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

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Computational predictions offer scalable initial assessments of variant health impacts.
• Discrepancies between computational predictions and experimental data are common.
• Inaccurate computational predictions are often presumed as the sole cause of these disagreements.

Purpose of the Study:

• To analyze methodological reasons for discrepancies between computational predictions and experimental functional impact data.
• To identify contributions of both computational and biological data limitations to these disagreements.
• To propose strategies for improving the concordance between computational and experimental assessments of variant effects.

Main Methods:

• Methodological analysis of computational prediction and experimental assay data.
• Investigation of multifunctional protein assaying completeness.
• Examination of variant prediction sensitivity to protein alignment construction.
• Customization of computational predictions for specific experimental contexts.

Main Results:

• Shortcomings in both computational and biological data contribute to prediction-experiment discrepancies.
• Incomplete assaying of multifunctional proteins weakens correlations between predictions and experiments.
• Considering multiple assays for protein functions improves impact quantification.
• Variant predictions are sensitive to alignment construction and can be tailored.

Conclusions:

• Inconsistencies often stem from computational and experimental methods testing different hypotheses.
• Aligning computational input design with experimental output design is crucial.
• Cooperation between computational and biological scientists will enhance prediction accuracy.
• Improved alignment will lead to better genotype-phenotype relationship understanding.