Search

The protein design problem involves finding polypeptide sequences folding into a given threedimensional structure. Its rigorous algorithmic solution is computationally demanding, involving a nested search in sequence and structure spaces. Structure searches can now be bypassed thanks to recent machine learning breakthroughs, which have enabled accurate and rapid structure predictions. Similarly, sequence searches might be entirely transformed by the advent of quantum annealing machines and by the required new encodings of the search problem, which could be performative even on classical machines. In this work, we introduce a general protein design scheme where algorithmic and technological advancements in machine learning and quantum-inspired algorithms can be integrated, and an optimal physics-based scoring function is iteratively learned. In this first proof-of-concept application, we apply the iterative method to a lattice protein model amenable to exhaustive benchmarks, finding that it can rapidly learn a physics-based scoring function and achieve promising design performances. Strikingly, our quantum-inspired reformulation outperforms conventional sequence optimization even when adopted on classical machines. The scheme is general and can be easily extended, e.g., to encompass off-lattice models, and it can integrate progress on various computational platforms, thus representing a new paradigm approach for protein design., Comment: main text: 11 pages, 5 figures supplementary material: 5 pages, 3 figures

The purpose of this clinical study is to test broad spectral range (635-1060 nm) time domain diffuse optical spectroscopy in monitoring the response of breast cancer patients to neoadjuvant chemotherapy (NAC). The broadband operation allows us to fully analyze tissue composition in terms of hemoglobin, water, lipids and collagen concentration, which has never been systematically studied until now during the course of therapy. Patients are subjected to multiple breast optical imaging sessions, each one performed at different stages of NAC, both on tumor-bearing and contralateral healthy breast. We correlate the optical results with conventional imaging techniques and pathological response. Preliminary outcomes on data of 10 patients show an average significant reduction in the concentrations of oxy-hemoglobin (-53%, p = 0.0020), collagen (-36%, p = 0.0039) and water (-15%, p = 0.0195), and increase in lipids (+39%, p = 0.0137) from baseline to the end of therapy in the tumor-bearing breast of patients who responded to therapy at least partially. With respect to scattering, the scattering amplitude, a, increases slightly by the end of the therapy compared to the baseline, while the scattering slope, b, shows no significant change. Some change in concentrations of constituents was noticed also in the contralateral healthy breast, even though it was significant only for oxy-hemoglobin concentration. Collagen seems to be the only component distinguishing between complete (p = 0.0423) and partial responders (p = 0.1012) by the end of 2-3 weeks after the start of therapy, suggesting that it could serve as a potential biomarker to measure NAC effectiveness early during treatment. These initial results are consistent with those of standard medical modalities and highlight the sensitivity of the technique to changes that occur in breast composition during NAC.

In this paper, we present results on improving out-of-domain weather prediction and uncertainty estimation as part of the \texttt{Shifts Challenge on Robustness and Uncertainty under Real-World Distributional Shift} challenge. We find that by leveraging a mixture of experts in conjunction with an advanced data augmentation technique borrowed from the computer vision domain, in conjunction with robust \textit{post-hoc} calibration of predictive uncertainties, we can potentially achieve more accurate and better-calibrated results with deep neural networks than with boosted tree models for tabular data. We quantify our predictions using several metrics and propose several future lines of inquiry and experimentation to boost performance., Comment: Presented at the Bayesian Deep Learning workshop at NeurIPS 2021

Recent developments in Machine Learning approaches for modelling physical systems have begun to mirror the past development of numerical methods in the computational sciences. In this survey, we begin by providing an example of this with the parallels between the development trajectories of graph neural network acceleration for physical simulations and particle-based approaches. We then give an overview of simulation approaches, which have not yet found their way into state-of-the-art Machine Learning methods and hold the potential to make Machine Learning approaches more accurate and more efficient. We conclude by presenting an outlook on the potential of these approaches for making Machine Learning models for science more efficient., Comment: 2nd AI4Science Workshop at the 39th International Conference on Machine Learning (ICML), 2022

Nondestructive testing (NDT) is widely applied to defect identification of turbine components during manufacturing and operation. Operational efficiency is key for gas turbine OEM (Original Equipment Manufacturers). Automating the inspection process as much as possible, while minimizing the uncertainties involved, is thus crucial. We propose a model based on RetinaNet to identify drilling defects in X-ray images of turbine blades. The application is challenging due to the large image resolutions in which defects are very small and hardly captured by the commonly used anchor sizes, and also due to the small size of the available dataset. As a matter of fact, all these issues are pretty common in the application of Deep Learning-based object detection models to industrial defect data. We overcome such issues using open source models, splitting the input images into tiles and scaling them up, applying heavy data augmentation, and optimizing the anchor size and aspect ratios with a differential evolution solver. We validate the model with $3$-fold cross-validation, showing a very high accuracy in identifying images with defects. We also define a set of best practices which can help other practitioners overcome similar challenges., Comment: Workshop on machine learning for engineering modeling, simulation and design @ NeurIPS 2020

Entanglement is assuming a central role in modern quantum many-body physics. Yet, for lattice gauge theories its certification remains extremely challenging. A key difficulty stems from the local gauge constraints underlying the gauge theory, which separate the full Hilbert space into a direct sum of subspaces characterized by different superselection rules. In this work, we develop the theoretical framework of entanglement witnessing for lattice gauge theories that takes this subtlety into account. We illustrate the concept at the example of a $\mathrm{U}(1)$ lattice gauge theory in 2+1 dimensions, without and with dynamical fermionic matter. As this framework circumvents costly state tomography, it opens the door to resource-efficient certification of entanglement in theoretical studies as well as in laboratory quantum simulations of gauge theories., Comment: 17+8 pages (single column), 4+2 figures; added references

Purpose: To describe the dosimetric commissioning and quality assurance (QA) of the actively scanned proton and carbon ion beams at the Italian National Center for Oncological Hadrontherapy. Methods: The laterally integrated depth-dose-distributions (IDDs) were acquired with the PTW Peakfinder, a variable depth water column, equipped with two Bragg peak ionization chambers. FLUKA MC code was used to generate the energy libraries, the IDDs in water, and the fragment spectra for carbon beams. EBT3 films were used for spot size measurements, beam position over the scan field, and homogeneity in 2D-fields. Beam monitor calibration was performed in terms of number of particles per MU using both a Farmer-type and an Advanced Markus ionization chamber. The beam position at the isocenter, beam monitor calibration curve, dose constancy in the center of the spread-out-Bragg-peak, dose homogeneity in 2D-fields, beam energy, spot size, and spot position over the scan field are all checked on a daily basis for both protons and carbon ions and on all beam lines. Results: The simulated IDDs showed an excellent agreement with the measured experimental curves. The measured FWHM of the pencil beam in air at the isocenter was from 0.7 to 2.2 cm for protons. For carbon ions, two sets of spot size are available: from 0.4 to 0.8 cm (for the smaller) and from 0.8 to 1.1 cm (for the larger one). The spot position was within +/-1 mm over the whole 20x20 cm2 scan field; homogeneity in a uniform squared field was within +/-5% for both particle types at any energy. QA results exceeding tolerance levels were rarely found. In the reporting period, the machine downtime was around 6%, of which 4.5% was due to planned maintenance shutdowns. Conclusions: After successful dosimetric beam commissioning, quality assurance measurements performed during a 24-month period show very stable beam characteristics.

The COVID-19 outbreak has stimulated the interest in the proposal of novel epidemiological models to predict the course of the epidemic so as to help planning effective control strategies. In particular, in order to properly interpret the available data, it has become clear that one must go beyond most classic epidemiological models and consider models that, like the recently proposed SIDARTHE, offer a richer description of the stages of infection. The problem of learning the parameters of these models is of crucial importance especially when assuming that they are time-variant, which further enriches their effectiveness. In this paper we propose a general approach for learning time-variant parameters of dynamic compartmental models from epidemic data. We formulate the problem in terms of a functional risk that depends on the learning variables through the solutions of a dynamic system. The resulting variational problem is then solved by using a gradient flow on a suitable, regularized functional. We forecast the epidemic evolution in Italy and France. Results indicate that the model provides reliable and challenging predictions over all available data as well as the fundamental role of the chosen strategy on the time-variant parameters., Comment: 12 pages, 7 figures

Experimental observations of droplet size sustained oscillations are reported in a two-phase flow between a lamellar and a sponge phase. Under shear flow, this system presents two different steady states made of monodisperse multilamellar droplets, separated by a shear-thinning transition. At low and high shear rates, the droplet size results from a balance between surface tension and viscous stress whereas for intermediate shear rates, it becomes a periodic function of time. A possible mechanism for such kind of oscillations is discussed.