Fig 1.
Schematic overview of process for assessing the functional effect of human disease-associated variants via complementation testing.
A. We selected paralog pairs where a human disease protein has a yeast paralog for which all protein domains are also found in the human protein. Homologous pairs of domains are connected by solid lines, while non-homologous domain pairs are connected by a dashed line. B. For a subset of those paralog pairs for which we identified complementation relationships, we used these relationships to assess whether the functionality of variants in these assays predicted variant pathogenicity.
Fig 2.
Protein domain architecture of yeast Kin28 and human paralogs.
Shown are yeast Kin28 (red text), and human paralogs tested for complementation (in blue text if we found complementation and black text otherwise). Protein domain patterns Pkinase_Tyr (PFAM pattern PF07714) and Pkinase (PFAM pattern PF00069) are indicated in light and dark blue, respectively.
Table 1.
Seven human genes can complement yeast Kin28.
Fig 3.
Functional assay and protein domain architecture of yeast Cak1 and its complementing human paralogs.
(A) Functional complementation assay results showing that expression of human proteins TBK1 and CDK7 complements defects in a strain (YFL029C_tsa650) that encodes a temperature sensitive variant of Cak1 (described as “cak1-ts” above). (B) Pkinase domains are shown in dark blue. Complementing paralogs indicated in blue text.
Fig 4.
Relating sequence similarity and ability of a paralog to complement.
The average percent identity (PID) score distribution is shown for human-yeast pairs such that multiple human paralogs were tested for a given yeast protein (A), and for human-yeast pairs such that multiple yeast paralogs were tested for a given human protein (B). In each case, the distribution is shown separately for complementing and non-complementing pairs. Each bin height is the count of human or yeast genes having a PID within the appropriate range for that bin. That complementing and non-complementing distributions are both shifted in positon relative to one another and highly overlapping suggests that sequence similarity is an informative but imperfect predictor of complementation.
Table 2.
Deleteriousness predictions from functional complemention (FC), Polyphen-2 (PPH2) and PROVEAN.
Table 3.
Pathogenicity prediction performance for the human disease gene paralog test set.
Fig 5.
Ability of functional complementation to predict pathogenicity.
(A) Distribution of FC scores for disease associated (red line) or non-disease-associated variants (blue line). FC scores from paralog-based complementation assays are significantly higher for disease-associated variants than non-disease-associated variants (P-value, Wilcoxon test). (B) Precision vs. recall performance for functional complementation scores (both paralog- and ortholog-based), PolyPhen-2 scores, and various options for combining the two approaches (see Methods).
Fig 6.
Performance of pathogenic variant identification does not strongly depend on whether the variant is in the aligned region.
Here we show precision vs recall performance for varants that either do (‘aligned’) or do not (non-aligned) fall within the sequence region that can be aligned between human and yeast homologs.
Fig 7.
The kinome tree of yeast Kin28 and its kinase paralogs tested here.
Kinases that can complement yeast Kin 28 were colored in pink, other kinases tested for ability to complement yeast Kin28 were colored in cyan. (The image was generated from the Kinome-Render Tool [49] hosted at Cell Signaling, Inc.).
Table 4.
Numbers of human disease-associated genes with orthologs and paralogs in five model species.