Our research spans from fundamental nonlinear dynamics and noise-induced phenomena to applied systems medicine, AI-based diagnostics, and computational neuroscience. Below we summarise the main research directions, organised by topic.
Early work on nonlinear acoustics, including sound wave interaction with obstacles, jet self-oscillation theory with acoustic feedback, and shock wave interaction with resonant absorbers (1, 2, 3, 4).
Discovery of several fundamental noise-induced phenomena: Doubly Stochastic Resonance (DSR), where multiplicative noise creates bistability and additive noise then optimises signal transmission (22, 23); Doubly Stochastic Coherence in neural models (37); System Size Resonance, where finite system size controls ordering (29, 35); noise-induced propagation in monostable media (32); and vibrational resonance in noise-induced structures (28, 42, 43).
Stochastic modelling of childhood epidemics (8), symbolic dynamics and FHN-based models for bimanual movements (9, 17), analysis of microseism oscillations (38), and bistability in rolling friction on viscoelastic surfaces (47).
Introduction of structural complexity measures (SCI, SDI, IGE, TNI) for quantifying trabecular bone architecture from CT images (6, 27, 49), development of bone resorption simulation algorithms (48), and a conceptual model of stochastic resonance in bone remodelling under microgravity (66).
Discovery of noise-induced excitability and wave propagation in excitable media (36), noise-memory-induced excitability (52), frequency-selective stochastic resonance (53). Rich dynamics in coupled synthetic genetic oscillators: multistability, clustering, oscillation death in repressilator networks (59, 62, 63). Speed-dependent cellular decision making (SdCDM) in gene regulatory networks (78, 80, 83, 124). Synchronization in complex networks with degree correlations and multiplex topologies (109, 118, 119, 189).
Development of the Brownian ratchet model for peptide transport in the proteasome (51), stochastic and analytical models for peptide length distributions (56, 57), the ProteaMAlg kinetic model for proteasomal cleavage (67), and SpliceMet -- the first computational method for predicting proteasome-catalysed peptide splicing (73).
Serial multi-marker longitudinal algorithms for ovarian cancer screening using UKCTOCS data, including Bayesian change-point and LSTM-based approaches (111, 122, 175). Proteomics-derived multimarker panels for ovarian cancer (132, 136). CA19-9 upregulation before pancreatic cancer diagnosis and PancRISK urine biomarker score (92, 131, 177). Sex steroid receptor bioactivity markers for breast cancer in BRCA carriers (79, 86). A 14-protein plasma signature for lung cancer prevention from UK Biobank data (200).
Prediction models for obstetric outcomes: preterm birth, fetal growth restriction, and preeclampsia using ultrasound, biochemical markers, and machine learning (164, 174, 187, 196, 202). A mathematical model of inflammageing incorporating bone marrow dynamics (125). TREC-based screening methods for primary immunodeficiency diseases (133). COVID-19 and post-COVID condition prediction models (146, 152, 186, 201). AI-based prediction of paediatric respiratory viral infections incorporating climate data (185). Biomedical statistics tutorial series (141, 150, 157, 170).
Introduction of per-gene methylation measures and meta-analysis across 14 cancer types (84). DNAm network correlation measures and prognostic subnetwork modules (89). Network community oncomarker detection pipeline (108). 93-100% cancer classification accuracy across 13 types using machine learning on methylation features (101, 104). Parenclitic networks applied to Down syndrome methylation (140). Demonstration that 90% of age-dependent DNA methylation variability is deterministic (145). Development of epicMI R package for Illumina methylation array reliability (179).
Time-series radiomics with parallel LSTM architecture for predicting prostate cancer progression on active surveillance (AUC 0.86) (162). PSA density superiority over PSA for dynamic monitoring (166). Explainable AI framework with SHAP-guided feature selection for progression prediction (ensemble AUC 0.947) (192). Multicentre validation of active surveillance safety in the MRI era, with cribriform GG2 as key risk factor (188). Cribriform morphology as a long-term prognostic marker from the 15-year ProtecT trial (199).
The genetic perceptron concept: gene regulatory circuits performing linear classification and learning by selective plasmid loss (97, 98, 173). Emergent digital bio-computation in bacterial populations via spatial diffusion (110, 176). Spiking neuron-astrocyte network (SNAN) models for working memory and associative memory (129, 144, 153, 171). Discovery of noise-induced artificial intelligence through astrocytic modulation (159). Integrated information (Phi) in neuron-astrocyte and genetic networks (127, 128, 138, 194). Modelling brain ageing, glymphatic clearance, and Alzheimer's disease (134, 137, 172, 190). Reservoir computing methods for time series classification (178). A thermodynamic model of the sleep cycle driven by brain temperature dynamics (181).
Systematic benchmarking of parenclitic and synolitic network approaches, demonstrating synolitic superiority for high-dimensional data (147, 148). Parenclitic networks applied to DNA methylation achieving 90-99% cancer classification from 12 topological features (114). 2DKDE parenclitic networks for early ovarian cancer detection (121). Proteomics-based parenclitic models for COVID-19 treatment prediction (AUROC 0.99) (146, 152). Synolitic Graph Neural Networks (SGNNs) for brain imaging classification (98.5-100% accuracy) and ovarian cancer detection (167, 168, 180, 193, 195). Demonstration of SGNN robustness to the curse of dimensionality across 15 UCI benchmark datasets (195).
We provide methodological support for medical groups and develop new statistical methods, especially in: