MAVE-NN: learning genotype-phenotype maps from multiplex assays of variant effect.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Multiplex assays of variant effect (MAVEs) are powerful tools for understanding genotype-phenotype relationships.
• Existing methods lack a general strategy for quantitative model inference from MAVE data.
• This limits the full potential of deep mutational scanning and massively parallel reporter assays.

Purpose of the Study:

• To introduce MAVE-NN, a novel Python package for learning quantitative genotype-phenotype maps from MAVE data.
• To provide a broadly applicable, information-theoretic framework for MAVE data analysis.
• To enable the inference of biophysically interpretable models.

Main Methods:

• Development of MAVE-NN, a neural-network-based Python package.
• Implementation of an information-theoretic framework for model learning.
• Application to diverse MAVE datasets, including protein deep mutational scanning and gene regulatory sequence assays.

Main Results:

• MAVE-NN successfully infers quantitative genotype-phenotype maps from various MAVE datasets.
• The approach effectively deconvolves true mutational effects from experimental noise and nonlinearities.
• Biophysically interpretable models can be learned, enhancing biological insights.

Conclusions:

• MAVE-NN offers a general and robust strategy for MAVE data analysis.
• The package facilitates deeper understanding of genotype-phenotype relationships.
• MAVE-NN advances the application of MAVE techniques in biological research.