Incremental Inverse Design of Desired Soybean Phenotypes.

Area of Science:

• * Agricultural Science
• * Computational Biology
• * Bioinformatics

Background:

• * Conventional forward design relies on trial-and-error for biological optimization.
• * Inverse design offers a paradigm shift by directly optimizing genotype for desired phenotypes.
• * Challenges in genotype-to-bulk phenotype (G-BP) mapping include the 'one-to-many' nature of inverse functions and biological viability constraints.

Purpose of the Study:

• * To present a foundational synthesis of inverse design principles applied to G-BP optimization.
• * To propose a novel design paradigm combining computational and experimental approaches for incremental phenotype optimization.
• * To automate both design and learning phases within the

Main Methods:

• * Developed a computational inverse design pipeline integrating a random forest (RF) model for genotype-to-phenotype (G-to-P) relationship prediction.
• * Employed a genetic algorithm to efficiently search for feasible genotypes with optimized phenotypes.
• * Utilized an in silico case study on a soybean nested association matrix dataset.

Main Results:

• * The proposed pipeline successfully optimized soybean protein content.
• * Achieved a mean protein content of 36.13% after 20 design, build, test, learn (DBTL) cycles.
• * Demonstrated a three standard deviation increase in protein content above the original population mean.

Conclusions:

• * Computational inverse design, when integrated with DBTL cycles, is effective for G-BP optimization.
• * The pipeline can suggest specific genotypic modifications or optimal parents for selective breeding.
• * This approach offers a powerful, data-driven strategy for accelerating crop improvement and other biological engineering applications.