Improving fine-mapping by modeling infinitesimal effects.

Area of Science:

• Human Genetics
• Statistical Genetics
• Genomic Medicine

Background:

• Bayesian fine-mapping algorithms are crucial for identifying causal genetic variants associated with phenotypes.
• Assessing the calibration of posterior probabilities in real-world genetic data is challenging due to potential model misspecification and unknown true causal variants.

Purpose of the Study:

• To develop a novel metric, replication failure rate (RFR), for evaluating fine-mapping consistency.
• To introduce new fine-mapping methods, SuSiE-inf and FINEMAP-inf, that model both infinitesimal and sparse causal effects.
• To improve the accuracy of causal variant identification and polygenic risk score prediction.

Main Methods:

• Introduced replication failure rate (RFR) by downsampling to assess fine-mapping consistency.
• Developed SuSiE-inf and FINEMAP-inf, which incorporate infinitesimal effects alongside sparse large effects.
• Conducted simulations to evaluate the impact of genetic architecture, imputation noise, and quality control on calibration.

Main Results:

• Existing methods (SuSiE, FINEMAP, COJO-ABF) exhibited high RFR, suggesting overconfidence.
• Non-sparse genetic architectures were identified as a cause of miscalibration.
• SuSiE-inf and FINEMAP-inf demonstrated superior calibration, lower RFR, enhanced functional enrichment, and competitive recall.
• Posterior effect sizes from the new methods significantly improved polygenic risk score accuracy.

Conclusions:

• Replication failure rate (RFR) provides a valuable assessment of fine-mapping consistency.
• The novel SuSiE-inf and FINEMAP-inf methods offer improved calibration and accuracy for causal variant identification.
• These advancements contribute to more reliable genetic insights for complex traits and personalized medicine.