Computational modeling of polymeric materials has traditionally been divided between quantum chemical methods that capture electronic structure at small scales and coarse-grained approaches that enable simulation of mesoscale phenomena. However, modern challenges in polymer material design require descriptions that capture electronic structure and mesoscale features simultaneously. In this contribution, I will present recent advances that integrate electronic structure directly into coarse-grained polymer simulations [1]. By leveraging machine learning techniques, we enable accurate predictions of material properties that depend on both electronic structure and mesoscale morphology. I will show how the developed scheme can be applied to study ionic, electronic, and structural properties of polymer electrodes for all-organic batteries [2]. Furthermore, I will describe current efforts to bridge this approach with transferable coarse-grained models [3], narrowing the traditional gap between quantum and mesoscale methods and enabling high-throughput [4] simulations of polymers with advanced electronic, optical, or reactive functionalities.

Color tuning of solid-state emissive materials is of great interest for both fundamental research and practical applications. [1] The development of structurally simple multicolor polymers that do not require sophisticated synthetic methodologies is appealing, but rather challenging. Despite enormous ef-forts on molecular design and engineering, a general and facile polymer platform that offer high flexi-bility and broad extensibility in emission color tuning remains scarce. In recent years, through-space charge transfer (TSCT) has emerged as a novel mechanism for the design of highly emissive molecules.[2] Our group discovered a structurally remote TSCT process in well-defined polymer systems, which enabled continuous color tuning of polymer fluorescence in solid state via controlled polymeri-zation.[3] Using a single-acceptor fluorophore as the initiator for atom transfer radical polymerization, a series of electron-donor groups containing simple aromatic moieties were introduced by facile copol-ymerization or post-functionalization (Figure 1a). Recently, guided by a machine learning model, we further developed a TSCT polymer library with precisely tailorable emission wavelength, covering full visible light spectra.[4] This was achieved by fine-manipulation of donor-acceptor interplay via simple controlled polymer synthesis. We further demonstrated this TSCT polymer platform can be used to design various functional materials [4,5], including photochromic inks for information encryption, self-reporting fluorescent reagents for capsule damage indicating, and highly emissive polymer coatings for luminescent waveguides and solar concentrators.

Acknowledgements
We thank the financial support from Fondation Glaude et Giuliana (research project no. 1-005137) and Swiss National Sci-ence Foundation (Spark grant no. 190313).

The molecular design of semiconducting polymers (SCPs) has traditionally relied on varying monomer combinations and sequences, guided by insights from charge transport mechanisms. However, the link between controllable structural features and electronic properties remains elusive, leaving design rules for next-generation SCPs undefined.

Using high-throughput computational methods, we analyze over 100 state-of-the-art p- and n-type polymer models to derive statistically significant design rules. Our study disentangles the effects of key structural features, testing existing hypotheses and uncovering new structure-property relationships. Notably, we find that polymer rigidity has minimal influence on charge transport, while the newly introduced planarity persistence length proves to be a superior descriptor. We also demonstrate the predictive power of machine learning models trained on our dataset, enabling a data-driven approach to SCP design and accelerating the discovery of materials with tailored electronic properties.

This work provides a hierarchy of methodologies that can be used to screen polymers depending on the desired accuracy: descriptors for thousands, the atomistic soup model for hundreds, and full atomistic models for tens of SCPs. The unprecedented throughput and automation of these methods create a continuous feedback loop between modeling and experimentation, making theoretical predictions significantly faster than synthesis and characterization.

Finally, these methods are versatile and can be extended to predict diverse properties, including optical, thermoelectric, and mechanical, advancing in-silico design beyond incremental optimization of existing materials.

The growing demand for enhanced material functionality has recently led to a surge of interest in complex matter systems. Soft colloidal suspensions, in particular, serve as smart building blocks for responsive materials. Their tunable size and capacity to swell and deform in response to various stimuli make them ideal for integration into emerging micro- and nanoscale applications. Although these systems are commonly used in various fields, such as paints, food, cosmetics, and pharmaceuticals, a deeper understanding of them has yet to be fully explored. Specifically, investigating their mechanical behavior numerically remains challenging, as most well-known methods are based on elasticity while neglecting permeability.
Our mesoscale approach to this problem combines both bulk properties at the continuum level and internal particle structure at the particle level using non-equilibrium thermodynamics. Our two-scale dynamic model, which decouples position and shape dynamics—unlike experiments—makes it possible to investigate elasticity and permeability independently [1-2]. Making particular choices for the particle-particle and particle-solvent interactions, we investigate the phase diagram of prolates in quasi-3D space using particle-based simulations. Comparing previously investigated hard spherical [3] and non-spherical [4] colloids reveals the potential of soft permeable colloids in modern applications such as 4D printing.

Artificial intelligence is becoming an increasingly valuable tool for predicting polymer properties, offering a powerful alternative to traditional characterization methods. Among these properties, the solubility parameter is particularly challenging to determine, as it depends on complex molecular interactions. However, it plays a crucial role in assessing polymer compatibility, miscibility, and dissolution behavior, making accurate predictions essential for applications such as coatings, adhesives, and drug delivery. [1] While some platforms and literature sources provide estimations, they often require simulations for training, are limited to specific polymer classes, or introduce uncertainties in their predictions. [2–4] In this study, both machine learning and neural network algorithms were explored to develop a predictive tool for solubility parameters. The dataset comprises theoretical values obtained from established methods such as Hansen, Van Krevelen, and Hoy, alongside experimental data sourced from literature-based databases. The results show a mean absolute error of approximately 0.3 MPa^1/2, demonstrating an improvement in accuracy compared to existing platforms. This approach not only enhances predictive performance but also extends applicability to polymers that are not well-represented in current models, all while eliminating the need for computationally expensive simulations. To further assess its reliability, the tool was applied to a set of waterborne polyurethanes as a case study, confirming its effectiveness in predicting solubility parameters for complex polymer systems.