Meta-EA and Meta-EA_clinical

Background

Computational methods for estimating missense variant impact suffer from inconsistent performance across genes, which poses a major challenge for their reliable use in clinical practice. While ensemble scores leverage multiple prediction methods to enhance consistency, the overrepresentation of certain genes in the training data can bias their outcomes.

Meta-EA Approach

To address this critical limitation, we generated Meta-EA ensemble scores. These ensemble scores were trained separately for each gene using as reference the Evolutionary Action scores and as features 20 other state-of-the-art scores (PROVEAN v1.1, VEST v4, MutPred v1.2, AlphaMissense, DEOGEN2, REVEL, SIFT4G, MPC v1, MCAP v1.3, CADD v1.4, PrimateAI, SIFT, ESM1b, FATHMM_XF, MutationAssessor v3, MVP v1.0, Polyphen2, Eigen v1.1, MetaLR, MetaSVM). Single gene annotations suggest that the performance of Meta-EA is comparable to the performance of the top individual predicting method for each gene. Gene-balanced ClinVar assessments suggest Meta-EA is the most competitive compared to its component methods.

Meta-EA_clinical Approach

We incorporated the effects of splicing (using SpliceAI scores) and the allele frequency of human polymorphisms (using data from the GnomAD and All of Us projects) to further enhance the performance of Meta-EA, achieving an area under the receiver characteristic curve of 0.97 for both gene-balanced and imbalanced clinical assessments.

A manuscript detailing the Meta-EA and Meta-EA_clinical approaches is currently under peer review at Nature Communications.