Whole-Genome Phenotype Prediction with Machine Learning: Open Problems in Bacterial Genomics

TL;DR

This paper reframes bacterial genotype-to-phenotype prediction as an ill-posed causal-inference problem driven by high-dimensional, LD-rich genomes. It introduces a formal GP mapping with representation pipelines and covariates, and outlines Task 1 for learning the GP mapping and Task 2 for fine-mapping causal variants under regularization. It then articulates open problems about reformulating the problem to be well-posed (Hadamard criteria), plus concrete challenges from limited sampling, information loss in representations, confounding, model choice, and the integration of domain knowledge. The authors argue that robust causal discovery in bacterial genomics requires carefully designed representations, constrained hypothesis spaces, explicit handling of population structure and observational noise, and the incorporation of spatial and prior causal information to improve identifiability and interpretability. The work provides a roadmap for future causal ML methods in bacterial genomics with implications for understanding antibiotic resistance and other traits.

Abstract

How can we identify causal genetic mechanisms that govern bacterial traits? Initial efforts entrusting machine learning models to handle the task of predicting phenotype from genotype return high accuracy scores. However, attempts to extract any meaning from the predictive models are found to be corrupted by falsely identified "causal" features. Relying solely on pattern recognition and correlations is unreliable, significantly so in bacterial genomics settings where high-dimensionality and spurious associations are the norm. Though it is not yet clear whether we can overcome this hurdle, significant efforts are being made towards discovering potential high-risk bacterial genetic variants. In view of this, we set up open problems surrounding phenotype prediction from bacterial whole-genome datasets and extending those to learning causal effects, and discuss challenges that impact the reliability of a machine's decision-making when faced with datasets of this nature.

Figures (2)

• Figure 1: Histogram of unitig presence count across 4140 Staphylococcus aureus isolates.
• Figure 2: A: Manhattan plot for bGWAS single nucleotide polymorphism Pyseer results. B: 3-dimensional structure obtained experimentally of a single Staphylococcus aureus protein. Figure adapted from wheeler2019contrasting.