Conditional analysis

In addition to fine-mapping, a more robust way is to use stepwise forward selection (also called iterative conditioning, or conditional analysis) to build iteratively a set S of SNPSs based on an initial "seed" variant, it usually being the lead variant (lowest pval) of a certain region. The next step is to perform a regular analysis where now variants in S are treated as covariates. If the new lead variant is significant, it's added to S and the analysis is repeated as long as lead variants remain within significance. This approach was made popular by GCTA’s Conditional & joint (COJO) analysis of GWAS results. The algorithm works as follows:

initially S is empty
repeat until all P-values outside S are > threshold
  add SNP with the lowest P-value to S
  update P-values of all SNPs 'l' outside S using joint model Y ~ X.S + X.l
end repeat