Help

fr
en
de

Zerara Mohammed

Responsable de l'Unité Projet

Main contract

Responsable de l'Unité Projet

HES-SO Genève Direction
Campus Battelle, Bâtiment F, Rue de la Tambourine 2, 1227 Carouge, CH

HESDI

Master Applied Mathematics - Data Science - Cergy Paris Université - Cours commun avec PSL (Université Paris Dauphine)

Deep Learning
Machine Learning

Ongoing

Combined machine learning and Raman spectroscopy for enterobacter sakazakii bacteria detection

Role: Partner

Description du projet :

The aim of this project is to detect Enterobacter sakazakii bacteria in powder infant formula milk using Raman spectroscopy. Machine learning and deep learning algorithms have been used to detect the presence of enterobacter sakazakii. Deep learning algorithms did not require prior pre-processing for background removal nor spectra smoothing.

Research team within HES-SO: Zerara Mohammed

Durée du projet: 01.03.2022 - 30.06.2023

Statut: Ongoing

Machine - Deep Learning and hyperspectral image classification and anomaly detection.

Role: Collaborator

Financement: The project is funded by an international organization.

Description du projet :

The aim of this project is to exploit hyperspectral imaging for features detection in compounds and products (food, biological samples). Machine learning methods are used to perform binary and multi-label classification, as well as anomaly detection.

Research team within HES-SO: Zerara Mohammed

Statut: Ongoing

2024

Miniaturized spectrometer with intrinsic long-term image memory

Scientific paper

Zerara Mohammed

Nature Communications, 2024 , vol. 15, no 676, pp. 1-11

Link to the publication

Summary:

Miniaturized spectrometers have great potential for use in portable optoelectronics and wearable sensors. However, current strategies for miniaturization rely on von Neumann architectures, which separate the spectral sensing, storage, and processing modules spatially, resulting in high energy consumption and limited processing speeds due to the storage-wall problem. Here, we present a miniaturized spectrometer that utilizes a single SnS2/ReSe2 van der Waals heterostructure, providing photodetection, spectrum reconstruction, spectral imaging, long-term image memory, and signal processing capabilities. Interface trap states are found to induce a gate-tunable and wavelength-dependent photogating effect and a non-volatile optoelectronic memory effect. Our approach achieves a footprint of 19 μm, a bandwidth from 400 to 800 nm, a spectral resolution of 5 nm, and a > 104 s long-term image memory. Our single-detector computational spectrometer represents a path beyond von Neumann architectures.

2023

Miniaturized Computational Spectrometer

Scientific paper

Zerara Mohammed

IEEE Nanotechnology Magazine, 2023 , vol. 17, no 6, pp. 36-42

Link to the publication

Summary:

Miniaturized computational spectrometers are opto-electronic instruments that can measure the intensity of light as a function of its wavelength, providing valuable information for applications such as material analysis, environmental monitoring, and medical diagnostics. In recent years, advances in nanotechnology, micro-electro-mechanical systems (MEMS), and computational algorithms have allowed significant miniaturization of spectrometers, vastly reducing their footprint, weight, and cost compared with traditional benchtop instruments. Despite these advances, several challenges still remain in the development and widespread adoption of miniaturized computational spectrometers. In this article, we begin by providing an overview of the benefits and potential applications of miniaturized computational spectrometers. Following that, we delve into detailed discussion on the materials utilized and the underlying physical mechanisms at play within these devices. We then review the computational algorithms employed for spectrum reconstruction. Lastly, we attempt to shed light on the outstanding challenges faced in this field.

2022

Hyperspectral images ConvNets based classification

Scientific paper

Zerara Mohammed

To be submitted, 2022

Summary:

Publication in preparation.

2021

Machine Learning and Statistical Methods for Time Series Forecasting: a Case Study with Water Demand

Scientific paper

Zerara Mohammed

TechRxiv, 2021

Link to the publication

Summary:

Nowadays, time series forecasting is commonly
performed using several methods which can be classified in
two groups; the stochastic methods and the machine learning
based approaches. The former group includes amongst other
auto-regressive (AR), moving average (MA), a combination of
AR and MA called ARIMA (with a differential term), the
seasonal ARIMA called SARIMA, the vector auto-regression
(VAR) technique for multi-variate series, and others. The latter
group includes regular neural networks such as the multi-layer
perceptron (MLP), more sophisticated deep learning algorithms
such as GRU and LSTM, two recurrent neural networks, support
vector machines (SVM) and its application to regression problems
(SVR), a more recent algorithm called XGBoost, and others.
In this study, most of these approaches and techniques will
be used to forecast water demand of a public company which
is in charge of distributing water, as potable water, to several
hundred thousand of end-customers. These different methods
will be briefly introduced, then assessed with the water demand
case study and discussed in details.

2015

Updates in Rhea - a manually curated resource of biochemical reactions

Scientific paper

Zerara Mohammed

Nucleic Acids Res., 2015 , vol. 43, no Database Issue, pp. 459-464

Link to the publication

Summary:

Rhea (http://www.ebi.ac.uk/rhea) is a comprehensive and non-redundant resource of expert-curated biochemical reactions described using species from the ChEBI (Chemical Entities of Biological Interest) ontology of small molecules. Rhea has been designed for the functional annotation of enzymes and the description of genome-scale metabolic networks, providing stoichiometrically balanced enzyme-catalyzed reactions (covering the IUBMB Enzyme Nomenclature list and additional reactions), transport reactions and spontaneously occurring reactions. Rhea reactions are extensively curated with links to source literature and are mapped to other publicly available enzyme and pathway databases such as Reactome, BioCyc, KEGG and UniPathway, through manual curation and computational methods. Here we describe developments in Rhea since our last report in the 2012 database issue of Nucleic Acids Research. These include significant growth in the number of Rhea reactions and the inclusion of reactions involving complex macromolecules such as proteins, nucleic acids and other polymers that lie outside the scope of ChEBI. Together these developments will significantly increase the utility of Rhea as a tool for the description, analysis and reconciliation of genome-scale metabolic models.

2009

Parameterization of an empirical model for the prediction of n -octanol, alkane and cyclohexane/water as well as brain/blood partition coefficients

Scientific paper

Zerara Mohammed

J. Comput. Aided Mol. Des., 2009 , vol. 23, no 2, pp. 105-111

Link to the publication

Summary:

Quantitative information of solvation and transfer free energies is often needed for the understanding of many physicochemical processes, e.g the molecular recognition phenomena, the transport and diffusion processes through biological membranes and the tertiary structure of proteins. Recently, a concept for the localization and quantification of hydrophobicity has been introduced (Jäger et al. J Chem Inf Comput Sci 43:237–247, 2003). This model is based on the assumptions that the overall hydrophobicity can be obtained as a superposition of fragment contributions. To date, all predictive models for the logP have been parameterized for n-octanol/water (logP oct ) solvent while very few models with poor predictive abilities are available for other solvents. In this work, we propose a parameterization of an empirical model for n-octanol/water, alkane/water (logP alk ) and cyclohexane/water (logP cyc ) systems. Comparison of both logP alk and logP cyc with the logarithms of brain/blood ratios (logBB) for a set of structurally diverse compounds revealed a high correlation showing their superiority over the logP oct measure in this context.

2008

Parametrization of the molecular free energy surface density (MolFESD) for different solvents and brain-blood barrier partitioning

Scientific paper

Zerara Mohammed

Chemistry Central Journal, 2008 , vol. 2, pp. 20-20

Link to the publication

Summary:

Quantitative information of solvation and transfer free energies is often needed for the understanding of many physicochemical processes, e.g the molecular recognition phenomena, the transport and diffusion processes through biological membranes and the tertiary structure of proteins. Recently, the molecular free energy surface density concept (MolFESD) has been introduced [1, 2]. This model is based on the assumptions that the overall hydrophobicity can be obtained as a superposition of fragment contributions and that the corresponding free energy can be calculated as an integral of the MolFESD over the molecular solvent accessible surface. The scalar quantity 3D-FED, the three-dimensional free energy density offers a physical basis to the establishment of a new predictive model with limited empirical character. Although this volume density is accessible from more accurate methods e.g. Monte-Carlo methods, the Grid program [3] has proved to be suitable for a rapid evaluation of the 3D-FED. It uses empirical force field to determine the interaction energy of a particular probe molecule (e.g water) for all points of a regular three dimensional grid in which the target molecule is enclosed. In the present study, parametrization of the MolFESD is done in two steps using the same strategy described in the previous work of Jäger et al. [1, 2]; some model parameters were fitted to computed data (provided by Grid program) while other parameters were fitted to experimental logP values. Unlike most fragment based predictive models for logP, the MolFESD concept calculates logP values which depend on the area and shape of molecular surfaces and in turn, on the corresponding molecular conformations. A parametrization of the MolFESD is done for three solvents (n-octanol/water, alkane/water and cyclohexane/water) followed by its application to the prediction of brain-blood barrier partition coefficients (logBB).

On the generation of intrinsic electric dipole fields as the basis for the understanding of the morphogenesis of fluoroapatite-gelatine nano-composites

Scientific paper

Zerara Mohammed

Chemistry Central Journal, 2008 , vol. 2, no S1, pp. 39-39

Link to the publication

Summary:

Systems of ordered fractal aggregates of fluoroapatite-gelatine composites were chosen to mimic the growth of the biosystem apatite-collagen, which plays an important role in the human body as functional material of teeth and bones. The morphogenesis of these particles starts with elongated hexagonal prismatic seeds, followed by fractal branching and the development of growing dump-bell states. In order to gain insight into structure formation a lot of experimental investigations were performed [1]. High resolution TEM micrograph of the [001] zone of a composite seed showed also the presence of triple-helical macromolecules oriented along the c-axis [2]. The general principles of the dramatic selforganization process are not yet understood. It is proposed that they should be manifested already in the structure of the seed. Molecular dynamics simulations on the basis of atomistic resolution models offer some ideas for the organisation principle. However, the effort is typically immense and unacceptable for systems with increasing size. The simulation scenario has to be drastically simplified. Our approach starts from the assumption of the formation of an intrinsic electric field - built by the permanent dipoles contained in each individual composite crystal [3] - which takes over control of the aggregate-growth. This assumption is consistent with the observation of the biological significance of electric fields (pyro- piezo electricity) during bones formation. The simulation strategy can be described as follows. Adopting a coarse grained model, the first main task is the calculation of the electric field [4] (on the basis of given dipole arrangements). Each collagen molecule, is represented by two beads with opposite charges, in the hypothesis that all the C-termini and N-termini are completely deprotonated and protonated, respectively. All the beads are arranged in the seed according to the experimental geometrical parameters. A Monte Carlo simulation is performed in order to optimise the still missing geometrical parameters. According to the resulting force field lines, the orientation of the seeds belonging to the following generations can be predicted.

pyVib, a computer program for the analysis of infrared and Raman optical activity

Scientific paper

Zerara Mohammed

J. Comput. Chem., 2008 , vol. 29, no 2, pp. 306-311

Link to the publication

Summary:

A new program called pyVib has been developed as a tool for the analysis of Gaussian (Gaussian 03, Gaussian Inc., Pittsburgh, PA) outputs of vibrational absorption (IR), Raman as well as vibrational optical activity (VOA) spectra calculations. This program has been designed to help the computational chemistry practitioner in the task of analyzing and visualizing molecular vibrations and cross sections. In particular, the analysis of absorption and scattering cross sections can be done using new tools such as group coupling matrices (GCMs) and atomic contribution patterns (ACPs) as either 2D or 3D representations, respectively (Hug, Chem Phys 2001, 264, 53). It reads the Hessian, the atomic polar tensors (APTs), the atomic axial tensors (AATs) (Nafie, J Chem Phys 1983, 79, 4950), and the gradients of the various polarizability tensors involved in VOA calculations and stored in Gaussian fchk ascii files. pyVib is capable of picking suitably chosen atoms or group of atoms for evaluating the contribution of each atom or defined groups of atoms to the calculated VOA scattered intensities. All the results generated by pyVib can be visualized in real-time but can also be transferred to text editors and electronic spreadsheets, which facilitate a detailed subsequent analysis and the visualization by other graphical user interfaces (GUIs).

2007

Optical investigation of spin-crossover in cobalt(II) bis-terpy complexes

Scientific paper

Zerara Mohammed

Inorganica Chimica Acta, 2007 , vol. 360, no 13, pp. 3945-3950

Link to the publication

Summary:

The spin transition of the [Co(terpy)2]2+ complex (terpy = 2,2′:6′,2″-terpyridine) is analysed based on experimental data from optical spectroscopy and magnetic susceptibility measurements. The single crystal absorption spectrum of [Co(terpy)2](ClO4)2 shows an asymmetric absorption band at 14 400 cm−1 with an intensity typical for a spin-allowed d–d transition and a temperature behaviour typical for a thermal spin transition. The single crystal absorption spectra of suggest that in this compound, the complex is essentially in the high-spin state at all temperatures. However, the increase in intensity observed in the region of the low-spin MLCT transition with increasing temperature implies an unusual partial thermal population of the low-spin state of up to about 10% at room temperature. Finally, high-spin → low-spin relaxation curves following pulsed laser excitation for [Co(terpy)2](ClO4)2 dispersed in KBr discs, and as a comparison for the closely related [Co(4-terpyridone)2](ClO4)2 spin-crossover compound are given.

Spin-crossover in cobalt(II) imine complexes

Scientific paper

Zerara Mohammed

Coordination Chemistry Reviews, 2007 , vol. 251, no 3-4, pp. 364-378

Link to the publication

Summary:

Whereas there are hundreds of known iron(II) spin-crossover compounds, only a handful of cobalt(II) spin-crossover compounds have been discovered to date, and hardly an in depth study on any of them exists. This review begins with an introduction into the theoretical aspects to be considered when discussing spin-crossover compounds in general and cobalt(II) systems in particular. It is followed by case studies on [Co(bpy)3]2+ and [Co(terpy)2]2+ (bpy = 2,2′-bipyridine, terpy = 2,2′:6′,2″-terpyridine) presenting and discussing results from magnetic susceptibility measurements, X-ray crystallography, optical spectroscopy, and EPR spectroscopy.

2006

Density-Functional Theory Investigation of the Geometric, Energetic, and Optical Properties of the Cobalt (II) tris (2, 2 ‘-bipyridine) Complex in the High-Spin and the Jahn- Teller Active Low-Spin States

Scientific paper

Zerara Mohammed

Journal of Chemical Theory and Computation, 2006 , vol. 2, no 5, pp. 1342-1359

Link to the publication

Summary:

State-of-the-art generalized gradient approximation (GGA) (PBE, OPBE, RPBE, OLYP, and HCTH), meta-GGA (VSXC and TPSS), and hybrid (B3LYP, B3LYP*, O3LYP, and PBE0) functionals are compared for the determination of the structure and the energetics of the D3 [Co(bpy)3]2+ complex in the 4A2 and 4E trigonal components of the high-spin 4T1g( ) state and in the low-spin 2E state of octahedral 2Eg( ) parentage. Their comparison extends also to the investigation of the Jahn−Teller instability of the 2E state through the characterization of the extrema of C2 symmetry of this spin state's potential energy surface. The results obtained for [Co(bpy)3]2+ in either spin manifold are very consistent among the functionals used and are in good agreement with available experimental data. The functionals, however, perform very differently with respect to the spin-state energetics because the calculated values of the high-spin/low-spin energy difference Δ vary between −3212 and 3919 cm-1. Semilocal functionals tend to give too large Δ values and thus fail to correctly predict the high-spin state as the ground state of the isolated complex, while hybrid functionals tend to overestimate the stability of the high-spin state with respect to the low-spin state. Reliable results are, however, obtained with the OLYP, HCTH, B3LYP*, and O3LYP functionals which perform best for the description of the isolated complex. The optical properties of [Co(bpy)3]2+ in the two spin states are also analyzed on the basis of electronic excitation calculations performed within time-dependent density functional response theory. The calculated absorption and circular dichroism spectra agree well with experimental results.

2004

Cobalt (II)-tris-2, 2′-bipyridine as a Spin-Crossover Complex: Evidence for Cooperative Effects in Three-Dimensional Oxalate Networks

Scientific paper

Zerara Mohammed

ChemPhysChem, 2004 , vol. 5, no 3, pp. 395-399

Link to the publication

Summary:

In a spin: In the three-dimensional oxalate network structures of composition [CoxMII1-x(bpy)3][MICr(ox)3], the spin state of the [Cox(bpy)3]2+ complex can be tuned by means of chemical pressure. With MI=Na it is a classic high-spin complex. Substitution of Na by Li stabilises the complex and it becomes a spin-crossover complex. Dilution with MII=Fe reinforces this effect, and MII=Zn reverses it.

Fine Tuning the Electronic Properties of [M (bpy) 3] 2+ Complexes by Chemical Pressure (M= Fe 2+, Ru 2+, Co 2+, bpy= 2, 2′-Bipyridine)

Book chapter

Zerara Mohammed

, Optical Spectra and Chemical Bonding in Inorganic Compounds. 2004, Berlin, Heidelberg : Springer

Link to the publication

Summary:

The optical and photophysical properties of three classic transition metal complexes, namely [Fe(bpy)3]2+, [Ru(bpy)3]2+ and [Co(bpy)3]2+, can be fine-tuned by doping them into a variety of inert host lattices. The effects of the guest-host interactions are discussed in terms of a chemical pressure. In this chapter we show how this chemical pressure can accelerate the high-spin→low-spin relaxation in the iron complex by more than three orders of magnitude, how it can increase the activation energy of the thermal quenching of the famous orange luminescence of the ruthenium complex by a factor of two, and how it can push the high-spin cobalt complex to become a spin-crossover complex. [Fe(bpy)3]2+ –[Ru(bpy)3]2+ –[Co(bpy)3]2+ –Intersystem crossing–Spin-crossover–Luminescence–Radiationless deactivation–Chemical pressure–Guest-host interactions

2003

Transition de spin dans le complexe cobalt(II)tris(2,2'-bipyridine)

Doctoral thesis

Zerara Mohammed

2003, Département de Chimie Physique : Université de Genève. 180 p.

Hauser Andreas

Link to the publication

Summary:

Etude théorique, expérimentale et simulations informatiques du phénomène de transition de spin d'un composé de cobalt(II). Les propriétés thermodynamiques, magnétiques et optiques ont été expliquées au moyen de la mécanique quantique, de la théorie des groupes et de la théorie des perturbations.

2002

Chemical pressure

Scientific paper

Zerara Mohammed

Chimia, 2002 , vol. 56, no 12, pp. 685-689

Link to the publication

Summary:

The physical and photophysical properties of three classic transition metal complexes, namely [Fe(bpy)3]2+, [Ru(bpy)3]2+, and [Co(bpy)3]2+, can be tuned by doping them into a variety of inert crystalline host lattices. The underlying guest-host interactions are discussed in terms of a chemical pressure.

Achievements