MEDICAL AND BEHAVIOURAL KNOWLEDGE DISCOVERY USING MULTI-OBJECTIVE ANALYSIS
TEMPORAL LEARNING ON EXPERIMENTAL DATA
BEHAVIOURAL TIME SERIES
Objective ranking and clustering of behavioural data that utilize different metrics is an essential foundation for understanding the relationship between behaviour and physiology, as well as neurological processes in medicine and neuroscience. However, the inherent complexity of biological systems being studied and the small data size usually produced by experimental studies present challenges for the existing statistical methods. In this project, we aim to take advantage of the ordinal nature of behavioural metrics and propose a multi-objective analysis-based framework, inspired by techniques used in multi-objective optimization, to better represent the levels of optimalities and relative trade-offs in behavioural performances exhibited by the test subjects.
We will combine this with human-understandable data science methods, encompassing exploratory analysis, prediction on temporal experimental data, and aids for the inspection of the results.
The proposed methodology will be validated on learning behavioural performance data collected on mice. The long term aim of this project is to produce an objective, reliable and robust framework for the modeling and analysis of multi-dimensional behavioural data, hence facilitating knowledge discovery in the life sciences.
What we want to achieve
Our Project Goals
Construct dominant behavioral metrics from time-series data
We propose a series of potential behavioral metrics and compute them for all test subjects from the recorded time-series data. The dominant metrics are constructed via principal component analysis data in the space spanned by the computed metrics, while preserving the ordinality of the metrics.
Develop a clustering framework based on non-dominated sorting of available metrics
The learning behavior of mice are then ranked using non-dominated sorting and clustered according to the relative optimalities or trade-offs of their learning behaviors. The clustering is done both on their average performances across the experiments and on their progression during the course of the experiments.
Evaluate the validity of the resulting non-dominated fronts on behavioural data collected
The validity of the resulting non-dominated fronts as a basis for clustering is evaluated by two proposed indicators, which are easily calculated and not affected by linear scaling of any constituent metrics.
Transfer knowledge on salient motifs of the subjects
We plan to investigate the potential of alternative representations of behavioural data for prediction and clustering.
Produce a software tool for easy application of the proposed methodology on other data sets
We plan to package the proposed methodology into an accessible software tool, which can be easily adopted by other researchers wishing to apply it to their data.
Project Team
Noor Jamaludeen
Felix Kuhn
Prof. Dr. Sanaz Mostaghim
Prof. Dr. Stefan Remy
Qihao Shan
Prof. Dr. Myra Spiliopoulou
Publications
Medical and Behavioral Knowledge Discovery using Multi-Objective Analysis
IEEE CIBCB 2023: Conference on Computational Intelligence in Bioinformatics and Computational BiologyInferring Salient Motifs during Learning Experiments
2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS)Support our research
Become a study participant!
How do our brains, our bodies and our environment interact? How do physical illnesses affect our mental performance? And why are we more efficient on some days than others?
We would like to get to the bottom of these questions together with you. Register now and take part in exciting studies.
