Artificial intelligence and machine learning
What is AI?
Artificial intelligence or AI is best described as intelligence demonstrated by machines, where the machine mimics learning - potentially making mistakes and using that mistake to correct future behaviour or problem solving.
Machine learning is a part of artificial intelligence, using computers that can themselves improve upon the task they have been set to perform. Computers are able to handle huge amounts of data and being able to ‘learn’ from that data makes them particularly good at exploring and analysing large amounts of complex information and data.
Here are a few examples of the exciting work where birdsong could help tackle human disease, food security is being delivered through a mobile phone and big brother is bringing improvements in animal health and welfare. All made possible by artificial intelligence and machine learning.
Professor Christine Orengo
Introduction by Professor Christine Orengo, University College London
Computational biologists have long utilised statistical and machine learning methodologies to decipher patterns in complex biological data and use these patterns predictively. Applications of machine learning have far reaching potential including discovering genotype-phenotype associations, drug discovery and the mining of health records for disease associations.
The recent deluge of data brought about by new high throughput techniques (for example in protein sequencing and imaging), together with advances in computing hardware, are enabling powerful new strategies. For example, deep learning, a type of machine learning that exploits artificial neural networks, can exploit huge, high quality datasets to detect important features in the data that the biologist may not even have considered. It offers considerable promise in finding key features driving predictions and has proven to be particularly powerful for interpreting imaging data.
Dr Robert Francis Lachlan
In order to deliver accurate predictions, very high quality experimental data is needed and it is essential that biologists and computer scientists work closely together. The Council’s strategic priorities on AI and machine learning are supporting this, enabling interesting discoveries.
From tweets to human disease: the power of learning songs
Dr Robert Francis Lachlan, Queen Mary, University of London
Songbirds, such as chaffinches and great tits, share an unusual ability with humans: vocal learning. Like us, birds need to hear and imitate others in order to develop their vocal communication signals.
Research from a number of disciplines has uncovered links between human speech and bird song and bird song currently represents the best animal model we have for understanding the biology of speech. Using machine learning approaches to study bird song syllables, Dr Lachlan’s research seeks to train “machine learning” computer algorithms using birds’ perceptual judgements of song similarity and investigate how birds learn their songs.
By introducing the first biologically validated method to compare songs, this research will enable a large change to the methods used by the research field as a whole, benefitting research in fields from neuroscience and human disease to biodiversity and biomonitoring.
Professor Liangxiu Han
Food security delivered through a mobile phone
Professor Liangxiu Han, Manchester Metropolitan University
Early accurate detection and identification of crop diseases plays an important role in effectively controlling and preventing diseases for sustainable agriculture and food security.
Using advanced image processing, machine learning and cloud computing approaches, Professor Han’s team has developed an innovative automated machine vision system for efficient crop disease diagnosis from images.
Through a BBSRC GCRF Translation Award, the team will now take this technology forward, working closely with partners in China, aiming to develop a tool that can run on mobile devices and enable farmers to perform immediate potato disease diagnosis.
This machine vision system will dramatically speed up diagnosis, giving growers more accurate information on which to base their disease control strategies and stop crop yields from being reduced by infection. This technology can help make a significant impact on agricultural productivity and farmer incomes, ensuring food security, and deliver highly cost-effective, long-term economic and social impact in China.
Professor Paul Rees
UK/US partnership provides a resource to unlock secrets within cells
Professor Paul Rees, Swansea University
Research funded through the UK BBSRC-US NSF/BIO Lead Agency pilot has enabled UK researchers at Swansea University to collaborate with colleagues at the Broad Institute of Harvard and MIT, Cambridge, USA to develop software with the potential to unlock hidden information within images of cells.
The project will develop and demonstrate software to mine data from imaging flow cytometers - instruments that capture thousands of images of cells per second. The research seeks to analyse these images to precisely measure hundreds of features related to cellular morphology. This project will develop advanced machine-learning software to accomplish this, using few or indeed no fluorescent biomarkers, eliminating the need to perturb cells.
The resulting open-source software will be freely available to scientists worldwide providing a valuable resource for both applied and clinical researchers.
Big brother: bringing improvements in animal health and welfare
Professor Ilias Kyriazakis, Newcastle University
Professor Ilias Kyriazakis
Subclinical and clinical disease is the main factor responsible for pig system inefficiency and reduction in productivity and welfare. Currently disease detection is done through human observation or diagnostic surveillance, but monitoring continuously involves significant costs and effort. Through a UKTI Agri-Tech Catalyst funded by BBSRC and Innovate UK, researchers at Newcastle University have been developing and validating technology to automatically monitor performance and behaviour in groups of growing pigs.
The work exploits the fact that health and welfare challenges lead to changes in behaviour long before clinical signs arise. Individual pig and group movements are automatically captured and analysed using low cost camera installations and computer vision and machine learning techniques, thereby providing information about pig performance, behaviour and group dynamics so as to allow rapid intervention to improve health and welfare and increase farm efficiency.
The future of AI
Professor Jonathan Ashmore FRS, Director LIDo Training Programme
The London Interdisciplinary Doctoral Programme (LIDo) is a BBSRC-sponsored project supporting over 190 4-year PhD studentships across six London universities; University College London, Kings College London, Birkbeck College, Queen Mary University of London, London School of Hygiene and Tropical Medicine and the Royal Veterinary College.
Professor Jonathan Ashmore
The highly competitive places on the programme draws students with a range of biology, engineering and physical sciences backgrounds and so highly motivated to engage with applications of Artificial Intelligence (AI) and Machine Learning (ML) technologies at the start of their careers. Drawing on the wide range of computer science research at the institutes as well as established industrial collaborations with, amongst others, Google, LIDo is able to offer projects with strong AI and ML components.
As examples, well aligned with the Key Challenges framework of BBSRC, there are projects which will be using de novo protein synthesis design by combining AI and synthetic biology at University College London and Birkbeck and, at the other end of the biological scale, an AI/ML project at the Royal Veterinary College for studying dynamic locomotor behaviour to improve poultry welfare using realistic musculoskeletal models (a collaboration with DeepMind).