We propose a new fast algorithm to estimate any sparse generalized linear model with convex or non-convex separable penalties. Our algorithm is able to solve problems with millions of samples and features in seconds, by relying on coordinate descent, working sets and Anderson acceleration. It handles previously unaddressed models, and is extensively shown to improve state-of-art algorithms. We provide a flexible, scikit-learn compatible package, which easily handles customized datafits and penalties.
Estimators based on non-convex sparsity-promoting penalties were shown to yield state-of-the-art solutions to the magneto-/electroencephalography (M/EEG) brain source localization problem. In this paper we tackle the model selection problem of these estimators: we propose to use a proxy of the Stein's Unbiased Risk Estimator (SURE) to automatically select their regularization parameters. The effectiveness of the method is demonstrated on realistic simulations and 30 subjects from the Cam-CAN dataset. To our knowledge, this is the first time that sparsity promoting estimators are automatically calibrated at such a scale. Results show that the proposed SURE approach outperforms cross-validation strategies and state-of-the-art Bayesian statistics methods both computationally and statistically.
e16580 Background: Gain of function mutations of the FGFR3 gene have been reported to be important driver mutations in a subset of muscle-invasive bladder urothelial cancer (MIBC). Subsequently, gene specific kinase inhibitors (e.g., erdafitinib) have been developed to target FGFR3 mutant metastasized MIBC with promising clinical activity. FGFR3 mutational status testing is required for erdafitinib treatment, but FGFR3 mutations are rare in metastatic MIBC (10-15%). Thus, cheap and fast pre-screening tools such as artificial intelligence (AI) based image classifiers could help to reduce turnover times of FGFR3 mutational testing and save costs arising from universal mutational testing. Herein, we propose an AI-based approach to identify MIBC FGFR3 mutated patients on routine histological slides. Methods: Slides from 612 patients with MIBC from two different sources (CCC-EMN Erlangen cohort n = 239, mutant = 20; TCGA-BLCA n = 373, mutant = 50) were used. The CCC-EMN cohort was used as a discovery set while TCGA-BLCA was used as an independent validation set. A two-step AI model was used to 1) extract features from each slide and 2) predict FGFR3 status from these features. The AUC, sensitivity and specificity were used as metrics to assess the performance of our model. Most predictive regions of FGFR3 mutation were reviewed by an expert pathologist. Results: The model reached an AUC/Sensitivity/Specificity of 0.82 [0.76-0.87] / 0.90 [0.82-0.97] / 0.57 [0.53-0.62] on TCGA-BLCA, substantially improving upon results previously reported on the same cohort (Loeffler et al. 2021: AUC = 0.70). Most predictive regions of FGFR3 mutation revealed monomorphous conventional urothelial tumor morphology and low desmoplastic tumor stroma content, while least predictive regions contained pleomorphic tumor morphology and high amounts of desmoplastic tumor content. Conclusions: By ruling out 57% of MIBC FGFR3 wild type tumors with high sensitivity, our model is the first to reach valuable performance for clinical use. Consequently, this study paves the way for the development of FGFR3 pre-screening tools to ease the management of patients with MIBC.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.