Abstract
Paralogous genes challenge short-read sequencing (SRS) due to high sequence similarity. Although long-read sequencing (LRS) improves resolution, the extent to which it resolves paralogous genes remains unclear. This study evaluates the capability of LRS by integrating in silico mappability-based predictions with clinical data to generate SRS- and LRS-unresolved gene lists, and by assessing whether a paralog-specific phasing, Paraphase, can overcome remaining limitations. Mappability was simulated across read lengths (250 bp to 14 kb) to predict unresolved regions and validated against mapping quality (MQ) from 66 high-fidelity LRS samples. Paraphase was applied to 79 paralog groups. Among 645 medically relevant (MR) genes unresolved by SRS, 419 (65.0%) were predicted to be resolved by LRS, while 226 (35.0%) remained unresolved. These predictions correlated with clinical MQ (χ² = 92.43, p < 2.2 × 10(-16); κ = 0.37), with significant differences between LRS-resolved and LRS-unresolved MR genes (W = 63,656, p < 2.2 × 10(-16); r = 0.36). Paraphase resolved 61 groups (77.2%), providing additional resolution beyond LRS. LRS improves paralogous gene resolution but cannot fully eliminate paralog blind spots. Curated gene lists define boundaries of LRS utility for clinical interpretation, while Paraphase adds complementary resolution, supporting an integrated framework combining predictive modeling with algorithmic strategies.