MAVE-NN

MAVE-NN infers quantitative genotype-phenotype (G-P) maps from multiplex assays of variant effect (MAVEs), including deep mutational scanning (DMS) for proteins and massively parallel reporter assays (MPRAs) for gene regulatory sequences.

Key Features:

• Neural Network Framework: Employs neural networks within an information-theoretic framework to model relationships in MAVE-derived data.
• Biophysically Interpretable Models: Produces quantitative models that are interpretable in biophysical terms linking genetic variants to phenotypic outcomes.
• Deconvolution of Mutational Effects: Deconvolves mutational effects from confounding factors such as experimental nonlinearities and noise present in high-throughput assays.

Scientific Applications:

• Inference of genotype-phenotype maps: Infers quantitative G-P maps from MAVE datasets to relate sequence variants to functional measurements.
• Analysis of DMS and MPRAs: Applies to deep mutational scanning of proteins and massively parallel reporter assays of gene regulatory sequences.
• Functional interpretation of variants: Characterizes the functional consequences of genetic variation at scale using high-throughput assay data.

Methodology:

MAVE-NN uses neural networks integrated with information-theoretic principles to learn G-P maps from MAVE data and to deconvolve mutational effects from experimental nonlinearities and noise, yielding biophysically interpretable models.