Combined complement gene analyses. Individual SNP analyses suggested several possible combinations of SNPs that protect an individual from developing AMD. To test these, an empirical model was applied first. (a,b) Fraction accounted for by hypothesized model. The model gives four possible combinations of genotypes (a) that would protect from AMD: (i) rs547154 (R32Q) is G/A AND rs1061170 (Y402H) is C/T, (ii) rs547154 is G/A AND rs1061170 is C/C, (iii) rs4151667 (L9H) is T/A AND rs1061170 is C/T, and (iv) rs4151667 is T/A AND rs1061170 is C/C. Application of this model resulted in the distributions shown in b for the Iowa, Columbia and combined cohorts. The case percentage is the percentage of cases for which the model was false; that is, they did not have protection as described by the model. The control percentage is the percentage of controls that did have the protective factors. (c,d) Fraction accounted for by the Exemplar machine-built model. c shows the best-performing model, which describes four possible individual or combinations of genotypes that protect from AMD (that is, combinations resulting in the model being ‘true’): (i) rs1048709 (R150R) is G/G AND rs1061170 is C/C, OR (ii) rs547154 is G/A, OR (iii) rs4151667 is T/A OR (iv) CFH intron 1 variant is delTT. The model performance is shown in d for the Iowa, Columbia, and combined cohorts.