StarBASE-GP: Biologically-Guided Automated Machine Learning for Genotype-to-Phenotype Association Analysis

TL;DR

StarBASE-GP presents a biologically guided AutoML framework that uses genetic programming to evolve multi-node genotype–phenotype analysis pipelines. By embedding nine inheritance encodings, a dynamic LD-pruning node, and a SNP database to guide variant selection, the method seeks to maximize explanatory power ($r^2$) while minimizing pipeline complexity, addressing the small $n$, large $p$ challenge. Across brown rat BMI_Tail data, StarBASE-GP yields superior Pareto fronts, higher QTL identification accuracy, and robust SNP-consistency signals, including potential novel loci, while modeling additive and non-additive genetic variance. The approach demonstrates the value of domain-informedAutoML for variant discovery in complex traits, with practical implications for scalable GPA analyses and downstream validation.

Abstract

We present the Star-Based Automated Single-locus and Epistasis analysis tool - Genetic Programming (StarBASE-GP), an automated framework for discovering meaningful genetic variants associated with phenotypic variation in large-scale genomic datasets. StarBASE-GP uses a genetic programming-based multi-objective optimization strategy to evolve machine learning pipelines that simultaneously maximize explanatory power (r2) and minimize pipeline complexity. Biological domain knowledge is integrated at multiple stages, including the use of nine inheritance encoding strategies to model deviations from additivity, a custom linkage disequilibrium pruning node that minimizes redundancy among features, and a dynamic variant recommendation system that prioritizes informative candidates for pipeline inclusion. We evaluate StarBASE-GP on a cohort of Rattus norvegicus (brown rat) to identify variants associated with body mass index, benchmarking its performance against a random baseline and a biologically naive version of the tool. StarBASE-GP consistently evolves Pareto fronts with superior performance, yielding higher accuracy in identifying both ground truth and novel quantitative trait loci, highlighting relevant targets for future validation. By incorporating evolutionary search and relevant biological theory into a flexible automated machine learning framework, StarBASE-GP demonstrates robust potential for advancing variant discovery in complex traits.

Figures (5)

• Figure 1: The eight "strict" inheritance models used in StarBASE-GP. On the x-axes are the three typical genotype encodings found in SNP datasets ($0$, $1$, and $2$: additive). On the y-axes is the expected phenotype increasing from $0$ to $1$. The red numbers above horizontal bars within plots represent the StarBASE-GP genotype encodings enforced under each respective inheritance model.
• Figure 2: StarBASE-GP pipeline structure. The various shapes for SNP nodes represent different inheritance model encodings. Colored circles in the legend (left) describe the node types each pipeline contains. All StarBASE-GP pipelines follow this defined structure. However, pipelines differ in their node hyperparameters, as well as in the number and composition of SNP nodes.
• Figure 3: Characteristics of the Pareto fronts for each experimental condition. Panels present raincloud plots for hypervolume (A), number of front solutions (B), and the count of unique bins (C).
• Figure 4: Manhattan plots of mean SNP consistency scores across replicates for Pareto front SNPs by experiment. Red "QTL" labels mark genomic locations of the four ground truth QTLs. Green stars denote potentially novel QTLs.
• Figure 5: Boxplots of BMI_TAIL residuals across genotype classes for ground truth and putative QTLs. The most common StarBASE-GP encoder selection is listed above the best-fit inheritance model from the full dataset. Bolded values above each boxplot represent the normalized (between 0 and 1) observed mean phenotype per genotype class in the full dataset.