Ashton Lu
Statistical approaches for integration of single-cell and spatial transcriptomics data at isoform resolution
Motivation:
A fundamental problem in molecular biology is identifying molecular signatures that
characterise cell identity and function. Advances in biotechnology now allow us to
address this problem with unprecedented cellular (e.g., single-cell), spatial, and
molecular (e.g., isoform-level) resolution. Long-read sequencing technologies,
combined with single-cell and spatial transcriptomics assays, now provide data
measuring isoform expression at single-cell or spatial resolution.
A key challenge in using this isoform-resolution data to characterise cell identity is that
each data type has inherent limitations: single-cell data provides isoform expression
at single-cell resolution but lacks spatial context, whereas spatial data captures spatial
organisation but only at multi-cell resolution and typically for a limited set of isoforms.
I propose to develop novel computational approaches that integrate both types of
isoform expression data, fully leveraging their complementary strengths to detect
isoform-level cell-identity signatures.
Proposed Methods:
I will adapt variants of factor models (e.g., [1, 2]) to build isoform-resolution factor
models using shrinkage prior approaches from Bayesian statistics. Specifically, I will
place gene-specific shrinkage priors on isoform-level parameters to stabilise estimates
from noisy isoform data while still capturing genuine biological variability among
isoforms.
Building on this, I will develop integrative factor models that jointly capture information
from both isoform data types through a shared subset of isoforms. To distinguish
between shared and data-type-specific isoform signatures of cell identity, I will model
these components separately. To incorporate spatial information, I will use multi-scale
methods (e.g., wavelets [3, 4]) to model spatial data.
I will perform the inference for the models using variational inference (VI). I will adapt
the VI techniques previously used in variants of factor models including [1] [2].
References:
1. Foo, Y. S., & Shim, H. (2021). A comparison of Bayesian inference techniques for sparse
factor analysis. arXiv preprint arXiv:2112.11719.
2. Wang, W. & Stephens, M (2021), The Journal of Machine Learning Research, 22(1), 5332-
5371.
Ashton Lu
The University of Melbourne
Before commencing his undergraduate studies at the University of Melbourne, Ashton Lu had always been fascinated by both the life sciences and the logic of mathematics. Initially majoring in Physiology, Ashton realised that his true passion lay in the mathematical structures that drive scientific understanding. After completing a Bioinformatics research project under the Metcalf Scholarship at the Walter and Eliza Hall Institute (WEHI), he realised that what captivated him most was not just the biological insights – but the mathematical reasoning behind them. This experience motivated him to switch his focus to a Mathematics major specialising in Statistics and Stochastic Processes, complemented by a Diploma in Computer Science.
As a recipient of the Melbourne Chancellor’s Scholarship and a Dean’s Honour List awardee, Ashton finds immense satisfaction in using quantitative approaches to make sense of complex real-world data. In particular, he is interested in stochastic modelling, statistical learning and inference – areas that connect rigorous mathematical thinking with applications in research and technology.
Outside of mathematics, Ashton has more recently been delving into the field of artificial intelligence, exploring how inference techniques and probabilistic reasoning contribute to the development of intelligent systems. He is fascinated by how AI integrates ideas from statistics, computer science, and cognitive science to replicate aspects of human learning and decision-making.
Beyond his academic pursuits, Ashton enjoys playing pool and bouldering whenever there is a gym nearby. He also finds great satisfaction in teaching mathematics, striving to explain abstract ideas in intuitive and engaging ways. For Ashton, science has always been about “making the world understandable” and he thinks of mathematics as a robust framework designed to cut through real-world noise to unveil the underlying mechanisms that govern it.
You may be interested in
Xiaojuan He
Advanced Optimisation Techniques to the Variable‐Sized 2‐dimensional Bin Packing Problem
