Improving detection of gene expression in microscopy, using Artificial Intelligence techniques in order to investigate how food affects ageing in C El

Lead Research Organisation: King's College London
Department Name: Developmental Neurobiology


This project investigates how food affects ageing via conserved neuroendocrine factors. In C. elegans and humans, neuroendocrine factors such as insulin-like peptides, growth factors, and biogenic amines regulate each other in complex networks to modulate ageing, metabolism, and other physiological outputs. However, the information processing mechanisms in these networks for discriminating food inputs are unclear. We address this question by enhancing high-throughput C. elegans experiments with artificial intelligence and machine learning.

These studies have two key limitations: first, the neurons are identified by hand, which is very labour intensive; and second, the need for highly standardised images means many images are discarded. The first issue offers an opportunity to automate cell identification, either by applying supervised machine learning (e.g. random forest classifier) to tens of thousands of annotated image stacks; or by using deep learning so that the algorithm can select the parameters to optimise. The second issue means that there are many unannotated image stacks that can be analysed by unsupervised deep learning and cross-validated against the annotated subset to check performance.

We will then apply decoding analysis and machine learning to relate food inputs, combinatorial patterns of gene expression, and ageing phenotypes. Using these results, we will predict the lifespans of novel mutants based on their effects on gene expression, and verify these hypotheses experimentally by exploiting high-throughput imaging and the automated image analysis developed above. This work thus decodes food-sensing gene networks in the nervous system that impact ageing and other health-related outputs.


10 25 50
Description We showed that it is possible for a neural network to learn the pose and structure of a protein complex within a certain size range, from a series of 2D images obtained using fluorescence microscopy. Furthermore, we showed that our approach is general but limited to certain imaging modalities and scales. We showed that the existing method is not suitable for analysing C. elegans neurons in it's current state, but suggested potential avenues for future research in this area.
Exploitation Route We anticipate that the work could be adapted to aid the field of Cryo-Electron-Microscopy and the reconstruction of biological structures smaller than these studied in this work.
Sectors Digital/Communication/Information Technologies (including Software),Pharmaceuticals and Medical Biotechnology

Title Hypothesised Objects from Light Localisations (HOLLy) 
Description HOLLy is a deep-learning neural network that attempts to learn structure and pose from images obtained using fluorescence microscopy. Given a number of images (either experimental or simulated) HOLLy will converge on a solution - a 3D structure of the object within these 2D images, becoming sensitive to the pose of the structure within these images. 
Type Of Technology Webtool/Application 
Year Produced 2023 
Open Source License? Yes  
Impact Unknown