Interpretable Vector Symbolic Architectures for scalable Genome Interpretation

We are looking for a motivated PhD student to work at the interface of Artificial Intelligence, Machine Learning, Bioinformatics, and Genomics. The project will investigate the use of Vector Symbolic Architectures (VSA), also known as Hyperdimensional Computing, as a new framework for predicting phenotypes and disease risk directly from genomic sequencing data.

Project description

Genome Interpretation aims to understand how genetic variation determines phenotypes, including quantitative traits and disease risk. This remains a challenging machine-learning problem because genomic datasets contain millions of variables but comparatively few samples, making conventional neural networks prone to overfitting and difficult to interpret.
The PhD project will develop a new approach based on Vector Symbolic Architectures, in which genomic information is represented using high-dimensional vectors and compositional algebraic operations.
The student will develop methods to encode genomic information at multiple biological scales:

variant → gene → pathway → individual

using operations such as binding, bundling, and permutation. These representations will then be used to build predictive models of genotype–phenotype relationships.
The project will first be prototyped using yeast whole-genome sequencing data and hundreds of quantitative phenotypes, allowing rapid comparison of different VSA representations and learning strategies. The methods will subsequently be applied to human exome sequencing data , using publicly available case-control cohorts.
A major component of the project will focus on interpretability. The student will develop methods based on VSA decoding and unbinding to identify the variants, genes, and biological pathways contributing to individual predictions and compare the discovered signals with known disease-associated loci.

The project therefore combines methodological development in machine learning with applications to real genomic datasets and clinically relevant problems.

Main research objectives

The PhD student will:
    • develop multi-scale VSA representations for genomic sequencing data; investigate different hypervector representations, including binary, bipolar, ternary, and continuous encodings;
    • design supervised and potentially self-supervised learning methods operating on genomic hypervectors;
    • benchmark VSA models against conventional machine-learning and deep-learning approaches;
    • develop interpretable decoding methods to quantify variant-, gene-, and pathway-level contributions;
    • apply the developed methods to yeast genotype–phenotype prediction and human IBD disease-risk prediction;
    • analyze the biological relevance of the associations discovered by the models;
    • publish the methodological and biological results in international journals and conferences.
    

Required skills

Candidates should have a Master's degree, or equivalent, in Computer Science, Artificial Intelligence, Machine Learning, Bioinformatics, Computational Biology, Applied Mathematics, or a related discipline.
Strong candidates should have:
    • solid Python programming skills and pytorch;
    • bioinformatics or computational genomics knowledge;
    • good knowledge of machine learning and statistical learning;
    • familiarity with linear algebra, vector representations, and optimization;
    • experience working with scientific datasets;
    • ability to independently design, implement, and evaluate computational methods;
    • good written and spoken English (at least B2). The job interview will be in english

Useful but not mandatory experience

Experience in one or more of the following would be advantageous:
    • deep learning and PyTorch;
    • hyperdimensional computing or Vector Symbolic Architectures;
    • genomic data formats such as VCF;
    • population genetics or genotype–phenotype prediction;
    • dimensionality reduction and representation learning;
    • interpretable or explainable machine learning;
    • high-performance computing and large-scale data analysis.

Previous biological training is not required, provided the candidate is interested in learning the necessary genomics and genetics concepts.

Candidate profile

We are particularly interested in candidates who enjoy developing new machine-learning methodology rather than only applying existing models. The project requires a combination of algorithmic thinking, mathematical reasoning, programming, and curiosity about biological problems.
The successful candidate will work on a highly interdisciplinary project at the frontier between AI and genome biology, with the opportunity to develop a largely unexplored computational paradigm for interpretable genome analysis.

Salary is 2300€ brut. The start date is the 1/11/2026 (mandatory).

Applications should be sent to daniele.raimondi@igmm.cnrs.fr with the following object:

[PHD VSA]: name applicant  

(Emails are automatically filtered).