Annika Bande
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Annika Bande has a total of 183 co-authors. Here are the top co-authors by number of shared publications:
- Federico M. Pont
Universidad Nacional de Córdoba
9 shared publications - Joachim Dzubiella
Albert-Ludwigs-Universität Freiburg
8 shared publications - Tristan Petit
Helmholtz-Zentrum Berlin für Materialien und Energie
6 shared publications - Fabian Langkabel
Algorithmiq Ltd. 1 , Kanavakatu 3C, FI-00160 Helsinki
5 shared publications - Karol Palczynski
Research Group Simulations of Energy Materials, Helmholtz-Zentrum Berlin für Materialien und Energie, Hahn-Meitner-Platz 1, D-14109 Berlin, Germany
5 shared publications - Thorren Kirschbaum
Theory of Electron Dynamics and Spectroscopy, Helmholtz-Zentrum Berlin für Materialien und Energie GmbH 1 , Hahn-Meitner-Platz 1, 14109 Berlin
5 shared publications
Latest Publications
- Open Access
- 10 September 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Comparative Evaluation of Graph Neural Networks for Molecular Lipophilicity Prediction
Predicting molecular properties is a central challenge in cheminformatics and plays a critical role in drug discovery and materials science. Lipophilicity, which describes the partitioning of a molecule between lipid and aqueous environments, is a key physicochemical property influencing drug behavior. Although quantum mechanical methods can provide accurate predictions, they are often computationally demanding. In contrast, large experimental databases can be leveraged to train data-driven models such as graph neural networks (GNNs), which learn molecular representations directly from chemical structures. In this study, we compare four GNN architectures—graph convolutional networks (GCN), graph isomorphism networks (GIN), Graph Sample and AggregatE (GraphSAGE), and Attentive Fingerprint (Attentive FP)—for predicting lipophilicity using both regression and classification approaches. The models are evaluated on a benchmark dataset under consistent preprocessing, optimization, and training conditions. Our results show that the Attentive FP model consistently outperforms the other architectures, yielding improved predictive accuracy across both tasks. We further examine the impact of outlier removal during preprocessing, observing reduced prediction errors and improved robustness. These findings provide insight into the capabilities of representative GNN models for molecular lipophilicity prediction.
- 10 September 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessComparative Evaluation of Graph Neural Networks for Molecular Lipophilicity Prediction
Predicting molecular properties is a central challenge in cheminformatics and plays a critical role in drug discovery and materials science. Lipophilicity, which describes the partitioning of a molecule between lipid and aqueous environments, is a key physicochemical property influencing drug behavior. Although quantum mechanical methods can provide accurate predictions, they are often computationally demanding. In contrast, large experimental databases can be leveraged to train data-driven models such as graph neural networks (GNNs), which learn molecular representations directly from chemical structures. In this study, we compare four GNN architectures—graph convolutional networks (GCN), graph isomorphism networks (GIN), Graph Sample and AggregatE (GraphSAGE), and Attentive Fingerprint (Attentive FP)—for predicting lipophilicity using both regression and classification approaches. The models are evaluated on a benchmark dataset under consistent preprocessing, optimization, and training conditions. Our results show that the Attentive FP model consistently outperforms the other architectures, yielding improved predictive accuracy across both tasks. We further examine the impact of outlier removal during preprocessing, observing reduced prediction errors and improved robustness. These findings provide insight into the capabilities of representative GNN models for molecular lipophilicity prediction.
- Open Access
- 3 July 2026
- Published by IOP Publishing in Nano Futures
Roadmap on 5D Photonics for nano and beyond
As photonic systems grow more complex, it becomes increasingly difficult to capture their behaviour within the conventional four dimensions of space and time, particularly for systems operating at the nanometer scale, where strong confinement effects, near-field interactions, and subwavelength structuring introduce additional layers of complexity. The concept of 5D photonics reflects this shift by incorporating additional physical, material, computational, adaptive, and quantum degrees of freedom as active components in design, control, and function. Rather than defining a single extra coordinate, higher-dimensional photonics is about operating photonic systems within expanded, dynamically accessible state spaces where multiple dimensions can interact and evolve together. This roadmap brings together perspectives ranging from modeling and design concepts to experimental platforms, materials, components, and system-level implementations. It covers a wide spectrum of synthetic and structured dimensions, nonlinear and strong-field regimes, adaptive and reconfigurable architectures, cyber-physical and engineering approaches, as well as inherently high-dimensional quantum and excitonic systems. Across all these areas, higher-dimensional thinking emerges not as an abstract construct but as a practical tool for enabling new functionalities, overcoming conventional design limitations, and bridging physical systems with digital and AI-driven layers. By framing these diverse developments within a shared higher-dimensional perspective, the roadmap aims to provide orientation in a rapidly expanding field, reveal conceptual connections between traditionally separate areas of photonics, and highlight common challenges and opportunities. In doing so, it positions higher-dimensional photonics as a central paradigm for developing future photonic technologies that are increasingly adaptive, intelligent, and integrated across physical and virtual domains.
- 3 July 2026
- Published by IOP Publishing in Nano Futures
Open AccessRoadmap on 5D Photonics for nano and beyond
As photonic systems grow more complex, it becomes increasingly difficult to capture their behaviour within the conventional four dimensions of space and time, particularly for systems operating at the nanometer scale, where strong confinement effects, near-field interactions, and subwavelength structuring introduce additional layers of complexity. The concept of 5D photonics reflects this shift by incorporating additional physical, material, computational, adaptive, and quantum degrees of freedom as active components in design, control, and function. Rather than defining a single extra coordinate, higher-dimensional photonics is about operating photonic systems within expanded, dynamically accessible state spaces where multiple dimensions can interact and evolve together. This roadmap brings together perspectives ranging from modeling and design concepts to experimental platforms, materials, components, and system-level implementations. It covers a wide spectrum of synthetic and structured dimensions, nonlinear and strong-field regimes, adaptive and reconfigurable architectures, cyber-physical and engineering approaches, as well as inherently high-dimensional quantum and excitonic systems. Across all these areas, higher-dimensional thinking emerges not as an abstract construct but as a practical tool for enabling new functionalities, overcoming conventional design limitations, and bridging physical systems with digital and AI-driven layers. By framing these diverse developments within a shared higher-dimensional perspective, the roadmap aims to provide orientation in a rapidly expanding field, reveal conceptual connections between traditionally separate areas of photonics, and highlight common challenges and opportunities. In doing so, it positions higher-dimensional photonics as a central paradigm for developing future photonic technologies that are increasingly adaptive, intelligent, and integrated across physical and virtual domains.
- Open Access
- 6 May 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry A
- Vol. 130 (20) , 3981-3995
From Local Atomic Structure to X-ray Spectra: Absorber-Centric Machine-Learning Encoding
X-ray spectroscopy provides sensitive, element-specific insight into local geometric and electronic structures, but predictive first-principles simulations can be computationally expensive for large and chemically diverse molecular systems. Recent machine-learning approaches have shown promise in accelerating structure-to-spectrum prediction; however, most directly regress discretized spectral intensities and rely on hand-crafted geometric descriptors centered on the absorbing atom. Herein, we introduce a machine learning framework that encodes a detailed, environment-aware representation of the nuclear structure beyond the absorbing site. The model combines these descriptors with a physically motivated, multiscale Gaussian spectral basis whose coefficients are obtained via ridge projection, enforcing smoothness and spectral consistency. To further enhance robustness across chemical and conformational diversity, we employ a multiscale structural similarity loss that couples geometric and spectral resolution. This integrated approach yields accurate and transferable predictions across a wide range of molecular geometries and chemical environments while maintaining physical interpretability. The proposed framework establishes a physically structured and scalable route to machine-learned X-ray spectroscopy.
- 6 May 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry A
- Vol. 130 (20) , 3981-3995
Open AccessFrom Local Atomic Structure to X-ray Spectra: Absorber-Centric Machine-Learning Encoding
X-ray spectroscopy provides sensitive, element-specific insight into local geometric and electronic structures, but predictive first-principles simulations can be computationally expensive for large and chemically diverse molecular systems. Recent machine-learning approaches have shown promise in accelerating structure-to-spectrum prediction; however, most directly regress discretized spectral intensities and rely on hand-crafted geometric descriptors centered on the absorbing atom. Herein, we introduce a machine learning framework that encodes a detailed, environment-aware representation of the nuclear structure beyond the absorbing site. The model combines these descriptors with a physically motivated, multiscale Gaussian spectral basis whose coefficients are obtained via ridge projection, enforcing smoothness and spectral consistency. To further enhance robustness across chemical and conformational diversity, we employ a multiscale structural similarity loss that couples geometric and spectral resolution. This integrated approach yields accurate and transferable predictions across a wide range of molecular geometries and chemical environments while maintaining physical interpretability. The proposed framework establishes a physically structured and scalable route to machine-learned X-ray spectroscopy.
- Open Access
- 5 May 2026
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 28 (20) , 12432-12443
Graph neural network architectures for predicting the electrophilicity index: insights from 2D and 3D molecular graph representations
Benchmarking 2D and 3D GNNs for predicting the electrophilicity index ( ω ) of QM9 molecules reveals that geometry-aware architectures such as ALIGNN achieve the highest accuracy while 2D models offer superior computational efficiency.
- 5 May 2026
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 28 (20) , 12432-12443
Open AccessGraph neural network architectures for predicting the electrophilicity index: insights from 2D and 3D molecular graph representations
Benchmarking 2D and 3D GNNs for predicting the electrophilicity index ( ω ) of QM9 molecules reveals that geometry-aware architectures such as ALIGNN achieve the highest accuracy while 2D models offer superior computational efficiency.
- Open Access
Efficient Photo-Driven Electron Transfer from Amino Group-Decorated Adamantane to Water
Nanodiamonds in water can generate solvated electrons under ultraviolet (UV) excitation, but UV light constitutes only a small portion of solar energy. To harvest solar energy in the visible range, we investigate band gap reduction via surface amino functionalization and examine its impact on photo-excited charge transfer to water. Adamantane, the smallest nanodiamond, is used as a model due to its electron emission properties. Liquid water is first represented using water dimers and then complete solvation shell structures surrounding the adamantane. By systematically analyzing different functionalized adamantane structures, we find that nitrogen serves as the primary electron donor to nearby water molecules. Furthermore, the negative electron affinity of adamantane, which determines its emission capability, is preserved with half of the amino group functionalization on the surface. Our findings motivate further experimental verification using nanodiamonds with amino-functionalized surfaces.
Open AccessEfficient Photo-Driven Electron Transfer from Amino Group-Decorated Adamantane to Water
Nanodiamonds in water can generate solvated electrons under ultraviolet (UV) excitation, but UV light constitutes only a small portion of solar energy. To harvest solar energy in the visible range, we investigate band gap reduction via surface amino functionalization and examine its impact on photo-excited charge transfer to water. Adamantane, the smallest nanodiamond, is used as a model due to its electron emission properties. Liquid water is first represented using water dimers and then complete solvation shell structures surrounding the adamantane. By systematically analyzing different functionalized adamantane structures, we find that nitrogen serves as the primary electron donor to nearby water molecules. Furthermore, the negative electron affinity of adamantane, which determines its emission capability, is preserved with half of the amino group functionalization on the surface. Our findings motivate further experimental verification using nanodiamonds with amino-functionalized surfaces.
- Open Access
- 10 July 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (2)
Inter-particle Coulombic decay of highly excited resonance states: A study on competing relaxation mechanisms
The interparticle Coulombic decay (ICD) is a nonradiative decay process, in which energy released by an electron’s transition within one species leads to the ionization of a neighboring species. Over the years, research has focused on ICD in quantum dots (QDs), whose size-dependent excitation makes them suitable to tailor ICD properties. This work introduces a more highly excited state in the relaxing QD, aiming to allow for several ICD pathways and to clarify their competition by energy and selection rules. As an unexpected result, it was found that the otherwise directive energetic ICD conditions alone are insufficient to predict behavior in an ICD setting with competing pathways. In general, the pathway with the higher-energy transition is found to be slower and, moreover, allowed due to the breaking of symmetry selection rules in the QD array. It could, hence, be used in time-resolved measurements of the electron dynamics in self-assembled QDs, where the state preparation via charging from a reservoir is straightforward.
- 10 July 2025
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 163 (2)
Open AccessInter-particle Coulombic decay of highly excited resonance states: A study on competing relaxation mechanisms
The interparticle Coulombic decay (ICD) is a nonradiative decay process, in which energy released by an electron’s transition within one species leads to the ionization of a neighboring species. Over the years, research has focused on ICD in quantum dots (QDs), whose size-dependent excitation makes them suitable to tailor ICD properties. This work introduces a more highly excited state in the relaxing QD, aiming to allow for several ICD pathways and to clarify their competition by energy and selection rules. As an unexpected result, it was found that the otherwise directive energetic ICD conditions alone are insufficient to predict behavior in an ICD setting with competing pathways. In general, the pathway with the higher-energy transition is found to be slower and, moreover, allowed due to the breaking of symmetry selection rules in the QD array. It could, hence, be used in time-resolved measurements of the electron dynamics in self-assembled QDs, where the state preparation via charging from a reservoir is straightforward.
- Open Access
- 28 May 2025
- Published by Royal Society of Chemistry (RSC) in Catalysis Science & Technology
- Vol. 15 (14) , 4266-4278
NH3-induced activation of hydrophilic Fe–N–C nanocages for enhanced oxygen reduction reaction
A scalable process for the synthesis Fe–N–C catalysts for the ORR is introduced. After annealing in ammonia, the Fe–N–C catalysts exhibit an onset potential of 0.97 V and a potential of 0.89 V vs. RHE at a current density of −1 mA cm−2 in 0.1 M KOH.
- 28 May 2025
- Published by Royal Society of Chemistry (RSC) in Catalysis Science & Technology
- Vol. 15 (14) , 4266-4278
Open AccessNH3-induced activation of hydrophilic Fe–N–C nanocages for enhanced oxygen reduction reaction
A scalable process for the synthesis Fe–N–C catalysts for the ORR is introduced. After annealing in ammonia, the Fe–N–C catalysts exhibit an onset potential of 0.97 V and a potential of 0.89 V vs. RHE at a current density of −1 mA cm−2 in 0.1 M KOH.
- Open Access
- 21 March 2025
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 27 (16) , 8202-8211
X-ray absorption spectroscopy reveals charge transfer in π-stacked aromatic amino acids
X-ray absorption spectroscopy (XAS) and quantum mechanical calculations bear great potential to unravel π stacking side-chain interaction properties and structure in, e.g., proteins. However, core-excited state calculations for proteins and their associated interpretation for π-π interactions are challenging due to the complexity of the non-covalent interactions involved. A theoretical analysis is developed to decompose the core-to-valence transitions into their atomic contributions in order to characterize the π stacking of aromatic amino acids as a function of their non-covalent distance change. Three models were studied as a non-covalent mixed dimers of the phenylalanine, tyrosine and tryptophan amino acids. We found that there are carbon 1s → π∗ charge transfer transitions associated with the non-covalently paired aromatic amino acids through their side chains. The atomic-centered contributions to the electronic transition density quantify the excited state charge transfer of the pairing amino acid models, highlighting the π stacking interactions between their aromatic side chains.
- 21 March 2025
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 27 (16) , 8202-8211
Open AccessX-ray absorption spectroscopy reveals charge transfer in π-stacked aromatic amino acids
X-ray absorption spectroscopy (XAS) and quantum mechanical calculations bear great potential to unravel π stacking side-chain interaction properties and structure in, e.g., proteins. However, core-excited state calculations for proteins and their associated interpretation for π-π interactions are challenging due to the complexity of the non-covalent interactions involved. A theoretical analysis is developed to decompose the core-to-valence transitions into their atomic contributions in order to characterize the π stacking of aromatic amino acids as a function of their non-covalent distance change. Three models were studied as a non-covalent mixed dimers of the phenylalanine, tyrosine and tryptophan amino acids. We found that there are carbon 1s → π∗ charge transfer transitions associated with the non-covalently paired aromatic amino acids through their side chains. The atomic-centered contributions to the electronic transition density quantify the excited state charge transfer of the pairing amino acid models, highlighting the π stacking interactions between their aromatic side chains.
- 7 October 2024
- Published by American Physical Society (APS) in Physical Review A
- Vol. 110 (4)
Impact of the nuclear motion on the interparticle Coulombic electron capture
- 7 October 2024
- Published by American Physical Society (APS) in Physical Review A
- Vol. 110 (4)
Impact of the nuclear motion on the interparticle Coulombic electron capture
- Open Access
- 1 October 2024
- Published by AIP Publishing in AIP Advances
- Vol. 14 (10)
Transfer learning for molecular property predictions from small datasets
Machine learning has emerged as a new tool in chemistry to bypass expensive experiments or quantum-chemical calculations, for example, in high-throughput screening applications. However, many machine learning studies rely on small datasets, making it difficult to efficiently implement powerful deep learning architectures such as message passing neural networks. In this study, we benchmark common machine learning models for the prediction of molecular properties on two small datasets, for which the best results are obtained with the message passing neural network PaiNN as well as SOAP molecular descriptors concatenated to a set of simple molecular descriptors tailored to gradient boosting with regression trees. To further improve the predictive capabilities of PaiNN, we present a transfer learning strategy that uses large datasets to pre-train the respective models and allows us to obtain more accurate models after fine-tuning on the original datasets. The pre-training labels are obtained from computationally cheap ab initio or semi-empirical models, and both datasets are normalized to mean zero and standard deviation one to align the labels’ distributions. This study covers two small chemistry datasets, the Harvard Organic Photovoltaics dataset (HOPV, HOMO–LUMO-gaps), for which excellent results are obtained, and the FreeSolv dataset (solvation energies), where this method is less successful, probably due to a complex underlying learning task and the dissimilar methods used to obtain pre-training and fine-tuning labels. Finally, we find that for the HOPV dataset, the final training results do not improve monotonically with the size of the pre-training dataset, but pre-training with fewer data points can lead to more biased pre-trained models and higher accuracy after fine-tuning.
- 1 October 2024
- Published by AIP Publishing in AIP Advances
- Vol. 14 (10)
Open AccessTransfer learning for molecular property predictions from small datasets
Machine learning has emerged as a new tool in chemistry to bypass expensive experiments or quantum-chemical calculations, for example, in high-throughput screening applications. However, many machine learning studies rely on small datasets, making it difficult to efficiently implement powerful deep learning architectures such as message passing neural networks. In this study, we benchmark common machine learning models for the prediction of molecular properties on two small datasets, for which the best results are obtained with the message passing neural network PaiNN as well as SOAP molecular descriptors concatenated to a set of simple molecular descriptors tailored to gradient boosting with regression trees. To further improve the predictive capabilities of PaiNN, we present a transfer learning strategy that uses large datasets to pre-train the respective models and allows us to obtain more accurate models after fine-tuning on the original datasets. The pre-training labels are obtained from computationally cheap ab initio or semi-empirical models, and both datasets are normalized to mean zero and standard deviation one to align the labels’ distributions. This study covers two small chemistry datasets, the Harvard Organic Photovoltaics dataset (HOPV, HOMO–LUMO-gaps), for which excellent results are obtained, and the FreeSolv dataset (solvation energies), where this method is less successful, probably due to a complex underlying learning task and the dissimilar methods used to obtain pre-training and fine-tuning labels. Finally, we find that for the HOPV dataset, the final training results do not improve monotonically with the size of the pre-training dataset, but pre-training with fewer data points can lead to more biased pre-trained models and higher accuracy after fine-tuning.