This document reports Data Science and Artificial Intelligence initiatives oriented towards Life Sciences applications that are in place at the Trieste node for the national AIIS Laboratory. involves more than 10 members that apply Data researchers, Post-Docs and PhD students, and are affiliated to the following departments: Ital-IA 2023: 3rd National Conference on Artificial Intelligence,
Science. The current members rank as Full •
(https://dia.units.it/it);
Medical Sciences (https://dsm.units.it/).
In this document we describe relevant ongoing research projects, as well as teaching in the areas of
led and
The Cancer Data Science (CDS) Laboratory is consists of computational scientists with different expertise backgrounds, spanning from computer science, physics, genetics and biology.
2023 Copyright for this paper by its authors. Use permitted under Creative and
response (www.caravagnalab.org). to therapy The
CDS
Among our (https://github.com/caravagnalab) contributions there and are model-based technologies i) to measure clonal evolution from high-throughput sequencing data 1,2, ii) to detect repeated evolutionary trajectories with prognostic power 3, and iii) to determine the temporal ordering of somatic mutations from cross-sectional cancer genomics assays 4–6.
Recently, the laboratory has been working on the definition of AI-based models that support the application of whole-genome sequencing technologies at scale 7, as well single-cell RNA sequencing 8.
Other projects are ongoing regarding the application of AI to integrate multiomics data and, more in general, spatio-temporal measurements.
The CDS Laboratory has direct collaborations with several research institutes that work on AIrelated projects, including Human Technopole (IT), the Institute of Cancer Research (UK) the Barts Cancer Institute (UK), University College London (UK), Milan-Bicocca (IT) and Memorial Sloan Kettering (US). The Laboratory also collaborates with several clinical institutes in Italy and abroad: the Royal Marsden Hospital (UK), the Hospital San Raffaele (IT), CRO Aviano (IT) and San Gerardo Monza (IT). 2.2
The Artificial Intelligence and Cyber Physical Systems Lab (AI-CPS) is led by Prof. Luca Bortolussi, and develops novel approaches based on artificial intelligence and machine learning with applications in medicine, in collaboration with the Biostatistics Unit of the Department of Medical Science of the University of Trieste, led by prof. Giulia Barbati. The lab is particularly interested in explainable AI-based techniques for the analysis of biomedical signals of pulmonary ventilation in intensive care and ECG.
The activity of the lab in the area of biomedical signals of assisted ventilatory respiration started several years ago in collaboration with the Intensive Care Unit of the Trieste University Hospital, and has led to an international patent for the detection of asynchronies between machine and patient efforts.
The lab is also collaborating with the Biostatistics Unit of the Department of Medical Sciences for the application of deep learning to prediction tasks involving ECG signals. Within the same collaboration, the lab has also a growing interest in solutions for high quality synthetic medical data generation based on deep learning generative models. Moreover, there are a series of ongoing projects related to extracting useful epidemiological knowledge by means of machine learning techniques exploiting the informational content of the regional health administrative data.
We describe relevant research projects and collaboration with other academic or industrial partners. 3.1
Through an AIRC funded My First AIRC grant (PI Giulio Caravagna) the CDS Laboratory uses AI-based statistical models to study clonal evolution and response to therapy in different types of leukemia.
This collaborative project involves the participation of the Centre of Omics Sciences of Hospital San Raffaele (Dr Giovanni Tonon), which provides sequencing expertise and wet-lab support, and two clinical units based at IRCCS hospitals to provide leukemia samples: the unit of Experimental Onco-hematology at CRO Aviano (Dr Valter Gattei), and the unit of Immunogenetics, Leukemia Genomics and Immunobiology of Hospital San Raffaele (Dr Luca Vago).
The project seeks to develop AI technologies that can better elucidate disease dynamics and relapse mechanisms in both Acute Myeloid Leukemia and Chronic Lymphocytic Leukemia. 3.2
The CDS Laboratory collaborates synergistically with Genomics England (https://www.genomicsengland.co.uk/), the NHSowned company that delivers whole-genome sequencing (WGS) to the clinic in the UK.
Genomics England implements, through a collaboration with Illumina (https://www.illumina.com/), large-scale UKwide genomics projects involving both patients with genetic diseases, and cancer. Giulio Caravagna from the CDS Laboratory is involved in specialised data analysis groups that study WGS data collected for colorectal, endometrial, glioblastoma and haematological cancers. 3.3
The AI-CPS lab has been active for several years in the analysis of respiratory signals (pressure, flow, volume) returned by pulmonary ventilators, particularly when they are used in assisted respiration mode 9. In this case, the ventilator responds to a trigger of the patient asking for air by a positive pressure that helps the patient to breathe. Sometimes, the trigger mechanism can be misinterpreted by the controller of the machine, resulting in a so-called asynchrony. While a single asynchrony is of limited concern, several repeated episodes can lead to long term damages of the respiratory ability. However, detection of asynchronies is an error prone, manual activity. We developed a method, patented, that combines geometrical methods to extract features from the signal pipelined with machine learning which has a very high specificity and sensitivity 10. We are currently working on extensions of this method and application of AI approaches for the detection of other issues related to patients under assisted and forced ventilation, also using multi-modal learning strategies, combining different kinds of signals. 3.4
In collaboration with the Biostatistics unit of
the Department of Medical Sciences and with the
cardiology medical unit, we are exploring the use
of deep learning approaches for the analysis of
ECG signals in order to predict the onset of
cardiovascular diseases. In particular, we
investigated the performances of two ML
approaches based on ECGs for the prediction of
new-onset atrial fibrillation (AF), in terms of
discrimination, calibration and sample size
dependence [
Within the survival analysis framework, we
implemented a deep-learning-based prognostic
model for incident heart failure (HF) in patients
with diabetes using electronic health records
(EHR) that takes into account a large and
heterogeneous set of clinical factors [
Dr Alberto d’Onofrio, leader of the newly formed Computer Science for Complex Systems Laboratory, works at the interface between AI and the modelling of biological systems, with particular focus on Computational Epidemiology of Infectious Diseases and related problems in Global Public Health. He is currently co-leading the team writing a report for WHO on this and related matters. 3.6
Dr Alejandro Rodriguez-Garcia is currently working on the application of AI to the understanding of the biological availability of radioactive markers. In concrete, Unsupervised Machine Learning techniques are applied to analyze molecular simulations of some markers and obtain a deeper understanding that could lead to the proposal of new active substances with better biological properties. 3.7
Dr Fabio Anselmi is currently working at the intersection between computational neuroscience and machine learning.
In particular he develops machine learning models of the visual cortex that embody biological constraints and priors on how humans understand visual scenes such as the concept of whole-and parts in object recognition
He also develops models that incorporate symmetries of the visual signal statistics into the processing pipeline such as transformations that preserve object classification with the aim to have more faithful models of how the visual cortex works
We offer initiatives at the crossings of AI and Machine Learning applied to Life Sciences at several undergraduate and postgraduate levels. Most specialized teaching is concentrated in the Data Science and Scientific Computing (https://dssc.units.it/) Master program and in the Applied Data Science and AI doctoral program (http://adsai.units.it).
Relevant courses to the health area are: - Genome Data Analytics (6 CFU), taught by Prof. Giulio Caravagna, covering topics on the application of Machine Learning to genome-sequencing assays, with a strong focus on cancer data analysis. - Health Data Analytics (6 CFU), taught by Prof. Giulia Barbati, covering topics at the crossings between biostatistics and Machine Learning, with application to medical tabular data and survival analysis.
[1] G. Caravagna, G. Sanguinetti, T.A. Graham, A.
Sottoriva. The MOBSTER R package for tumour subclonal deconvolution from bulk DNA whole-genome sequencing data. BMC Bioinformatics. 2020;21(1):531..