Probabilistic Machine Learning for Genome Recovery and Microbial Data

This opportunity focuses on probabilistic machine learning for large-scale metagenomic data, inspired by the DarkScience research project on microbial dark matter . I offer expertise in probabilistic generative models, uncertainty-aware deep learning, and Bayesian methods applied to microbial genome recovery, graph-based embeddings, and high-dimensional biological signals.

Recent advances—such as long-read Nanopore sequencing, graph-structured assembly data, and epi-genetic modification signals—create a unique setting where classical ML struggles. Probabilistic ML provides the principled foundation needed to capture uncertainty, guide clustering, and support reliable genome binning at massive scale.

Potential collaboration topics include:
• Uncertainty-aware joint representation learning (e.g., variational autoencoders + probabilistic clustering)
• Probabilistic embeddings for assembly graphs and heterogeneous genomic-environmental data
• Bayesian and PAC-Bayesian approaches for model reliability in high-noise biological domains
• Deep probabilistic models for structure discovery in raw sequencing signals
• Probabilistic methods for identifying “dark spots”: missing species, unexplored regions of the tree of life, and high-value samples
• Uncertainty-driven exploration strategies for large-scale comparative genomics

This work is relevant to researchers in ML, computational biology, bioinformatics, probabilistic modelling, and data-centric bioscience who are interested in pushing the frontier of uncertainty-aware biological AI.