scispace - formally typeset
Search or ask a question
Author

Frederick Klauschen

Bio: Frederick Klauschen is an academic researcher from Humboldt University of Berlin. The author has contributed to research in topics: Medicine & Cancer. The author has an hindex of 45, co-authored 170 publications receiving 12451 citations. Previous affiliations of Frederick Klauschen include Charité & National Institutes of Health.


Papers
More filters
Journal ArticleDOI
10 Jul 2015-PLOS ONE
TL;DR: This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of nonlinear classifiers by introducing a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks.
Abstract: Understanding and interpreting classification decisions of automated image classification systems is of high value in many applications, as it allows to verify the reasoning of the system and provides additional information to the human expert. Although machine learning methods are solving very successfully a plethora of tasks, they have in most cases the disadvantage of acting as a black box, not providing any information about what made them arrive at a particular decision. This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of nonlinear classifiers. We introduce a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks. These pixel contributions can be visualized as heatmaps and are provided to a human expert who can intuitively not only verify the validity of the classification decision, but also focus further analysis on regions of potential interest. We evaluate our method for classifiers trained on PASCAL VOC 2009 images, synthetic image data containing geometric shapes, the MNIST handwritten digits data set and for the pre-trained ImageNet model available as part of the Caffe open source package.

3,330 citations

Journal ArticleDOI
TL;DR: Current data on the clinical validity and utility of TILs in BC are reviewed in an effort to foster better knowledge and insight in this rapidly evolving field, and to develop a standardized methodology for visual assessment on H&E sections.

1,971 citations

Journal ArticleDOI
TL;DR: The hypothesis that breast cancer is immunogenic and might be targetable by immune-modulating therapies is supported and increased TIL concentration predicted response to neoadjuvant chemotherapy in all molecular subtypes assessed.
Abstract: Summary Background Tumour-infiltrating lymphocytes (TILs) are predictive for response to neoadjuvant chemotherapy in triple-negative breast cancer (TNBC) and HER2-positive breast cancer, but their role in luminal breast cancer and the effect of TILs on prognosis in all subtypes is less clear. Here, we assessed the relevance of TILs for chemotherapy response and prognosis in patients with TNBC, HER2-positive breast cancer, and luminal–HER2-negative breast cancer. Methods Patients with primary breast cancer who were treated with neoadjuvant combination chemotherapy were included from six randomised trials done by the German Breast Cancer Group. Pretherapeutic core biopsies from 3771 patients included in these studies were assessed for the number of stromal TILs by standardised methods according to the guidelines of the International TIL working group. TILs were analysed both as a continuous parameter and in three predefined groups of low (0–10% immune cells in stromal tissue within the tumour), intermediate (11–59%), and high TILs (≥60%). We used these data in univariable and multivariable statistical models to assess the association between TIL concentration and pathological complete response in all patients, and between the amount of TILs and disease-free survival and overall survival in 2560 patients from five of the six clinical trial cohorts. Findings In the luminal–HER2-negative breast cancer subtype, a pathological complete response (pCR) was achieved in 45 (6%) of 759 patients with low TILs, 48 (11%) of 435 with intermediate TILs, and 49 (28%) of 172 with high TILs. In the HER2-positive subtype, pCR was observed in 194 (32%) of 605 patients with low TILs, 198 (39%) of 512 with intermediate TILs, and 127 (48%) of 262 with high TILs. Finally, in the TNBC subtype, pCR was achieved in 80 (31%) of 260 patients with low TILs, 117 (31%) of 373 with intermediate TILs, and 136 (50%) of 273 with high TILs (p 2 test for trend). In the univariable analysis, a 10% increase in TILs was associated with longer disease-free survival in TNBC (hazard ratio [HR] 0·93 [95% CI 0·87–0·98], p=0·011) and HER2-positive breast cancer (0·94 [0·89–0·99], p=0·017), but not in luminal–HER2-negative tumours (1·02 [0·96–1·09], p=0·46). The increase in TILs was also associated with longer overall survival in TNBC (0·92 [0·86–0·99], p=0·032), but had no association in HER2-positive breast cancer (0·94 [0·86–1·02], p=0·11), and was associated with shorter overall survival in luminal–HER2-negative tumours (1·10 [1·02–1·19], p=0·011). Interpretation Increased TIL concentration predicted response to neoadjuvant chemotherapy in all molecular subtypes assessed, and was also associated with a survival benefit in HER2-positive breast cancer and TNBC. By contrast, increased TILs were an adverse prognostic factor for survival in luminal–HER2-negative breast cancer, suggesting a different biology of the immunological infiltrate in this subtype. Our data support the hypothesis that breast cancer is immunogenic and might be targetable by immune-modulating therapies. In light of the results in luminal breast cancer, further research investigating the interaction of the immune system with different types of endocrine therapy is warranted. Funding Deutsche Krebshilfe and European Commission.

1,154 citations

Journal ArticleDOI
14 Dec 2012-PLOS ONE
TL;DR: The functionality of Cutoff Finder is illustrated by the analysis of the gene expression of estrogen receptor and progesterone receptor in breast cancer tissues, which is analyzed and correlated with immunohistologically determined ER status and distant metastasis free survival.
Abstract: Gene or protein expression data are usually represented by metric or at least ordinal variables. In order to translate a continuous variable into a clinical decision, it is necessary to determine a cutoff point and to stratify patients into two groups each requiring a different kind of treatment. Currently, there is no standard method or standard software for biomarker cutoff determination. Therefore, we developed Cutoff Finder, a bundle of optimization and visualization methods for cutoff determination that is accessible online. While one of the methods for cutoff optimization is based solely on the distribution of the marker under investigation, other methods optimize the correlation of the dichotomization with respect to an outcome or survival variable. We illustrate the functionality of Cutoff Finder by the analysis of the gene expression of estrogen receptor (ER) and progesterone receptor (PgR) in breast cancer tissues. This distribution of these important markers is analyzed and correlated with immunohistologically determined ER status and distant metastasis free survival. Cutoff Finder is expected to fill a relevant gap in the available biometric software repertoire and will enable faster optimization of new diagnostic biomarkers. The tool can be accessed at http://molpath.charite.de/cutoff.

934 citations

Journal ArticleDOI
TL;DR: Investigating the immunogenicity of human epidermal growth factor receptor 2 -positive and triple-negative breast cancers and immunologically relevant genes in the neoadjuvant GeparSixto trial found immunologic factors were highly significant predictors of therapy response in the trial, particularly in patients treated with Cb.
Abstract: Purpose Modulation of immunologic interactions in cancer tissue is a promising therapeutic strategy. To investigate the immunogenicity of human epidermal growth factor receptor 2 (HER2) –positive and triple-negative (TN) breast cancers (BCs), we evaluated tumor-infiltrating lymphocytes (TILs) and immunologically relevant genes in the neoadjuvant GeparSixto trial. Patients and Methods GeparSixto investigated the effect of adding carboplatin (Cb) to an anthracycline-plus-taxane combination (PM) on pathologic complete response (pCR). A total of 580 tumors were evaluated before random assignment for stromal TILs and lymphocyte-predominant BC (LPBC). mRNA expression of immune-activating (CXCL9, CCL5, CD8A, CD80, CXCL13, IGKC, CD21) as well as immunosuppressive factors (IDO1, PD-1, PD-L1, CTLA4, FOXP3) was measured in 481 tumors. Results Increased levels of stromal TILs predicted pCR in univariable (P < .001) and multivariable analyses (P < .001). pCR rate was 59.9% in LPBC and 33.8% for non-LPBC (P < .001). pC...

822 citations


Cited by
More filters
Proceedings Article
04 Dec 2017
TL;DR: In this article, a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations), is presented, which assigns each feature an importance value for a particular prediction.
Abstract: Understanding why a model makes a certain prediction can be as crucial as the prediction's accuracy in many applications. However, the highest accuracy for large modern datasets is often achieved by complex models that even experts struggle to interpret, such as ensemble or deep learning models, creating a tension between accuracy and interpretability. In response, various methods have recently been proposed to help users interpret the predictions of complex models, but it is often unclear how these methods are related and when one method is preferable over another. To address this problem, we present a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations). SHAP assigns each feature an importance value for a particular prediction. Its novel components include: (1) the identification of a new class of additive feature importance measures, and (2) theoretical results showing there is a unique solution in this class with a set of desirable properties. The new class unifies six existing methods, notable because several recent methods in the class lack the proposed desirable properties. Based on insights from this unification, we present new methods that show improved computational performance and/or better consistency with human intuition than previous approaches.

7,309 citations

Journal ArticleDOI
TL;DR: This review summarizes the clinical efficacy, perspectives, and future challenges of using PD-1/PD-L1-directed antibodies in the treatment of breast cancer.
Abstract: Immune checkpoint inhibition represents a major recent breakthrough in the treatment of malignant diseases including breast cancer. Blocking the programmed death receptor-1 (PD-1) and its ligand, PD-L1, has shown impressive antitumor activity and may lead to durable long-term disease control, especially in the triple-negative subtypes of breast cancer (TNBC). Although immune checkpoint blockade is generally well tolerated, specific immune-related adverse events (irAEs) may occur. This review summarizes the clinical efficacy, perspectives, and future challenges of using PD-1/PD-L1-directed antibodies in the treatment of breast cancer.

5,777 citations

Journal ArticleDOI
TL;DR: The Perseus software platform was developed to support biological and biomedical researchers in interpreting protein quantification, interaction and post-translational modification data and it is anticipated that Perseus's arsenal of algorithms and its intuitive usability will empower interdisciplinary analysis of complex large data sets.
Abstract: A main bottleneck in proteomics is the downstream biological analysis of highly multivariate quantitative protein abundance data generated using mass-spectrometry-based analysis. We developed the Perseus software platform (http://www.perseus-framework.org) to support biological and biomedical researchers in interpreting protein quantification, interaction and post-translational modification data. Perseus contains a comprehensive portfolio of statistical tools for high-dimensional omics data analysis covering normalization, pattern recognition, time-series analysis, cross-omics comparisons and multiple-hypothesis testing. A machine learning module supports the classification and validation of patient groups for diagnosis and prognosis, and it also detects predictive protein signatures. Central to Perseus is a user-friendly, interactive workflow environment that provides complete documentation of computational methods used in a publication. All activities in Perseus are realized as plugins, and users can extend the software by programming their own, which can be shared through a plugin store. We anticipate that Perseus's arsenal of algorithms and its intuitive usability will empower interdisciplinary analysis of complex large data sets.

5,165 citations

01 Jan 2006

3,012 citations

Journal ArticleDOI
TL;DR: In this paper, a taxonomy of recent contributions related to explainability of different machine learning models, including those aimed at explaining Deep Learning methods, is presented, and a second dedicated taxonomy is built and examined in detail.

2,827 citations