Predicting phenotype from genotype: Improving accuracy through more robust experimental and computational modeling

Abstract

Computational prediction yields efficient and scalable initial assessments of how variants of unknown significance (VUS) may affect human health. However, when discrepancies between these predictions and direct experimental measurements of functional impact arise, inaccurate computational predictions are frequently assumed as the source. Here we present a methodological analysis indicating that shortcomings in both computational and biological data can contribute to these disagreements. We demonstrate that incomplete assaying of multifunctional proteins can affect the strength of correlations between prediction and experiments; a variant's full impact on function is better quantified by considering multiple assays that probe an ensemble of protein functions. Additionally, many variants predictions are sensitive to protein alignment construction and can be customized to maximize relevance of predictions to a specific experimental question. We conclude that inconsistencies between computation and experiment can often be attributed to the fact that they do not test identical hypotheses. Aligning the design of the computational input with the design of the experimental output will require cooperation between computational and biological scientists, but will also lead to improved estimations of computational prediction accuracy and a better understanding of the genotype-phenotype relationship.

CAGI5: Objective performance assessments of predictions based on the Evolutionary Action equation

“…The CAGI assessments consistently found the Evolutionary Action submissions are amongst the best predictors of the fitness effect of variants. This was shown in older CAGI experiments (Katsonis & Lichtarge, ) and in the most recent one: Best assessor's score for CALM1 , second best correlation for GAA , and best multiclass ROC in ENIGMA amongst submissions that did not rely on prior (circular) annotation data. This last challenge is clinically relevant as variants of unknown significance bedevil the results of BRCA1/2 sequencing in breast cancer clinics.…”

“…We participated in several CAGI challenges as predictors, where we estimated the fitness effect of variants with the Evolutionary Action (EA) method. In older CAGI experiments (CAGI2 to CAGI4) we only submitted predictions on challenges that asked for the impact of individual variants (most often on enzymatic function), where EA was consistently one of the top methods (Katsonis & Lichtarge, ). In the CAGI5 experiment we participated in ten diverse challenges, which also included predictions of protein stability and matching exomes to phenotype.…”

Evolutionary Action ofde novoMissense Variants across Pathways Prioritizes Genes Linked to Autism and Predicts Patient Phenotypic Severity

“…Differentiation of this fitness function yields the EA equation to predict variant impact, in which the perturbation of the fitness landscape is equal to the product of the evolutionary fitness gradient, estimated by Evolutionary Trace (Lichtarge et al 1996), and the substitution log-odds of the amino acid change(Katsonis and Lichtarge 2014). These values are calculable from sequence data and predictions have been shown to correlate well to experimental assessments of protein fitness (Gallion et al 2017) (Katsonis and Lichtarge 2014), consistently outperform machine learning methods (Katsonis and Lichtarge 2017), and to stratify patient morbidity (Katsonis and Lichtarge 2014) and mortality (Neskey et al 2015) in other disease contexts. Here this EA theory is extended by considering the distribution of variant EA scores over a pathway.…”

Prediction and interpretation of deleterious coding variants in terms of protein structural stability

“…Classification methods developed in the literature usually integrate different kinds of biological data in view of gathering the complexity of the phenomenon, and require only the amino acid sequence as input. Some of them, such as Provean, 3,4 Sift 11 or the Evolutionary Action method, 5,6 use as sole ingredient the evolutionary conservation in homologous proteins at the mutated position and in the neighborhood along the sequence. Though this information is represented by a single score, it corresponds to a mixture of several molecular effects, among which protein stability, solubility, function, and interactions.…”