BPS2026 – Modeling relationships in multi-phenotype deep mutational scanning using graph theory and ensemble view

Deep mutational scanning (DMS) enables high-throughput mapping of protein sequence-to-function relationships. It can simultaneously measure the effects of millions of mutations of a protein at the same time. Multi-phenotype DMS designs, such as those that measure both protein abundance and activity, are becoming increasingly common. We catalogued over 50 such data sets, whose dates range from 2012 to 2025. However, the methodology for designing multi-phenotype DMS experiments and subsequently analyzing the data remains highly ad hoc and underdeveloped. To more systematically address this issue, we introduce the Bee framework. The Bee framework represents a multi-phenotype DMS experiment as a directed acyclic graph (DAG), where each vertex corresponds to an ensemble-level protein phenotype (i.e., the average behavior of a protein variant across the cell pool) and edges indicate biological dependencies (e.g., protein activity requires proper folding; surface expression depends on trafficking). Based on the graph structure, we decompose multi-phenotype DMS experiments as combinations of atom types: parallel, sequential, and multi-condition. We categorize existing studies under this framework and show through simulation and real data that the reporter assay’s selection criterion and its dynamic range can introduce variant-specific, numerically nonlinear relationships between phenotypes. Insights through the Bee model and the accompanying simulation framework can provide guidance for more efficient experimental design and data analysis of multi-phenotype DMS experiments.