ArXiv最新文献

Quantitative descriptors convert high-dimensional medical images into low-dimensional features capable of differentiating organ shapes that correlate with injury or disease progression for diagnostic purposes. An important example is aortic dissections, which can be imaged using high-resolution CT scans and for which the shape of the true and false lumens of the aorta has long been used to predict disease state and the potential for positive surgical outcomes (namely thoracic endovascular repair or TEVAR). Here we present a method for calculating the topographical pair correlation function (TPCF), a descriptor of the spatial correlation of point estimates for Gaussian curvature, mean curvature, shape index, and bending ratio constrained to the surface of a meshed image. We used the TPCF as a metric to describe aortic shape and extracted quantitative features from the resulting curves. When the TPCF was parameterized by shape index, the area under the curve of the correlation function contributed to a classification accuracy of 0.95 for disease presence and/or impending TEVAR success. Comparison with single-point statistics suggests that the TPCF provides powerful features for classifying the disease state of aortas and more broadly in capturing structural correlations in anatomical data.

Representational Similarity Analysis (RSA) is a popular method for analyzing neuroimaging and behavioral data. Here we evaluate the accuracy and reliability of RSA in the context of model selection, and compare it to that of regression. Although RSA offers flexibility in handling high-dimensional, cross-modal, and cross-species data, its reliance on a transformation of raw data into similarity structures may result in the loss of critical stimulus-response information. Across extensive simulation studies and empirical analyses, we show that RSA leads to lower model selection accuracy, regardless of sample size, noise level, feature dimensionality, or multicollinearity, relative to regression. While principal component analysis and feature reweighting mitigate RSA's deficits driven by multicollinearity, regression remains superior in accurately distinguishing between models. Empirical data and a follow-up fMRI simulation further support these conclusions. Our findings suggest that researchers should carefully consider which approach to use: RSA is less effective than linear regression for model selection and fitting when direct stimulus-response mappings are available.

Bayesian spatial modeling provides a flexible framework for whole-brain fMRI analysis by explicitly incorporating spatial dependencies, overcoming the limitations of traditional massive univariate approaches that lead to information waste. In this work, we introduce SIMBA, a Scalable Image Modeling using a Bayesian Approach, for group-level fMRI analysis, which places Gaussian process (GP) priors on spatially varying functions to capture smooth and interpretable spatial association patterns across the brain volume. To address the significant computational challenges of GP inference in high-dimensional neuroimaging data, we employ a low-rank kernel approximation that enables projection into a reduced-dimensional subspace. This allows for efficient posterior computation without sacrificing spatial resolution, and we have developed efficient algorithms for this implemented in Python that achieve fully Bayesian inference either within minutes using the Gibbs sampler or within seconds using mean-field variational inference (VI). Through extensive simulation studies, we first show that SIMBA outperforms competing methods in estimation accuracy, activation detection sensitivity, and uncertainty quantification, especially in low signal-to-noise settings. We further demonstrate the scalability and interpretability of SIMBA in large-scale task-based fMRI applications, analyzing both volumetric and cortical surface data from the NARPS and ABCD studies.

Motivation: Low-complexity (LC) DNA sequences are compositionally repetitive sequences that are often associated with spurious homologous matches and variant calling artifacts. While algorithms for identifying LC sequences exist, they either do not define complexity mathematically or are inefficient with long or variable context windows.

Results: Longdust is a new algorithm that efficiently identifies long LC sequences including centromeric satellite and tandem repeats with moderately long motifs. It defines string complexity by statistically modeling the $k$ -mer count distribution with the parameters: the $k$ -mer length, the context window size and a threshold on complexity. Longdust exhibits high performance on real data and high consistency with existing methods.

Availability and implementation: https://github.com/lh3/longdust.

The theranostic paradigm enables personalization of treatment by selecting patients with a diagnostic radiopharmaceutical and monitoring therapy using a matched therapeutic isotope. This strategy relies on accurate image reconstruction of both pretherapy and post-therapy images for patient selection and monitoring treatment. However, traditional reconstruction methods are hindered by challenges such as crosstalk in multi-isotope imaging and extremely low-count measurements data when imaging of alpha-emitting isotopes. Additionally, to fully realize the benefits of new imaging systems being developed for theranostic applications, advanced reconstruction techniques are needed. This review highlights recent progress and discusses critical challenges and unmet needs in theranostic image reconstruction.

The PTPN11 gene encodes the Src homology 2 domain-containing protein tyrosine phosphatase (SHP2), a key regulator of cell growth, differentiation, and apoptosis through its modulation of various signaling pathways, including the RAS/MAPK signaling pathway. Missense variants in PTPN11 disrupt SHP2's proper catalytic activity and the regulation of signaling pathways, leading to disorders such as Noonan syndrome (NS), LEOPARD syndrome (LS), or juvenile myelomonocytic leukemia (JMML). These missense variants have molecular disruptions resulting in gains and losses of function at both the molecular and phenotypic levels. Depending on their location within SHP2, missense substitutions disrupt inter-domain regulation or impair phosphatase function, resulting in altered phosphatase activity. In this study, we investigate the molecular basis underlying the differential pathogenicity of PTPN11 missense variants and predict the structural consequences of these variants using MutPred2 and AlphaFold2. We find that LOF and GOF variants display distinct functional mechanisms in sodium and DNA binding, and that NS-associated missense variants identified in fetuses with ultrasound-detected anomalies and familiar cases are more likely to be pathogenic.

Primates utilize distributed neural circuits to learn habits in uncertain environments, but the underlying mechanisms remain poorly understood. We propose a formal theory of network energetics explaining how brain states influence sequential behavior. We test our theory on multi-unit recordings from the caudate nucleus and cortical regions of macaques performing a motor habit task. The theory predicts the energy required to transition between brain states represented by trial-specific firing rates across channels, assuming activity spreads through effective connections. We hypothesized that habit formation would correlate with lower control energy. Consistent with this, we observed smaller energy requirements for transitions between similar saccade patterns and those of intermediate complexity, and sessions exploiting fewer patterns. Simulations ruled out confounds from neurons' directional tuning. Finally, virtual lesioning demonstrated robustness of observed relationships between control energy and behavior. This work paves the way for examining how behavior arises from changing activity in distributed circuitry.

Alzheimer's disease (AD) is characterized by the progressive spread of pathology across brain networks, yet forecasting this cascade at the individual level remains challenging. We present a personalized graph-based dynamical model that captures the spatiotemporal evolution of cortical atrophy from longitudinal MRI and PET data. The approach constructs individualized brain graphs and learns the dynamics driving regional neurodegeneration. Applied to 1,891 participants from the Alzheimer's Disease Neuroimaging Initiative, the model accurately predicts key AD biomarkers-including amyloid- $\beta$ , tau, neurodegeneration, and cognition-outperforming clinical and neuroimaging benchmarks. Patient-specific parameters reveal distinct progression subtypes and anticipate future cognitive decline more effectively than standard biomarkers. Sensitivity analysis highlights regional drivers of disease spread, reproducing known temporolimbic and frontal vulnerability patterns. This network-based digital-twin framework offers a quantitative, personalized paradigm for AD trajectory prediction, with implications for patient stratification, clinical trial design, and targeted therapeutic development.

Generative machine learning models for exploring chemical space have shown immense promise, but many molecules they generate are too difficult to synthesize, making them impractical for further investigation or development. In this work, we present a novel approach by fine-tuning Meta's Llama3 Large Language Models (LLMs) to create SynLlama, which generates full synthetic pathways made of commonly accessible building blocks and robust organic reaction templates. SynLlama explores a large synthesizable space using significantly less data, and offers strong performance in both forward and bottom-up synthesis planning compared to other state-of-the-art methods. We find that SynLlama, even without training on external building blocks, can effectively generalize to unseen yet purchasable building blocks, meaning that its reconstruction capabilities extend to a broader synthesizable chemical space than the training data. We also demonstrate the use of SynLlama in a pharmaceutical context for synthesis planning of analog molecules and hit expansion leads for proposed inhibitors of target proteins, offering medicinal chemists a valuable tool for discovery.

Nearly all cell models explicitly or implicitly deal with the biophysical constraints that must be respected for life to persist. Despite this, there is almost no systematicity in how these constraints are implemented, and we lack a principled understanding of how cellular dynamics interact with them and how they originate in actual biology. Computational cell biology will only overcome these concerns once it treats the life-death boundary as a central concept, creating a theory of cellular viability. We lay the foundation for such a development by demonstrating how specific geometric structures can separate regions of qualitatively similar survival outcomes in our models, offering new global organizing principles for cell fate. We also argue that idealized models of emergent individuals offer a tractable way to begin understanding life's intrinsically generated limits.