PhD student in computational biology · University of Edinburgh · PRIMED

Machine learning for stochastic single-cell biology.

I am Xi Yang (Ian), AMRSC. My PhD research uses stochastic modelling, time-series analysis, self-supervised learning, and time-lapse microscopy to characterise metabolic oscillations in single cells.

View research GitHub LinkedIn
Xi Yang (Ian) headshot
Single-cell dynamics Stochastic simulation Contrastive learning

Current Focus

My project sits between mathematical biology and machine learning: modelling biological rhythms in baker's yeast, extracting information from noisy transcription-factor localisation traces, and testing whether simulation-trained encoders can transfer to real microscopy data.

Research groups
Swain lab and Grima group
PhD period
October 2024 - September 2028
Methods
SSA, telegraph models, SimCLR, SVM, transformers

Research

From synthetic trajectories to living-cell signal interpretation.

A contrastive encoder trained only on stochastic gene-expression simulations can be reused on real yeast time-lapse microscopy traces, helping separate transcription-factor identity and stress conditions without manual feature design.

Label-free contrastive learning reads stress in yeast.

The workflow starts with two-state telegraph model simulations generated by SSA. SimCLR learns an embedding space from paired synthetic trajectories, the projection head is discarded, and the frozen encoder is evaluated on real transcription-factor localisation traces.

  • 79%6-class experimental accuracy
  • 51%12-class experimental accuracy
  • 1,024Sobol-sampled parameter sets
Experimental transcription-factor localisation traces under glucose conditions
Experimental TF localisation traces and benchmark classification results.
Budding yeast cells in a microfluidic device
Time-lapse microscopy of fluorescently tagged yeast cells.
t-SNE embedding comparing catch22 features with SimCLR features
Embedding spaces show clearer separation after SimCLR feature extraction.
Model accuracy comparison for synthetic and experimental time-series classifiers
Model comparisons across raw features, catch22, LSTM, transformers, and SimCLR pipelines.

Writing And Recognition

Science communication, awards, and community work.

Awards and leadership

  • Student Experience Grant - BioHackathon Edinburgh
  • Student Experience Grant - ChemPALS
  • 2023 Community Prize, School of Chemistry
  • Edinburgh Award: Leadership in Student Opportunities
  • Saltire Award: The Ascent

Contact

Open to collaborations across computational biology, modelling, and scientific software.

I am based in Edinburgh and especially interested in conversations around single-cell time-series analysis, self-supervised learning for biology, and reproducible research tooling.

LinkedIn profile GitHub profile