ACC (Ton) Coolen

Professor,
Physics of machine learning and complex systems
Department of Biophysics,
Radboud University 
Director, Saddle Point Science Europe BV

Email:

a.coolen@science.ru.nl

ton.coolen@saddlepointscience.com

Office address:

Department of Biophysics, Donders Institute

Huygens Gebouw, Heyendaalseweg 135

6525 AJ Nijmegen, The Netherlands

RESEARCH

Active Themes

Active Projects

Mathematical biomedicine and survival analysis, with applications to cancer research

 

Statistical mechanics of disordered and heterogeneous many-particle systems

 

Stochastic processes on complex networks or graphs, and generation of tailored random graphs

Medical statistics

Bayesian analysis of survival data in presence of latent cohort heterogeneity, beyond the visible covariates (e.g. heterogeneity in associations or base hazard rates), and with competing risks (i.e. informative censoring). Theory of overfitting and overfitting-correction protocols in time-to-event regression models, and Bayesian discriminant analysis for very high-dimensional covariates.

Mathematical immunology

Dynamics and statics of large immune system networks in the finite connectivity regime. We want to improve our understanding of the adaptive immune system, with a long-term view to aid intelligent reprogramming of this powerful but so far clinically under-utilised system.

Currently dormant themes

Mathematical analysis of natural and synthetic neural information processing systems

Non-equilibrium statistical mechanics of minority games - agent based models for financial markets

Generation of primary and secondary structure in heteropolymers

Stationary state probability distributions of nonequilibrium stochastic processes

Theory of cellular signalling processes

Cellular signalling networks (proteomic, gene regulatory, metabolic) are highly complex interacting many-variable systems. I try to adapt the mathematical tools from heterogeneous many-particle systems in theoretical physics to proteome and interactome cell biology, such as generating functional analysis

(based on path integrals).

Analysis of loopy sparse random graph ensembles

Nearly all of our mathematical methods for analysing sparse graphs (cavity methods, belief propagation, standard replica methods, etc) assume that these graphs are locally tree-like. No exact methods exist for calculating e.g. eigenvalue spectra or ensemble entropies for graphs with many short loops. I am developing a new method for this, based on a flavour of replica analysis with imaginary replica dimension.

Equilibrium and non-equilibrium analysis of processes on complex loopy graphs

I try to use the methods of the previous project to derive exact results for complex dynamical systems on finitely connected loopy graphs. The idea is to approximate any given graph by an ensemble of random graphs which share with the original graph the full degree sequence and the eigenvalue spectrum (so they have identical local connectivity and identical closed path statistics of any length). 

 
 
PUBLICATIONS

Books

Books in preparation

Principles of survival analysis (with JE Barrett and L Holmberg)

All research and review papers

ACC Coolen, R Kuehn and P Sollich 

Theory of neural information processing systems

 (Oxford University Press, July 2005)

ACC Coolen, A Annibale and  K Roberts

Generating random networks and graphs

(Oxford University Press, April 2017)

Selected recent research papers

   

ACC Coolen, J Barrett, P Paga and C Perez Vicente. J Phys A 50 (2017), 375001

Replica analysis of overfitting in regression models for time-to-event data.

M Sheikh and ACC Coolen. Journal of Classification (2017), submitted. 

Accurate Bayesian data classification without hyperparameter cross-validation.

F Aguirre Lopez, P Barucca, M Fekom and ACC Coolen. J Phys A 51 (2018), 085101

Exactly solvable random graph ensemble with extensively many short cycles.

A Grigoriadis et al. J Path: Clin Res 4 (2018), 39-54

Histological scoring of immune and stromal features in breast and axillary lymph

nodes is prognostic for distant metastasis in lymph node-positive breast cancers

A Mozeika and ACC Coolen. Phys Rev E98 (2018), 042133

Mean-field theory of Bayesian clustering

PUBLICATIONS

Books

Books in preparation

Principles of survival analysis (with M Jonker and L Holmberg)

Statistical physics of heterogeneous many-particle systems (with A Annibale)

Books planned

Introduction to the replica method

Dynamics of heterogeneous many-particle systems

        (with A Annibale and A Mozeika)

All research and review papers

Below you find links to:

A full list of my publications

PDFs of recently published papers, organised by year

PDFs of recently published papers, organised by topic

ACC Coolen, R Kuehn and P Sollich 

Theory of neural information processing systems

 (Oxford University Press, July 2005)

ACC Coolen, A Annibale and  K Roberts

Generating random networks and graphs

(Oxford University Press, April 2017)

Selected recent research papers

ACC Coolen, T Nikoletopoulos, S Arai and K Tanaka, in press (Sublinear Computation Paradigm, Springer, 2022) 

Dynamical analysis of quantum annealing

A Mozeika, M Sheikh, F Aguirre-Lopez, F Antenucci and ACC Coolen, Phys rev E 103 (2021), 042142

Exact results on high-dimensional linear regression via statistical physics

ACC Coolen, M Sheikh, A Mozeika, F Aguirre-Lopez and F Antenucci,

J Phys A 53 (2020) 365001

Replica analysis of overfitting in generalized linear regression models

F Aguirre Lopez and ACC Coolen, J Phys A 53 (2020), 065002

Imaginary replica analysis of loopy regular random graphs

 

M Sheikh and ACC Coolen. J Classification (2019), doi.org/10.1007/s00357-019-09316-6

Accurate Bayesian data classification without hyperparameter cross-validation.

Evora (Portugal), 16 Sept 2011. Controlled Markovian dynamics of graphs.

London, 21 Feb 2012, 26 March 2012. What you see is not what you get: how sampling affects macroscopic features of biological networks

Bad Honnef (DPG Physics School), Sept 2012. Counting and generating tailored random graphs

Paris, 11 Dec 2013. Solvable immune network models on finitely connected graphs with many short loops

Pisa, 8 Dec 2014. Bayesian clinical classification from high-dimensional data: signatures versus variability.

       

Boston, 20 May 2015, London, June 9 2015. New analytical tools for loopy sparse random graphs.

Kyoto, 15 December 2015, London, June 9 2015. Replica methods for loopy sparse random graphs

New York, 18 April 2016. Towards a theory of overfitting in proportional hazards regression for survival data.

SOME SELECTED RECENT SEMINARS
TEACHING

Lecture courses

Postgraduate collaborators

Emanuele Massa

Survival analysis for heterogeneous cohorts 

Jasper Hof 

Recurrent event models for bladder cancer outcome prediction from SNV data

SUPERVISION
AND

Former postgraduate collaborators

Postdoctoral collaborators

Mark Rowley

Bayesian latent class analyses of clinical data, and their computational implementations

 

Alexander Mozeika

Predictive regression and survival analysis in the high-dimensional regime

 

Fabrizio Antenucci

Bayesian predictive inference and overfitting correction for high-dimensional biomedical data

Theodore Nikoletopoulos

Predictive regression from longitudinal data.

Harm Jonker, Stephen Laughton, Charles Mace, Nikos Skantzos, Alexander Heimel, Hirak Chakravorty, Jon Hatchett, Theodore Nikoletopoulos, Bastian Wemmenhove, Nima Shayeghi, Sabrina Rabello, Alexander Mozeika, Panos Papadopoulos, Mark Rowley, Kate Roberts, James Barrett, Akram Shalabi, Mansoor Sheikh, Fabian Aguirre-Lopez

Lecture notes