arXiv subjects

Matthew Lyon

Publications and source records attributed to Matthew Lyon.

4 recordsLinked to original sources

arXiv2026

Single-Cell Cross-Modal Transfer by Adversarial Fine-Tuning of Foundation Models

Spatial transcriptomics (ST) is a powerful tool for exploring biological properties dependent on structure, proximity, and interaction in tissue. The methods underpinning ST are developing rapidly but are limited in their ability to profile many thousands of genes at a subcellular scale. Although dissociated from tissue, it is known that the whole-transcriptome readouts of cells in single-cell RNA sequencing (scRNA-seq) retain information about their former in situ neighbourhoods, motivating computational methods to recover it. While paired ST and scRNA-seq datasets are scarce, each modality in its own right is abundantly available. We therefore propose to perform cross-modal translation between unpaired ST and scRNA-seq data. In this work we show that a single-cell foundation model can perform this translation via adversarial fine-tuning. We demonstrate that our method performs favourably against methods built for multi-omics translation.

q-bio.GN↗

arXiv2025

Beyond Automation: Redesigning Jobs with LLMs to Enhance Productivity

The adoption of generative artificial intelligence (AI) is predicted to lead to fundamental shifts in the labour market, resulting in displacement or augmentation of AI-exposed roles. To investigate the impact of AI across a large organisation, we assessed AI exposure at the task level within roles at the UK Civil Service (UKCS). Using a novel dataset of UKCS job adverts, covering 193,497 vacancies over 6 years, our large language model (LLM)-driven analysis estimated AI exposure scores of 1,542,411 tasks. By aggregating AI exposure scores for tasks within each role, we calculated the mean and variance of job-level exposure to AI, highlighting the heterogeneous impacts of AI, even for seemingly identical jobs. We then use an LLM to redesign jobs, focusing on task automation, task optimisation, and task reallocation. We find that the redesign process leads to tasks where humans have comparative advantage over AI, including strategic leadership, complex problem resolution, and stakeholder management. Overall, automation and augmentation are expected to have nuanced effects across all levels of the organisational hierarchy. Most economic value of AI is expected to arise from productivity gains rather than role displacement. We contribute to the automation, augmentation and productivity debates as well as advance our understanding of job redesign in the age of AI.

econ.GN↗

arXiv2023

Spatio-Angular Convolutions for Super-resolution in Diffusion MRI

Diffusion MRI (dMRI) is a widely used imaging modality, but requires long scanning times to acquire high resolution datasets. By leveraging the unique geometry present within this domain, we present a novel approach to dMRI angular super-resolution that extends upon the parametric continuous convolution (PCConv) framework. We introduce several additions to the operation including a Fourier feature mapping, global coordinates, and domain specific context. Using this framework, we build a fully parametric continuous convolution network (PCCNN) and compare against existing models. We demonstrate the PCCNN performs competitively while using significantly less parameters. Moreover, we show that this formulation generalises well to clinically relevant downstream analyses such as fixel-based analysis, and neurite orientation dispersion and density imaging.

eess.IV↗

arXiv2022

Angular Super-Resolution in Diffusion MRI with a 3D Recurrent Convolutional Autoencoder

High resolution diffusion MRI (dMRI) data is often constrained by limited scanning time in clinical settings, thus restricting the use of downstream analysis techniques that would otherwise be available. In this work we develop a 3D recurrent convolutional neural network (RCNN) capable of super-resolving dMRI volumes in the angular (q-space) domain. Our approach formulates the task of angular super-resolution as a patch-wise regression using a 3D autoencoder conditioned on target b-vectors. Within the network we use a convolutional long short term memory (ConvLSTM) cell to model the relationship between q-space samples. We compare model performance against a baseline spherical harmonic interpolation and a 1D variant of the model architecture. We show that the 3D model has the lowest error rates across different subsampling schemes and b-values. The relative performance of the 3D RCNN is greatest in the very low angular resolution domain. Code for this project is available at https://github.com/m-lyon/dMRI-RCNN.

eess.IV↗