impMKT: the imputed McDonald and Kreitman test, a straightforward correction that significantly increases the evidence of positive selection of the McDonald and Kreitman test at the gene level

Abstract

The McDonald and Kreitman test is one of the most powerful and widely used methods to detect and quantify recurrent natural selection in DNA sequence data. One of its main limitations is the underestimation of positive selection due to the presence of slightly deleterious variants segregating at low frequencies. Although several approaches have been developed to overcome this limitation, most of them work on gene pooled analyses. Here, we present the imputed McDonald and Kreitman test (impMKT), a new straightforward approach for the detection of positive selection and other selection components of the distribution of fitness effects at the gene level. We compare imputed McDonald and Kreitman test with other widely used McDonald and Kreitman test approaches considering both simulated and empirical data. By applying imputed McDonald and Kreitman test to humans and Drosophila data at the gene level, we substantially increase the statistical evidence of positive selection with respect to previous approaches (e.g. by 50% and 157% compared with the McDonald and Kreitman test in Drosophila and humans, respectively). Finally, we review the minimum number of genes required to obtain a reliable estimation of the proportion of adaptive substitution (α) in gene pooled analyses by using the imputed McDonald and Kreitman test compared with other McDonald and Kreitman test implementations. Because of its simplicity and increased power to detect recurrent positive selection on genes, we propose the imputed McDonald and Kreitman test as the first straightforward approach for testing specific evolutionary hypotheses at the gene level. The software implementation and population genomics data are available at the web-server imkt.uab.cat.

Keywords: McDonald and Kreitman test; natural selection; nucleotide variation; positive selection; protein-coding genes.

Fig. 1.

Hypothetical SFS and fixed differences from Hahn (2018).

Fig. 2.

α MKT estimations by the different MKT approaches under different SLiM simulated scenarios, specifically different simulated fractions of adaptive mutations. Equivalent results under other SLiM simulated scenarios are available in Supplementary Fig. 1.

Fig. 3.

Error biases associated with the α estimations for all of the scenarios and MKT approaches.

Fig. 4.

α estimations at simulations accounting for weak adaptation. Any of the proposed methods can correct linkage and weak adaptation at the estimations. Although the method proposed by Uricchio et al. (2019) can overcome linkage and weak adaptation, α estimations at the gene level remain unexplored and new approaches are required.

Fig. 5.

a) Estimated number of genes under positive selection in the D. melanogaster Zambian population detected by each MKT approach. b) Estimated number of genes under positive selection in the human lineage African populations detected by each MKT approach.

Fig. 6.

Gene pooled analysis. A total of 3,500 random protein-coding genes were sampled from the ZI dataset. We pooled the genes to obtain SFS of 1, 2, 5, 10, 25, 50, 75, 100, 250, 750, and 1,000 genes by resampling them 1,000 times with replacement. a) α estimates with MKT correction. b) Proportion of analysis performed by impMKT. c) Proportion of analysis performed by aMKT. d) Proportion of analysis performed by aMKT. e) Proportion of analysis performed by Grapes.

Fig. 7.

Gene pooled analysis. A total of 3,500 random protein-coding genes were sampled from the human dataset. We pooled the genes to obtain SFS of 1, 2, 5, 10, 25, 50, 75, 100, 250, 750, and 1,000 genes by resampling them 1,000 times with replacement. a) α estimates with MKT correction. b) Proportion of analysis performed by impMKT. c) Proportion of analysis performed by aMKT. d) Proportion of analysis performed by aMKT. e) Proportion of analysis performed by Grapes.