Relevant news regarding my research activity.

Detecting repeated cancer evolution in human tumours from multi-region sequencing data

posted Jun 29, 2017, 12:53 AM by Giulio Caravagna

My last effort to study cancer evolution is finally out as preprint (biorXiv). The problem is to infer regularities in the development of tumours (precisely, recurrent evolutionary trajectories), with clear implications for the detection of prognostic markers, and evolutionary subgroups. The problem is hard but we have observed that, since we see cancer evolution happening multiple times in several patients, we can use a cool type of Machine Learning called Transfer Learning to make better inferences. So we have developed REVOLVER (Repeated EVOLution in cancER), and applied to lung, breast and renal cancers. We found subgroups of tumours that are characterised by recurrent evolutionary trajectories, and that have prognostic value. Joint work with Ylenia, GuidoDanieleTrevor and Andrea!

Learning the structure of Bayesian Networks and submodular function maximization

posted Jun 11, 2017, 4:27 AM by Giulio Caravagna

We are happy to announce that we have finished one of the latest manuscript on model-selection for Bayesian Networks with no hidden variables. This paper shows some new connections between a well-known class of optimization problems and model-selection, and it involves concepts from information theory, empirical Bayes and non-parametric Bootstrap. 

Text Box

 Learning the structure of dependencies among multiple random variables is a problem of considerable theoretical and practical interest. In practice, score optimisation with multiple restarts provides a practical and surprisingly successful solution, yet the conditions under which this may be a well founded strategy are poorly understood. In this paper, we prove that the problem of identifying the structure of a Bayesian Network via regularised score optimisation can be recast, in expectation, as a submodular optimisation problem, thus guaranteeing optimality with high probability. This result both explains the practical success of optimisation heuristics, and suggests a way to improve on such algorithms by artificially simulating multiple data sets via a bootstrap procedure. We show on several synthetic data sets that the resulting algorithm yields better recovery performance than the state of the art, and illustrate in a real cancer genomic study how such an approach can lead to valuable practical insights.

Invited speaker: Data Science approaches to Cancer Evolution

posted May 2, 2017, 5:43 PM by Giulio Caravagna   [ updated May 2, 2017, 5:46 PM ]

I am an invited speaker at the CODATA-RDA Research Data Science Summer School organized by the International Center for Theoretical Physics (Trieste, Italy; July 2017). There, I will present theory and practice of Data Science approaches to Cancer Evolution; very soon I will provide updates!

10th Open Source SoftwareWorld Challenge 2016 -- Silver prize!

posted Nov 22, 2016, 3:19 AM by Giulio Caravagna

Our team TRONCO-PiCnIc won the Silver prize (Korea Open Source Software Association) at the “Open Source Software (OSS) World Challenge 2016”, the annual competition hosted by The Ministry of Science, ICT and Future Planning of Korea!

Talk: Mathematics and Computer Science at the University of Glasgow.

posted Nov 7, 2016, 4:05 PM by Giulio Caravagna   [ updated Nov 7, 2016, 4:06 PM ]

On Friday 11 Nov at 15-16 I will be talking at the seminars of Mathematics and Computer Science at the University of Glasgow. 

PiCnIc featured on NPR (US)

posted Sep 10, 2016, 5:00 AM by Giulio Caravagna

PiCnIc in PNAS

posted Jun 29, 2016, 2:59 AM by Giulio Caravagna

I am very happy to say that our Pipeline for Cancer Inference PiCnIc is now out in PNAS. The tool is part of our R package for Translational OncologySome media coverage is happening, and will be mirrored on this webpage.

Paper accepted at CMSB 2016 (Cambridge)

posted Jun 21, 2016, 9:03 AM by Giulio Caravagna

My paper with Luca Bortolussi and Guido Sanguinetti on matching models across abstraction levels with Gaussian Processes was accepted for presentation (with very positive reviews) at the 14th. International Conference on Computational Methods in Systems Biology to be held in Cambridge, 21-23 September 2016.

New paper on gene switching and extrinsic noise is out in Scientific Reports

posted Jun 11, 2016, 6:01 AM by Giulio Caravagna

Our paper that studies a common transcriptional network motif in various operational settings appears in Scientific Reports. The paper shows how the response to random extrinsic perturbations is regulated to the rate of gene switching among active and inactive states. The paper is a joint work with physicists from Milan/Lyon. 

Gaussian Processes, Heteroscedasticity and Statistical Emulation

posted May 7, 2016, 10:05 AM by Giulio Caravagna   [ updated May 7, 2016, 10:06 AM ]

We just finished a paper about the reconciliation of quantitative models' predictions across abstraction levels. This draws interesting connections to the area of statistical emulation, heteroscedastic regression and model reduction.

1-10 of 15