Max Planck Computing and Data Facility

Acute myeloid leukemia (AML) remains challenging to treat due to extensive genetic heterogeneity, high relapse rates, and treatment-related toxicity. Although drug combinations offer therapeutic promise, their selection is often empirical. Here, we introduce Combinatorial Proteome Integral Solubility/Stability Alteration analysis (CoPISA), a high-throughput proteomics workflow that captures protein solubility/stability alterations uniquely induced by drug combinations. We applied CoPISA to two rationally designed AML drug pairs, LY3009120-sapanisertib (LS) and ruxolitinib-ulixertinib (RU), previously identified as the most effective and least toxic combinations among many candidates and validated in AML cell lines, patient-derived samples and zebrafish xenograft models. We uncovered an emergent mechanism termed "conjunctional targeting", in which combinatorial drug action induces combination-exclusive protein targets consistent with an AND-gate logic model. LS-specific converged on SUMOylation, chromatin condensation, and VEGF-linked adhesion, while RU-specific targets disrupted DNA-damage checkpoints, mitochondrial bioenergetics, and RNA-splicing. Post-translational modification analysis revealed combination-induced acetylation, methylation, and phosphorylation of key AML proteins, including NPM1. Network analysis demonstrated that a substantial fraction of AML-associated proteins targeted by CoPISA are unique to combinations, including DNMT3A, NPM1, and TP53. By uncovering a mechanistic layer beyond classical synergy, CoPISA provides a robust framework for the precision-guided design of combinatorial therapies in heterogeneous cancers.

Relating billions of proteins across the tree of life remains a challenging task for comparative biosphere genomics and artificial intelligence-driven structure prediction. Here we present DIAMOND DeepClust, a cascaded, ultra-fast clustering method enabling planetary-scale organization of protein space, scaling to trillions of sequences while retaining sensitivity at low identity. Aggregating 19 billion biosphere proteins into 544 million nonsingleton clusters, we show that using our DeepClust database, available for download, can enhance structure prediction with AlphaFold2.

Abstract Besides classical feed-forward neural networks such as multilayer perceptrons, also neural ordinary differential equations (neural ODEs) have gained particular interest in recent years. Neural ODEs can be interpreted as an infinite depth limit of feed-forward or residual neural networks. We study the input–output dynamics of finite and infinite depth neural networks with scalar output. In the finite-depth case, the input is a state associated with a finite number of nodes, which maps under multiple non-linear transformations to the state of one output node. In analogy, a neural ODE maps an affine linear transformation of the input to an affine linear transformation of its time- T map. We show that, depending on the specific structure of the network, the input–output map has different properties regarding the existence and regularity of critical points. These properties can be characterized via Morse functions, which are scalar functions where every critical point is non-degenerate. We prove that critical points cannot exist if the dimension of the hidden layer is monotonically decreasing or the dimension of the phase space is smaller than or equal to the input dimension. In the case that critical points exist, we classify their regularity depending on the specific architecture of the network. We show that, except for a Lebesgue measure zero set in the weight space, each critical point is non-degenerate if for finite-depth neural networks, the underlying graph has no bottleneck, and if for neural ODEs, the affine linear transformations used have full rank. For each type of architecture, the proven properties are comparable in the finite and infinite depth cases. The established theorems allow us to formulate results on universal embedding and universal approximation, i.e., on the exact and approximate representation of maps by neural networks and neural ODEs. Our dynamical systems viewpoint on the geometric structure of the input–output map provides a fundamental understanding of why certain architectures perform better than others.

Abstract A major challenge when investigating intrinsically disordered proteins (IDPs) pertains to understanding how secondary structure formation across otherwise disordered ensembles, relates to function. While coarse-grained (CG) simulation methods are instrumental to describe ensemble-to-function relationships in IDPs, they remain largely incapable of sampling secondary structure. We here present NapshiftC α , an artificial neural network capable of modelling NMR chemical shifts from C α -mapped CG protein conformations, leveraging its predictions to restrain simulations and yield experimentally-informed IDP ensembles that contain secondary structure. We applied NapshiftC α on a large set of proteins in isolation or condensed phase, and found that, when secondary structure is present, CS incorporation yields improved agreements with smFRET spectroscopy and resolves the over-reliance of CG models on near linear conformers. By incorporating observables that are routinely collected for IDPs but never coupled to CG simulations, NapshiftC α ultimately informs on both local and global conformational features of function-defining IDP ensembles.

We present very high-energy optical photometry and spectroscopic observations of SN 2024bch in the nearby galaxy NGC 3206 (∼20 Mpc). We used gamma-ray observations performed with the first Large-Sized Telescope (LST-1) of the Cherenkov Telescope Array Observatory (CTAO) and optical observations with the Liverpool Telescope (LT) combined with data from public repositories to evaluate the general properties of the event and the progenitor star. No significant emission above the LST-1 energy threshold for this observation (∼100 GeV) was detected in the direction of SN 2024bch, and we computed an integral upper limit on the photon flux of F γ (> 100 GeV)≤3.61 × 10 −12 cm −2 s −1 based on six nonconsecutive nights of observations with the LST-1, between 16 and 38 days after the explosion. Employing a general model for the gamma-ray flux emission, we found an upper limit on the mass-loss-rate to wind-velocity ratio of Ṁ / u w ≤ 10 −4 M ⊙ / yr s/km, although gamma-gamma absorption could potentially have skewed this estimation, effectively weakening our constraint. From spectro-photometric observations we found progenitor parameters of M pr = 11 – 20 M ⊙ and R pr = 531 ± 125 R ⊙ . Finally, using archival images from the Hubble Space Telescope, we constrained the luminosity of the progenitor star to log (L pr / L ⊙ ) ≤ 4.82 and its effective temperature to T pr ≤ 4000 K. Our results suggest that SN 2024bch is a type IIn-L supernova that originated from a progenitor star consistent with a red supergiant. We show how the correct estimation of the mass-loss history of a supernova will play a major role in future multiwavelength observations.

Abstract In this work, we describe validation efforts conducted for the stellarator Monte-Carlo neutral beam and fast ion code BEAMS3D. This code is used at Wendelstein 7-X to analyze fast ion physics and has numerous integrated interfaces facilitating a wide range of applications. We use experimental fast-ion D$_\alpha$ data from the ASDEX Upgrade tokamak, covering on- and off-axis neutral beam heating phases. We obtained synthetic D$_\alpha$ spectra using the FIDASIM code, leveraging a newly developed interface that allows a distribution function from BEAMS3D to be used. Comparing the simulated distributions from BEAMS3D and the well-established NUBEAM codes to the data shows that both codes reproduce the experiment equally well given the same initial conditions. The experimental FIDA spectra are quantitatively matched. The slowing down model of BEAMS3D is thus validated and can be applied to quantitative stellarator investigations in future studies.

Abstract In the design process of gear systems, various computational tools are used to optimize the operational behavior. Due to time constraints, the initial stages of development typically assume ideal, deviation-free gear geometries. However, deviations resulting from manufacturing significantly affect performance in later applications. This publication presents a concept to incorporate the influence of manufacturing tolerances during early development phases systematically. Based on an initial set of detailed simulations, a surrogate model is developed and trained. This model can quickly generate a wide range of results, efficiently accounting for the large variance of deviations. The shown approach combines the advantages of detailed simulations with the speed of machine learning models.

We benchmarked the performance of the GROMACS 2024 molecular dynamics (MD) code on a modern high-performance computing (HPC) cluster with AMD CPUs on up to 65,536 CPU cores. We used five different MD systems, ranging in size from about 82,000 to 204 million atoms, and evaluated their performance using two different Message Passing Interface (MPI) libraries, Intel-MPI and Open-MPI. The largest system showed near-perfect strong scaling up to 512 nodes or 65,536 cores, maintaining a parallel efficiency above 0.9 even at the highest level of parallelization. Energy efficiency for a given number of nodes was generally equal to or slightly better than parallel efficiency. We achieved peak performances of 687 ns/d for the 82k atom system, 116 ns/d for the 53M atom system, and about 35 ns/d for the largest 204M atom system. These results demonstrate that highly optimized software running on a state-of-the-art HPC cluster provides sufficient computing power to simulate biomolecular systems at the mesoscale of viruses and organelles, and potentially small cells in the near future.

We report an , nonrelativistic QED method that couples light and matter self-consistently beyond the electric dipole approximation and without multipolar truncations. This method is based on an extension of the Maxwell-Pauli-Kohn-Sham approach to a full minimal coupling Hamiltonian, where the space- and time-dependent vector potential is coupled to the matter system, and its back reaction to the radiated fields is generated by the full current density. The implementation in the open-source code is designed for massively parallel multiscale simulations considering different grid spacings for the Maxwell and matter subsystems. Here, we show applications of this framework to simulate renormalized Cherenkov radiation of an electronic wave packet, magneto-optical effects with nonchiral light in nonchiral molecular systems, and renormalized plasmonic modes in a nanoplasmonic dimer. We show that in some cases, the beyond-dipole effects cannot be captured by a multipolar expansion Hamiltonian in the length gauge. Finally, we discuss further opportunities enabled by the framework in the field of twisted light and orbital angular momentum, inelastic light scattering, and strong-field physics.