Validation of a translatable chronobiological signature of early relapse in bipolar disorder (active)
The aim of this multi-national project is to provide a quantum advance in understanding the mechanisms of sleep and circadian rhythm disruption amongst people with established bipolar disorder (BD). Our methodological focus is a high-resolution signal of specific relevance to BD – the 24-hour rest-activity rhythm as measured by actigraphy. Across four work packages, distinct sleep and circadian features from this signal will be parsed through a machine learning approach called network analysis, and validated as a predictor of early relapse amongst inter-episode patients (Study 1 Australia), as a covariate of recovery from acute manic and depressive illness (Study 2 New Zealand), and as a proxy of endogenous circadian pathogenesis of BD (Study 3 India). In the integrative Work Package 4, findings from these complementary investigations will be cross-validated and synthesised into a theoretically and empirically grounded chronobiological signature of early relapse in BD. This biosignature could be the basis for a future automated early warning technology for BD (our long-term goal). Work Package 4 will also generate a new multi-national dataset and data processing pipelines to be shared with future researchers. Our multi-disciplinary team is uniquely qualified to undertake this project in collaboration with our long-standing lived experience collaborators.
Gradients of Reciprocity in Biological and Artificial Neural Networks (active)
We introduce efficient Network Reciprocity Control (NRC) algorithms for steering the degree of asymmetry and reciprocity in binary and weighted networks while preserving fundamental network properties. Our methods maintain edge density in binary networks and cumulative edge weight in weighted graphs. We test these algorithms on synthetic benchmark networks—including random, small-world, and modular structures— as well as brain connectivity maps (connectomes) from various species. We demonstrate how adjusting the asymmetry-reciprocity balance under edge density and total weight constraints influences key network features, including spectral properties, degree distributions, community structure, clustering, and path lengths.
Cross-modal Learning-A02
Linking biological and artificial neural networks, I develop structured and heterogenous reservoir computing models.
DFG- SFB 1328-A02
Ca2+ microdomain formation within the first seconds of T cell receptor activation determines downstream T cell signaling, but detailed mechanisms and involved players still remain to be identified. Novel image analysis approaches allow a deeper insight into the spatio-temporal Ca2+ microdomain architecture. In this project an unsupervised conventional cell segmentation technique is compared with two supervised machine learning solutions. A Reservoir Computing (RC) scheme is indeed adopted for cell segmentation. Besides, we proposed a reservoir computing (RC) framework for efficient and temporally consistent segmentation. Experiments were conducted with Jurkat T-cells and anti-CD3 coated beads used for T-cell activation. We compared the RC performance with a standard U-Net and a convolutional long short-term memory (LSTM) model.
NeuRAM3
The objective of this EU project is to fabricate a chip implementing a neuromorphic architecture that supports state-of-the-art machine learning and spike-based learning mechanisms. In the framework of this project, I'm working on implementing spiking reservoir computing on Dynapse chip to achieve a biomedical signal processing task.
PhD thesis (Systemic Modelling of Mood Variations in Bipolar Disorder)
In the absence of a coherent neural model of its aetiology, mathematical representation is a critical tool for understanding the complex temporal phenomenology of bipolar disorder. In my PhD thesis, manic and depressive episodes of bipolar disorder were proposed to be mathematically representable as two “strange sub-attractors” of a nonlinear discrete dynamical system through which the mood trajectory moves. We developed a map based chaotic system, which demonstrates intermittent recurrent episodes of the illness via changes in its parameters. The core of my phenomenological model was the notion of competition between recurrent maps, which mathematically represent the dynamics of activation in excitatory (Glutamatergic) and inhibitory (GABAergic) pathways in the brain.
Master thesis (Identification of Diabetic Retinopathy Stage Using Retinal Image Processing)
The process and knowledge of digital image processing was applied to develop a software which automatically identifies different stages of diabetic retinopathy and risk of diabetic macular edema from fundus images.
Thesis advisees:
1- Reservoir Computing Analysis of Information Transmission and Fusion in Laminar-Specific Cortical Microcircuits
2- Exploring potential ADHD biomarkers through advanced machine learning: An examination of audiovisual integration networks
3- Beyond Feature Attribution: Quantifying Unit Contributions using Multidimensional Shapley Value Analysis
4- fMRI analysis on emotion regulation to investigate the causal connectivities between cortical and sub-cortical regions of the brain in response to emotional images.
5- Including the effect of ephaptic coupling in a reaction- diffusion model of excitable tissue using the solution of Poisson’s equation to investigate the role of ephaptic coupling in dynamics of wave propagation in a syncytium.
6- Quantifying the changes in recurrence plot of actigraphic recordings in bipolar patients in comparison with the normal subjects.
7- Proposing a network of coupled excitable cells to investigate the effects of demyelination of each neuronal population on the dynamical response of its neighbors.
8- developing a simple mathematical model of interaction between psychological parameters in bipolar disorder based on the idea of prey-predator relationship.