探索全球前沿学术脉络

Bridging and analytical validation of the Prosigna(R) Breast Risk of Recurrence Test as a whole-transcriptome NGS lab developed test

Background: The Prosigna Breast Risk of Recurrence test is based on the PAM50 classifier and was originally validated as an in vitro diagnostic (IVD) test on the Dx enabled nCounter(R) Analysis System. The Prosigna test is intended for early-stage, hormone receptor+ (HR+) breast cancer and provides the risk of recurrence (ROR) score (0-100), intrinsic subtype (Luminal A, Luminal B, HER2-enriched, and Basal-like), and the 10-year probability of distant recurrence. We describe the performance of the Prosigna test as a whole transcriptome RNA sequencing laboratory developed test (LDT) for measuring the Prosigna ROR score and intrinsic subtypes on tissue from surgical resection and core needle biopsy as compared to the Prosigna test on the nCounter system. Methods: We evaluated three separate breast cancer cohorts to 1) bridge the IVD test on the nCounter system and NGS LDT test (n = 245), 2) validate the bridged algorithm on an independent biobank sample set (n = 187), and 3) retrospectively test performance on long-term archival samples from a previous study (n = 109). Results: Bridging analysis showed minimal score variability and robust correlation of Prosigna ROR scoring in surgical resections (SR) (2.459, SD; 0.981, R2) and core needle biopsy (CNB) (2.338, SD; 0.970, R2) samples. In the validation set, the Prosigna NGS LDT ROR scores maintained high correlation to the scores of the nCounter system (SR = 0.968, CNB = 0.966, R2), exhibited minimal score variability (SR = 2.488, CNB = 2.558, SD), and demonstrated high concordance in subtype classifications (SR = 92.3% CNB = 92.8%). Further testing demonstrated comparable performance across tumor fractions, a lower limit of detection (LLOD) of 5 ng, and robustness to exogenous ethanol or genomic DNA contamination. When testing previously extracted RNA from the clinical cohort, we observed high correlation (0.974, R2) and low variance (3.078, SD) of ROR scores with original values on the nCounter system, along with strong risk group (95.4%) and subtype (94.5%) concordance. Conclusions: This study describes the analytical validation of the Prosigna NGS-based LDT measuring the Prosigna ROR score and intrinsic subtypes with robust analytical performance on SR and CNB specimens, providing confidence for clinicians utilizing the NGS-based version of this well-established test.

Convex training of Lipschitz-regularized shallow neural networks

arXiv:2606.19652v1 Announce Type: new Abstract: In this work, we introduce a training procedure for shallow neural networks that promotes robustness against adversarial attacks. We solve a non-convex Lipschitz-regularized training program by introducing a convex restriction that can be efficiently solved to global optimality. Our approach can be employed as a post-processing step by taking a pre-trained network as an initial solution to then solving the convex program whose optimal network is guaranteed to be no worse than the initial one. We illustrate the improvements of our training procedure with experiments using real world datasets for regression tasks under an adversarial setting. We show numerically that solving our proposed convex program yields networks with lower objective values on the Lipschitz-regularized program compared to existing methods. Additionally, we show that on certain datasets, networks obtained using our convex training program are both more accurate and robust with respect to adversarial attacks.

SMLMFlow: Improving Structural Resolution in Single Molecule Localization Microscopy with Flow Matching

While Single Molecule Localization Microscopy (SMLM) aims to generate precise coordinates of molecular targets in cells, the resulting point clouds are inherently blurred by additive noise sources across the experimental, imaging, and processing workflow. This blurring often limits SMLM's ability to accurately quantify complex assembled structures required to address biological issues, despite reported localization precision down to a couple of nanometers. Here, we present SMLMFlow, a machine learning framework for improving structural resolution in SMLM datasets that combines a graph neural network and a hierarchical transformer with flow matching. We show that SMLMFlow improves structural resolution and downstream quantification across different structures, including filaments and protein nano-clusters, and generalizes to new unseen photophysics models.