed.2023.102523

Figure 7. FQ-HP embeddings improve recognition of expert-curated phenotype similarities.

Figure 7.

(A) A comparison on the overall accuracy of the similarity-measuring algorithms on 100 expert-curated phenotypic trios. The accuracy of our method (FQ-HP) surpasses the other similarity-measuring techniques where black bars show 95% confidence intervals, and the green dashed line represents the experts’ agreement level. (B) Histogram of experts’ agreement level for the 100 trios. Almost half of the records belong to the fair-level agreement category, indicating a wide variability in the clinical assessment of the experts and the challenging nature of such evaluations. (C) Performance of the similarity-measuring algorithms on the records of each agreement level. While all techniques have almost an increasing pattern in their performance as the agreement level improves, FQ-HP achieves better accuracy than other techniques, including Resnik, in the substantial and high agreement levels.