A Euclidean transformer for fast and stable machine learned force fields.

Recent years have seen vast progress in the development of machine learned force fields (MLFFs) based on ab-initio reference calculations. Despite achieving low test errors, the reliability of MLFFs in molecular dynamics (MD) simulations is facing growing scrutiny due to concerns about instability over extended simulation timescales. Our findings suggest a potential connection between robustness to cumulative inaccuracies and the use of equivariant representations in MLFFs, but the computational cost associated with these representations can limit this advantage in practice. To address this, we propose a transformer architecture called SO3KRATES that combines sparse equivariant representations (Euclidean variables) with a self-attention mechanism that separates invariant and equivariant information, eliminating the need for expensive tensor products. SO3KRATES achieves a unique combination of accuracy, stability, and speed that enables insightful analysis of quantum properties of matter on extended time and system size scales. To showcase this capability, we generate stable MD trajectories for flexible peptides and supra-molecular structures with hundreds of atoms. Furthermore, we investigate the PES topology for medium-sized chainlike molecules (e.g., small peptides) by exploring thousands of minima. Remarkably, SO3KRATES demonstrates the ability to strike a balance between the conflicting demands of stability and the emergence of new minimum-energy conformations beyond the training data, which is crucial for realistic exploration tasks in the field of biochemistry.

Bone remodeling in survivors of pediatric hematopoietic stem cell transplantation: Impact of heavy resistance training.

BACKGROUND: Early-onset osteoporosis is a frequent late effect after pediatric hematopoietic stem cell transplantation (HSCT). It remains unknown if physical training can improve bone formation in these patients, as the transplantation procedure may cause sustained dysregulation of the bone-forming osteoblast progenitor cells. OBJECTIVE: We aimed to explore the effect of resistance training on bone remodeling in long-term survivors of pediatric HSCT. PROCEDURE: In this prospective, controlled intervention study, we included seven HSCT survivors and 15 age- and sex-matched healthy controls. The participants completed a 12-week heavy load, lower extremity resistance training intervention with three weekly sessions. We measured fasting serum levels of the bone formation marker "N-terminal propeptide of type I procollagen" (P1NP), and the bone resorption marker "C-terminal telopeptide of type I collagen" (CTX). The hypothesis was planned before data collection began. The trial was registered at Clinicaltrials.gov before including the first participant, with trial registration no. NCT04922970. RESULTS: Resistance training led to significantly increased levels of fasting P1NP in both patients (from 57.62 to 114.99 ng/mL, p = .03) and controls (from 66.02 to 104.62 ng/mL, p < .001). No significant changes in fasting CTX levels were observed. CONCLUSIONS: Despite previous high-dose cytotoxic therapy, long-term survivors of pediatric HSCT respond to resistance training with improvement of bone formation, comparable to that of healthy controls. This suggests that resistance training might be a promising non-pharmacological approach to prevent the early decline in bone mass, and should be considered as part of a follow-up program to counteract long-term sequela after pediatric HSCT.

Cytomegalovirus infection is associated with thymic dysfunction and chronic graft-versus-host disease after pediatric hematopoietic stem cell transplantation.

Pediatric hematopoietic stem cell transplantation (HSCT) is challenged by chronic graft-versus-host disease (cGvHD) significantly affecting survival and long-term morbidity, but underlying mechanisms including the impact of post-HSCT CMV infection are sparsely studied. We first investigated the impact of CMV infection for development of cGvHD in 322 children undergoing standard myeloablative HSCT between 2000 and 2018. Clinically significant CMV infection (n = 61) was an independent risk factor for chronic GvHD in a multivariable Cox regression analysis (HR = 2.17, 95% CI = 1.18-3.97, P = 0.013). We next explored the underlying mechanisms in a subcohort of 39 children. CMV infection was followed by reduced concentration of recent thymic emigrants (17.5 vs. 51.9 × 106/L, P = 0.048) and naïve CD4+ and CD8+ T cells at 6 months post-HSCT (all P < 0.05). Furthermore, CD25highFOXP3+ Tregs tended to be lower in patients with CMV infection (2.9 vs. 9.6 × 106/L, P = 0.055), including Tregs expressing the naivety markers CD45RA and Helios. CD8+ T-cell numbers rose after CMV infection and was dominated by exhausted PD1-expressing cells (66% vs. 39%, P = 0.023). These findings indicate that post-HSCT CMV infection is a main risk factor for development of chronic GvHD after pediatric HSCT and suggest that this effect is caused by reduced thymic function with a persistently impaired production of naïve and regulatory T cells in combination with increased peripheral T-cell exhaustion.

Disentangled Explanations of Neural Network Predictions by Finding Relevant Subspaces.

Explainable AI aims to overcome the black-box nature of complex ML models like neural networks by generating explanations for their predictions. Explanations often take the form of a heatmap identifying input features (e.g. pixels) that are relevant to the model's decision. These explanations, however, entangle the potentially multiple factors that enter into the overall complex decision strategy. We propose to disentangle explanations by extracting at some intermediate layer of a neural network, subspaces that capture the multiple and distinct activation patterns (e.g. visual concepts) that are relevant to the prediction. To automatically extract these subspaces, we propose two new analyses, extending principles found in PCA or ICA to explanations. These novel analyses, which we call principal relevant component analysis (PRCA) and disentangled relevant subspace analysis (DRSA), maximize relevance instead of e.g. variance or kurtosis. This allows for a much stronger focus of the analysis on what the ML model actually uses for predicting, ignoring activations or concepts to which the model is invariant. Our approach is general enough to work alongside common attribution techniques such as Shapley Value, Integrated Gradients, or LRP. Our proposed methods show to be practically useful and compare favorably to the state of the art as demonstrated on benchmarks and three use cases.

Biomolecular dynamics with machine-learned quantum-mechanical force fields trained on diverse chemical fragments.

The GEMS method enables molecular dynamics simulations of large heterogeneous systems at ab initio quality.

[Explainable artificial intelligence in pathology]. / Erklärbare Künstliche Intelligenz in der Pathologie.

With the advancements in precision medicine, the demands on pathological diagnostics have increased, requiring standardized, quantitative, and integrated assessments of histomorphological and molecular pathological data. Great hopes are placed in artificial intelligence (AI) methods, which have demonstrated the ability to analyze complex clinical, histological, and molecular data for disease classification, biomarker quantification, and prognosis estimation. This paper provides an overview of the latest developments in pathology AI, discusses the limitations, particularly concerning the black box character of AI, and describes solutions to make decision processes more transparent using methods of so-called explainable AI (XAI).

Markers of intestinal mucositis to predict blood stream infections at the onset of fever during treatment for childhood acute leukemia.

Despite chemotherapy-induced intestinal mucositis being a main risk factor for blood stream infections (BSIs), no studies have investigated mucositis severity to predict BSI at fever onset during acute leukemia treatment. This study prospectively evaluated intestinal mucositis severity in 85 children with acute leukemia, representing 242 febrile episodes (122 with concurrent neutropenia) by measuring plasma levels of citrulline (reflecting enterocyte loss), regenerating islet-derived-protein 3α (REG3α, an intestinal antimicrobial peptide) and CCL20 (a mucosal immune regulatory chemokine) along with the general neutrophil chemo-attractants CXCL1 and CXCL8 at fever onset. BSI was documented in 14% of all febrile episodes and in 20% of the neutropenic febrile episodes. In age-, sex-, diagnosis- and neutrophil count-adjusted analyses, decreasing citrulline levels and increasing REG3α and CCL20 levels were independently associated with increased odds of BSI (OR = 1.6, 1.5 and 1.7 per halving/doubling, all p < 0.05). Additionally, higher CXCL1 and CXCL8 levels increased the odds of BSI (OR = 1.8 and 1.7 per doubling, all p < 0.0001). All three chemokines showed improved diagnostic accuracy compared to C-reactive protein and procalcitonin. These findings underline the importance of disrupted intestinal integrity as a main risk factor for BSI and suggest that objective markers for monitoring mucositis severity may help predicting BSI at fever onset.

Toward Explainable Artificial Intelligence for Precision Pathology.

The rapid development of precision medicine in recent years has started to challenge diagnostic pathology with respect to its ability to analyze histological images and increasingly large molecular profiling data in a quantitative, integrative, and standardized way. Artificial intelligence (AI) and, more precisely, deep learning technologies have recently demonstrated the potential to facilitate complex data analysis tasks, including clinical, histological, and molecular data for disease classification; tissue biomarker quantification; and clinical outcome prediction. This review provides a general introduction to AI and describes recent developments with a focus on applications in diagnostic pathology and beyond. We explain limitations including the black-box character of conventional AI and describe solutions to make machine learning decisions more transparent with so-called explainable AI. The purpose of the review is to foster a mutual understanding of both the biomedical and the AI side. To that end, in addition to providing an overview of the relevant foundations in pathology and machine learning, we present worked-through examples for a better practical understanding of what AI can achieve and how it should be done.

Diffeomorphic Counterfactuals With Generative Models.

Counterfactuals can explain classification decisions of neural networks in a human interpretable way. We propose a simple but effective method to generate such counterfactuals. More specifically, we perform a suitable diffeomorphic coordinate transformation and then perform gradient ascent in these coordinates to find counterfactuals which are classified with great confidence as a specified target class. We propose two methods to leverage generative models to construct such suitable coordinate systems that are either exactly or approximately diffeomorphic. We analyze the generation process theoretically using Riemannian differential geometry and validate the quality of the generated counterfactuals using various qualitative and quantitative measures.

sVEGF-R1 in acute non-infectious toxicity syndromes after pediatric allogeneic hematopoietic stem cell transplantation.

BACKGROUND: Allogeneic hematopoietic stem cell transplantation (HSCT) is challenged by acute non-infectious toxicities, including sinusoidal obstruction syndrome (SOS), engraftment syndrome (ES) and capillary leak syndrome (CLS) among others. These complications are thought to be driven by a dysfunctional vascular endothelium, but the pathophysiological mechanisms remain incompletely understood, and the diagnoses are challenged by purely clinical diagnostic criteria that are partly overlapping, limiting the possibilities for progress in this field. There is, however, increasing evidence suggesting that these challenges may be met through the development of diagnostic biomarkers to improve diagnostic accuracy of pathogenetically homogenous entities, improved pre-transplant risk assessment and the early identification of patients with increased need for specific treatment. Soluble vascular endothelial growth factor receptor-1 (sVEGF-R1) is emerging as an important biomarker of endothelial damage in patients with trauma and sepsis but has not been studied in HSCT. OBJECTIVES: To investigate sVEGF-R1 as a marker of endothelial damage in pediatric HSCT patients by exploring associations with SOS, CLS, ES, and acute graft-versus-host disease (aGvHD). METHODS: We prospectively included 113 children undergoing myeloablative HSCT and measured sVEGF-R1 in plasma samples obtained weekly during the early period of transplantation and 3 months post-transplant. RESULTS: All over, sVEGF-R1 levels were significantly increased from day +7 after graft infusion, peaking at day +30, most pronounced in patients receiving busulfan. Patients considered to be at increased risk of SOS and therefore commenced on prophylactic defibrotide had significantly elevated levels of sVEGF-R1 before start of conditioning (446 pg/mL vs. 281 pg/mL, p = 0.0035), and this treatment appeared to stabilize sVEGF-R1 levels compared to patients not treated with defibrotide. Thirteen (11.5%) children meeting the modified Seattle criteria for SOS at median day +8 (1-18), had significantly elevated sVEGF-R1 levels on day +14 (489 pg/mL vs. 327 pg/mL, p = 0.007). In contrast. sVEGF-R1 levels in the much broader group of patients (45.1%) meeting EBMT-SOS criteria, including patients with very mild disease, did not overall differ in sVEGF-R1 levels, but higher sVEGF-R1 levels were seen in EBMT-SOS patients with an increased need for diuretic treatment. Importantly, sVEGF-R1 levels were not associated with ES and CLS but were significantly increased on day +30 in patients with grade III-IV aGvHD (OR = 4.2 pr. quartile, p = 0.023). CONCLUSION: VEGF-R1 levels are found to be increased in pediatric patients developing SOS, reflecting the severity of morbidity. sVEGF-R1 were unassociated with both CLS and ES. The potential of sVEGF-R1 as a clinically useful biomarker for SOS should be further explored to improve pre-transplant SOS-risk assessment, SOS-severity grading, and to guide treatment.

Physical capacity in children and adolescents with newly diagnosed cancer: A systematic review and meta-analysis.

OBJECTIVE: To review the body of evidence on cardiorespiratory fitness, muscle strength, and physical performance in children with newly diagnosed cancer, five databases (MEDLINE, Embase, CINAHL, CENTRAL, and Web of Science) were searched on December 19, 2022. METHODS: Thirteen studies, embodying 594 participants within 1 month of cancer diagnosis and 3674 healthy controls were included. Eighteen different outcomes on cardiorespiratory fitness (n = 2), muscle strength (n = 5), physical performance (n = 10), and adverse events (n = 1) were analyzed. RESULTS: Fifteen out of 17 outcomes on physical capacity showed severe impairments compared with healthy controls. Where possible, random-effects meta-analysis was conducted to synthesize the results. No adverse events were reported related to testing. CONCLUSION: Children with cancer have impaired cardiorespiratory fitness, muscle strength, and physical performance within the first month after diagnosis. However, the evidence is based on a small number of studies with large clinical heterogeneity, limiting the certainty of evidence.

Automatic identification of chemical moieties.

In recent years, the prediction of quantum mechanical observables with machine learning methods has become increasingly popular. Message-passing neural networks (MPNNs) solve this task by constructing atomic representations, from which the properties of interest are predicted. Here, we introduce a method to automatically identify chemical moieties (molecular building blocks) from such representations, enabling a variety of applications beyond property prediction, which otherwise rely on expert knowledge. The required representation can either be provided by a pretrained MPNN, or be learned from scratch using only structural information. Beyond the data-driven design of molecular fingerprints, the versatility of our approach is demonstrated by enabling the selection of representative entries in chemical databases, the automatic construction of coarse-grained force fields, as well as the identification of reaction coordinates.

Learning domain invariant representations by joint Wasserstein distance minimization.

Domain shifts in the training data are common in practical applications of machine learning; they occur for instance when the data is coming from different sources. Ideally, a ML model should work well independently of these shifts, for example, by learning a domain-invariant representation. However, common ML losses do not give strong guarantees on how consistently the ML model performs for different domains, in particular, whether the model performs well on a domain at the expense of its performance on another domain. In this paper, we build new theoretical foundations for this problem, by contributing a set of mathematical relations between classical losses for supervised ML and the Wasserstein distance in joint space (i.e. representation and output space). We show that classification or regression losses, when combined with a GAN-type discriminator between domains, form an upper-bound to the true Wasserstein distance between domains. This implies a more invariant representation and also more stable prediction performance across domains. Theoretical results are corroborated empirically on several image datasets. Our proposed approach systematically produces the highest minimum classification accuracy across domains, and the most invariant representation.

Interpolating Nonadiabatic Molecular Dynamics Hamiltonian with Bidirectional Long Short-Term Memory Networks.

Essential for understanding far-from-equilibrium processes, nonadiabatic (NA) molecular dynamics (MD) requires expensive calculations of the excitation energies and NA couplings. Machine learning (ML) can simplify computation; however, the NA Hamiltonian requires complex ML models due to its intricate relationship to atomic geometry. Working directly in the time domain, we employ bidirectional long short-term memory networks (Bi-LSTM) to interpolate the Hamiltonian. Applying this multiscale approach to three metal-halide perovskite systems, we achieve two orders of magnitude computational savings compared to direct ab initio calculation. Reasonable charge trapping and recombination times are obtained with NA Hamiltonian sampling every half a picosecond. The Bi-LSTM-NAMD method outperforms earlier models and captures both slow and fast time scales. In combination with ML force fields, the methodology extends NAMD simulation times from picoseconds to nanoseconds, comparable to charge carrier lifetimes in many materials. Nanosecond sampling is particularly important in systems containing defects, boundaries, interfaces, etc. that can undergo slow rearrangements.

Concept for the Real-Time Monitoring of Molecular Configurations during Manipulation with a Scanning Probe Microscope.

A bold vision in nanofabrication is the assembly of functional molecular structures using a scanning probe microscope (SPM). This approach requires continuous monitoring of the molecular configuration during manipulation. Until now, this has been impossible because the SPM tip cannot simultaneously act as an actuator and an imaging probe. Here, we implement configuration monitoring using experimental data other than images collected during the manipulation process. We model the manipulation as a partially observable Markov decision process (POMDP) and approximate the actual configuration in real time using a particle filter. To achieve this, the models underlying the POMDP are precomputed and organized in the form of a finite-state automaton, allowing the use of complex atomistic simulations. We exemplify the configuration monitoring process and reveal structural motifs behind measured force gradients. The proposed methodology marks an important step toward the piece-by-piece creation of supramolecular structures in a robotic and possibly automated manner.

Gastrointestinal barrier integrity and mucosal inflammation as risk factors of blood stream infections in children treated for acute lymphoblastic leukaemia.

Chemotherapy-induced mucositis increases the risk of blood stream infections (BSI) due to translocation of bacteria across the intestinal epithelium. Our study investigated if quantitative measures of intestinal mucositis severity, including plasma citrulline (a marker of functional enterocytes) and CCL20 (an intestinal immune homeostatic chemokine), could identify patients at risk of BSI. A total of 106 children with ALL undergoing induction treatment (NOPHO ALL 2008) were included and information regarding BSI episodes was collected from the patients' medical records. Twenty-seven patients (25%) developed BSI during induction. Patients with BSI had a larger decrease in citrulline after chemotherapy than patients without BSI, and nearly all BSI episodes (25/27) occurred in the group of patients exhibiting a drop in citrulline (OR = 6.4 [95% CI: 1.4-29.3], P = .008). Patients who developed BSI had higher plasma CCL20 levels on days 8, 15 and 22 than patients without BSI (all P < .05), and elevated CCL20 levels on day 8 increased the risk of subsequent BSI (OR = 1.57 [1.11-2.22] per doubling of CCL20 level, P = .01) in a multivariable logistic regression analysis. These findings suggest that children with ALL who develop BSI during chemotherapy are characterised by more severe intestinal mucositis, as measured by plasma citrulline and CCL20. These markers may be useful in early risk stratification to guide treatment decisions.

Reconstructing Kernel-Based Machine Learning Force Fields with Superlinear Convergence.

Kernel machines have sustained continuous progress in the field of quantum chemistry. In particular, they have proven to be successful in the low-data regime of force field reconstruction. This is because many equivariances and invariances due to physical symmetries can be incorporated into the kernel function to compensate for much larger data sets. So far, the scalability of kernel machines has however been hindered by its quadratic memory and cubical runtime complexity in the number of training points. While it is known that iterative Krylov subspace solvers can overcome these burdens, their convergence crucially relies on effective preconditioners, which are elusive in practice. Effective preconditioners need to partially presolve the learning problem in a computationally cheap and numerically robust manner. Here, we consider the broad class of Nyström-type methods to construct preconditioners based on successively more sophisticated low-rank approximations of the original kernel matrix, each of which provides a different set of computational trade-offs. All considered methods aim to identify a representative subset of inducing (kernel) columns to approximate the dominant kernel spectrum.

Canonical Response Parameterization: Quantifying the structure of responses to single-pulse intracranial electrical brain stimulation.

Single-pulse electrical stimulation in the nervous system, often called cortico-cortical evoked potential (CCEP) measurement, is an important technique to understand how brain regions interact with one another. Voltages are measured from implanted electrodes in one brain area while stimulating another with brief current impulses separated by several seconds. Historically, researchers have tried to understand the significance of evoked voltage polyphasic deflections by visual inspection, but no general-purpose tool has emerged to understand their shapes or describe them mathematically. We describe and illustrate a new technique to parameterize brain stimulation data, where voltage response traces are projected into one another using a semi-normalized dot product. The length of timepoints from stimulation included in the dot product is varied to obtain a temporal profile of structural significance, and the peak of the profile uniquely identifies the duration of the response. Using linear kernel PCA, a canonical response shape is obtained over this duration, and then single-trial traces are parameterized as a projection of this canonical shape with a residual term. Such parameterization allows for dissimilar trace shapes from different brain areas to be directly compared by quantifying cross-projection magnitudes, response duration, canonical shape projection amplitudes, signal-to-noise ratios, explained variance, and statistical significance. Artifactual trials are automatically identified by outliers in sub-distributions of cross-projection magnitude, and rejected. This technique, which we call "Canonical Response Parameterization" (CRP) dramatically simplifies the study of CCEP shapes, and may also be applied in a wide range of other settings involving event-triggered data.