Project AI-driven model cross-species smORF prediction and classification; Owen Rackham

Only a fraction of the genome codes for the proteins in standard reference catalogues. Many small open reading frames (smORFs) scattered across the genome are, in fact, translated into microproteins that those catalogues miss, and distinguishing a genuinely translated smORF from background noise using sequence and experimental data remains an open computational problem at the heart of dark proteome research.

This project will build an AI foundation model that predicts which smORFs are translated and how translation behaves at the codon level, combining state-of-the-art sequence embeddings with cell-type information and ribosome-profiling data. It is the computational backbone of ORFeus Work Package 1 and a hub for the whole network: the model is trained and validated on experimental data from partner projects across the consortium and released as open tools.

The work connects closely to partner projects on rare translation events, genetic variation, codon substitutions, isoform-level translation, and evolutionary analysis.

• Develop a foundation model that captures how mRNA sequence and cellular context determine translation at codon-level resolution, drawing on state-of-the-art sequence-embedding architectures.
• Integrate sequence-based predictions with multi-omics evidence from ribosome profiling and mass spectrometry, including genetic-variant data from the projects of other doctoral candidates (DCs) in the network, here DC3 and DC12, and codon-substitution data from DC13, and refine the model iteratively using isoform-level translation data from DC2.
• Deploy interactive visualisation tools for cross-species and cell-type-specific translation predictions, refined by experimental validation feedback from across the network.
• Collaborate across the ORFeus network, contribute models and tools to the shared ORFeome platform and to Hugging Face (or similar), and produce the project's scientific report.

Methods and platforms
: machine learning and deep learning for genomics (sequence-embedding and foundation-model architectures), ribosome profiling (Ribo-seq) data analysis, integration of mass-spectrometry proteomics and other multi-omics data, and interactive data visualisation, working with public and consortium datasets via the ORFeome platform and Hugging Face.

Secondment
: you will spend around three months at EMBL-EBI (Hinxton, near Cambridge, UK), integrating large-scale mass-spectrometry datasets for cross-species applications of the model. You will also be guided by an independent academic advisor, with the possibility of a short, primarily virtual research exchange to strengthen the experimental validation of the model.

MSCA eligibility

You must meet all of the following on your recruitment date:

• You do not already hold a doctoral degree. If you have defended a doctoral thesis but the degree has not yet been formally awarded, you are not eligible.
• Mobility rule: you must not have lived or carried out your main activity (work, studies, and so on) in the United Kingdom for more than 12 months in the 36 months immediately before your recruitment date. Compulsory national service, short stays such as holidays, and time spent in a procedure to obtain refugee status under the Geneva Convention do not count toward the 12 months.
• You hold, or will hold before the start date, a degree that formally entitles you to enrol in a doctorate, and you can enrol in the doctoral program at the University of Southampton.

Candidates of any nationality may apply. There is no limit on prior research experience, as long as you do not already hold a doctorate.

Project-specific profile

• A relevant degree in bioinformatics, computational biology, computer science, data science, machine learning, or a related quantitative or life-science field with strong computational skills.
• Strong programming skills (for example Python) and hands-on experience with machine learning or deep learning.
• Desirable: experience with genomics or sequencing data (for example ribosome profiling or next-generation sequencing), sequence models and embeddings (including large language models), mass spectrometry or multi-omics data, and an interest in microproteins and the dark proteome.
• Good written and spoken English.
• Motivation for interdisciplinary, collaborative research, and willingness to travel for the secondment and network events.

A full-time employment contract for 36 months as a salaried researcher, with full social security…