Invited Talk by G. Volpe at ICTP, 8 September 2021

Neural net with input layer (left), dense internal layers, and output layer (right). (Image from the article Machine Learning for Active Matter)
Machine Learning for Active Matter: Opportunities and Challenges

Giovanni Volpe
Invited Talk
(online at) ICTP, Trieste, Italy
8 September 2021, 11:30 CEST

Machine-learning methods are starting to shape active-matter research. Which new trends will this start? Which new groundbreaking insight and applications can we expect? More fundamentally, what can this contribute to our understanding of active matter? Can this help us to identify unifying principles and systematise active matter? This presentation addresses some of these questions with some concrete examples, exploring how machine learning is steering active matter towards new directions, offering unprecedented opportunities and posing practical and fundamental challenges. I will illustrate some most successful recent applications of machine learning to active matter with a slight bias towards work done in my research group: enhancing data acquisition and analysis; providing new data-driven models; improving navigation and search strategies; offering insight into the emergent dynamics of active matter in crowded and complex environments. I will discuss the opportunities and challenges that are emerging: implementing feedback control; uncovering underlying principles to systematise active matter; understanding the behaviour, organisation and evolution of biological active matter; realising active matter with embodied intelligence. Finally, I will highlight how active matter and machine learning can work together for mutual benefit.

Links:
Conference: Statistical Physics of Complex Systems | (smr 3624)
Program of Day 1 (8 September)

The environment topography alters the transition from single-cell populations to multicellular structures in Myxococcus xanthus published in Science Advances

M. xanthus cell-cell and cell-particle local interactions during cellular aggregation.
The environment topography alters the transition from single-cell populations to multicellular structures in Myxococcus xanthus
Karla C. Hernández Ramos, Edna Rodríguez-Sánchez, Juan Antonio Arias del Angel, Alejandro V. Arzola, Mariana Benítez, Ana E. Escalante, Alessio Franci, Giovanni Volpe, Natsuko Rivera-Yoshida
Sci. Adv. 7(35), eabh2278 (2021)
bioRxiv: 10.1101/2021.01.27.428527
doi: 10.1126/sciadv.abh2278

The social soil-dwelling bacteria Myxococcus xanthus can form multicellular structures, known as fruiting bodies. Experiments in homogeneous environments have shown that this process is affected by the physico-chemical properties of the substrate, but they have largely neglected the role of complex topographies. We experimentally demonstrate that the topography alters single-cell motility and multicellular organization in M. xanthus. In topographies realized by randomly placing silica particles over agar plates, we observe that the cells’ interaction with particles drastically modifies the dynamics of cellular aggregation, leading to changes in the number, size and shape of the fruiting bodies, and even to arresting their formation in certain conditions. We further explore this type of cell-particle interaction in a minimal computational model. These results provide fundamental insights into how the environment topography influences the emergence of complex multicellular structures from single cells, which is a fundamental problem of biological, ecological and medical relevance.

The Cognitive Connectome in Healthy Aging published in Frontiers in Aging Neuroscience

Age-independent cognitive connectome in the whole cohort.
The Cognitive Connectome in Healthy Aging
Eloy Garcia-Cabello, Lissett Gonzalez-Burgos, Joana B. Pereira, Juan Andres Hernández-Cabrera, Eric Westman, Giovanni Volpe, José Barroso, & Daniel Ferreira
Front. Aging Neurosci. 13, 530 (2021)
doi: 10.3389/fnagi.2021.694254

Objectives: Cognitive aging has been extensively investigated using both univariate and multivariate analyses. Sophisticated multivariate approaches such as graph theory could potentially capture unknown complex associations between multiple cognitive variables. The aim of this study was to assess whether cognition is organized into a structure that could be called the “cognitive connectome,” and whether such connectome differs between age groups.

Methods: A total of 334 cognitively unimpaired individuals were stratified into early-middle-age (37–50 years, n = 110), late-middle-age (51–64 years, n = 106), and elderly (65–78 years, n = 118) groups. We built cognitive networks from 47 cognitive variables for each age group using graph theory and compared the groups using different global and nodal graph measures.

Results: We identified a cognitive connectome characterized by five modules: verbal memory, visual memory—visuospatial abilities, procedural memory, executive—premotor functions, and processing speed. The elderly group showed reduced transitivity and average strength as well as increased global efficiency compared with the early-middle-age group. The late-middle-age group showed reduced global and local efficiency and modularity compared with the early-middle-age group. Nodal analyses showed the important role of executive functions and processing speed in explaining the differences between age groups.

Conclusions: We identified a cognitive connectome that is rather stable during aging in cognitively healthy individuals, with the observed differences highlighting the important role of executive functions and processing speed. We translated the connectome concept from the neuroimaging field to cognitive data, demonstrating its potential to advance our understanding of the complexity of cognitive aging.

Enhanced prediction of atrial fibrillation and mortality among patients with congenital heart disease using nationwide register-based medical hospital data and neural networks published in European Heart Journal – Digital Health

Neural network prediction of mortality and atrial fibrillation. (Image taken from the article’s graphical abstract.)
Enhanced prediction of atrial fibrillation and mortality among patients with congenital heart disease using nationwide register based medical hospital data and neural networks
Kok Wai Giang, Saga Helgadottir, Mikael Dellborg, Giovanni Volpe, Zacharias Mandalenakis
European Heart Journal – Digital Health (2021)
doi: 10.1093/ehjdh/ztab065

Aims: To improve short-and long-term predictions of mortality and atrial fibrillation (AF) among patients with congenital heart disease (CHD) from a nationwide population using neural networks (NN).

Methods and results: The Swedish National Patient Register and the Cause of Death Register were used to identify all patients with CHD born from 1970 to 2017. A total of 71 941 CHD patients were identified and followed-up from birth until the event or end of study in 2017. Based on data from a nationwide population, a NN model was obtained to predict mortality and AF. Logistic regression (LR) based on the same data was used as a baseline comparison. Of 71 941 CHD patients, a total of 5768 died (8.02%) and 995 (1.38%) developed AF over time with a mean follow-up time of 16.47 years (standard deviation 12.73 years). The performance of NN models in predicting the mortality and AF was higher than the performance of LR regardless of the complexity of the disease, with an average area under the receiver operating characteristic of >0.80 and >0.70, respectively. The largest differences were observed in mortality and complexity of CHD over time.

Conclusion: We found that NN can be used to predict mortality and AF on a nationwide scale using data that are easily obtainable by clinicians. In addition, NN showed a high performance overall and, in most cases, with better performance for prediction as compared with more traditional regression methods.

Extracting quantitative biological information from brightfield cell images using deep learning featured in AIP SciLight

The article Extracting quantitative biological information from brightfield cell images using deep learning
has been featured in: “Staining Cells Virtually Offers Alterative Approach to Chemical Dyes”, AIP SciLight (July 23, 2021).

Scilight showcases the most interesting research across the physical sciences published in AIP Publishing Journals.

Scilight is published weekly (52 issues per year) by AIP Publishing.

Microscopic Metavehicles Powered and Steered by Embedded Optical Metasurfaces published in Nature Nanotechnology

Metavehicles.
Microscopic Metavehicles Powered and Steered by Embedded Optical Metasurfaces
Daniel Andrén, Denis G. Baranov, Steven Jones, Giovanni Volpe, Ruggero Verre, Mikael Käll
Nat. Nanotechnol. (2021)
doi: 10.1038/s41565-021-00941-0
arXiv: 2012.10205

Nanostructured dielectric metasurfaces offer unprecedented opportunities to manipulate light by imprinting an arbitrary phase gradient on an impinging wavefront. This has resulted in the realization of a range of flat analogues to classical optical components, such as lenses, waveplates and axicons. However, the change in linear and angular optical momentum associated with phase manipulation also results in previously unexploited forces and torques that act on the metasurface itself. Here we show that these optomechanical effects can be utilized to construct optical metavehicles – microscopic particles that can travel long distances under low-intensity plane-wave illumination while being steered by the polarization of the incident light. We demonstrate movement in complex patterns, self-correcting motion and an application as transport vehicles for microscopic cargoes, which include unicellular organisms. The abundance of possible optical metasurfaces attests to the prospect of developing a wide variety of metavehicles with specialized functional behaviours.

Extracting quantitative biological information from brightfield cell images using deep learning published in Biophysics Reviews

Virtually-stained generated image for lipid-droplet.
Extracting quantitative biological information from brightfield cell images using deep learning
Saga Helgadottir, Benjamin Midtvedt, Jesús Pineda, Alan Sabirsh, Caroline B. Adiels, Stefano Romeo, Daniel Midtvedt, Giovanni Volpe
Biophysics Rev. 2, 031401 (2021)
arXiv: 2012.12986
doi: 10.1063/5.0044782

Quantitative analysis of cell structures is essential for biomedical and pharmaceutical research. The standard imaging approach relies on fluorescence microscopy, where cell structures of interest are labeled by chemical staining techniques. However, these techniques are often invasive and sometimes even toxic to the cells, in addition to being time-consuming, labor-intensive, and expensive. Here, we introduce an alternative deep-learning-powered approach based on the analysis of brightfield images by a conditional generative adversarial neural network (cGAN). We show that this approach can extract information from the brightfield images to generate virtually-stained images, which can be used in subsequent downstream quantitative analyses of cell structures. Specifically, we train a cGAN to virtually stain lipid droplets, cytoplasm, and nuclei using brightfield images of human stem-cell-derived fat cells (adipocytes), which are of particular interest for nanomedicine and vaccine development. Subsequently, we use these virtually-stained images to extract quantitative measures about these cell structures. Generating virtually-stained fluorescence images is less invasive, less expensive, and more reproducible than standard chemical staining; furthermore, it frees up the fluorescence microscopy channels for other analytical probes, thus increasing the amount of information that can be extracted from each cell.

Neural Network Training with Highly Incomplete Datasets on ArXiv

Working principles for training neural networks with highly incomplete dataset: vanilla (upper panel) vs GapNet (lower panel) (Image by Yu-Wei Chang.)
Neural Network Training with Highly Incomplete Datasets
Yu-Wei Chang, Laura Natali, Oveis Jamialahmadi, Stefano Romeo, Joana B. Pereira, Giovanni Volpe
arXiV: 2107.00429

Neural network training and validation rely on the availability of large high-quality datasets. However, in many cases only incomplete datasets are available, particularly in health care applications, where each patient typically undergoes different clinical procedures or can drop out of a study. Since the data to train the neural networks need to be complete, most studies discard the incomplete datapoints, which reduces the size of the training data, or impute the missing features, which can lead to artefacts. Alas, both approaches are inadequate when a large portion of the data is missing. Here, we introduce GapNet, an alternative deep-learning training approach that can use highly incomplete datasets. First, the dataset is split into subsets of samples containing all values for a certain cluster of features. Then, these subsets are used to train individual neural networks. Finally, this ensemble of neural networks is combined into a single neural network whose training is fine-tuned using all complete datapoints. Using two highly incomplete real-world medical datasets, we show that GapNet improves the identification of patients with underlying Alzheimer’s disease pathology and of patients at risk of hospitalization due to Covid-19. By distilling the information available in incomplete datasets without having to reduce their size or to impute missing values, GapNet will permit to extract valuable information from a wide range of datasets, benefiting diverse fields from medicine to engineering.

Raman Tweezers for Tire and Road Wear Micro- and Nanoparticles Analysis on ChemRxiv

Optical beam focused into the liquid: the tire particles are pushed away from the laser focus.
Raman Tweezers for Tire and Road Wear Micro- and Nanoparticles Analysis
Pietro Giuseppe Gucciardi, Gillibert Raymond, Alessandro Magazzù, Agnese Callegari, David Brente Ciriza, Foti Antonino, Maria Grazia Donato, Onofrio M. Maragò, Giovanni Volpe, Marc Lamy de La Chapelle & Fabienne Lagarde
ChemRxiv: https://doi.org/10.33774/chemrxiv-2021-h59n1

Tire and Road Wear Particles (TRWP) are non-exhaust particulate matter generated by road transport means during the mechanical abrasion of tires, brakes and roads. TRWP accumulate on the roadsides and are transported into the aquatic ecosystem during stormwater runoffs. Due to their size (sub-millimetric) and rubber content (elastomers), TRWP are considered microplastics (MPs). While the amount of the MPs polluting the water ecosystem with sizes from ~ 5 μm to more than 100 μm is known, the fraction of smaller particles is unknown due to the technological gap in the detection and analysis of < 5 μm MPs. Here we show that Raman Tweezers, a combination of optical tweezers and Raman spectroscopy, can be used to trap and chemically analyze individual TWRPs in a liquid environment, down to the sub-micrometric scale. Using tire particles mechanically grinded from aged car tires in water solutions, we show that it is possible to optically trap individual sub-micron particles, in a so-called 2D trapping configuration, and acquire their Raman spectrum in few tens of seconds. The analysis is then extended to samples collected from a brake test platform, where we highlight the presence of sub-micrometric agglomerates of rubber and brake debris, thanks to the presence of additional spectral features other than carbon. Our results show the potential of Raman Tweezers in environmental pollution analysis and highlight the formation of nanosized TRWP during wear.

Colloquium by G. Volpe at TU-Darmstadt, 18 June 2021, Online

Deep learning for particle tracking. (Image by Aykut Argun)
Deep learning for microscopy, optical trapping, and active matter

Giovanni Volpe
Colloquium
(online at) TU-Darmstadt, Germany
18 June 2021, 14:00 CEST

After a brief overview of artificial intelligence, machine learning and deep learning, I will present a series of recent works in which we have employed deep learning for applications in photonics and active matter.
In particular, I will explain how we employed deep learning to enhance digital video microscopy, to estimate the properties of anomalous diffusion, to characterize microscopic force fields, to improve the calculation of optical forces, and to characterize nanoparticles.
Finally, I will provide an outlook for the application of deep learning in photonics and active matter.