University of Tübingen

About: University of Tübingen is a education organization based out in Tübingen, Germany. It is known for research contribution in the topics: Population & Transplantation. The organization has 40555 authors who have published 84108 publications receiving 3015320 citations. The organization is also known as: Eberhard Karls University & Eberhard-Karls-Universität Tübingen.

Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Models

Abstract: Many machine learning algorithms are vulnerable to almost imperceptible perturbations of their inputs. So far it was unclear how much risk adversarial perturbations carry for the safety of real-world machine learning applications because most methods used to generate such perturbations rely either on detailed model information (gradient-based attacks) or on confidence scores such as class probabilities (score-based attacks), neither of which are available in most real-world scenarios. In many such cases one currently needs to retreat to transfer-based attacks which rely on cumbersome substitute models, need access to the training data and can be defended against. Here we emphasise the importance of attacks which solely rely on the final model decision. Such decision-based attacks are (1) applicable to real-world black-box models such as autonomous cars, (2) need less knowledge and are easier to apply than transfer-based attacks and (3) are more robust to simple defences than gradient- or score-based attacks. Previous attacks in this category were limited to simple models or simple datasets. Here we introduce the Boundary Attack, a decision-based attack that starts from a large adversarial perturbation and then seeks to reduce the perturbation while staying adversarial. The attack is conceptually simple, requires close to no hyperparameter tuning, does not rely on substitute models and is competitive with the best gradient-based attacks in standard computer vision tasks like ImageNet. We apply the attack on two black-box algorithms from this http URL. The Boundary Attack in particular and the class of decision-based attacks in general open new avenues to study the robustness of machine learning models and raise new questions regarding the safety of deployed machine learning systems. An implementation of the attack is available as part of Foolbox at this https URL .

Cancer classification using the Immunoscore: a worldwide task force

Abstract: Prediction of clinical outcome in cancer is usually achieved by histopathological evaluation of tissue samples obtained during surgical resection of the primary tumor. Traditional tumor staging (AJCC/UICC-TNM classification) summarizes data on tumor burden (T), presence of cancer cells in draining and regional lymph nodes (N) and evidence for metastases (M). However, it is now recognized that clinical outcome can significantly vary among patients within the same stage. The current classification provides limited prognostic information, and does not predict response to therapy. Recent literature has alluded to the importance of the host immune system in controlling tumor progression. Thus, evidence supports the notion to include immunological biomarkers, implemented as a tool for the prediction of prognosis and response to therapy. Accumulating data, collected from large cohorts of human cancers, has demonstrated the impact of immune-classification, which has a prognostic value that may add to the significance of the AJCC/UICC TNM-classification. It is therefore imperative to begin to incorporate the 'Immunoscore' into traditional classification, thus providing an essential prognostic and potentially predictive tool. Introduction of this parameter as a biomarker to classify cancers, as part of routine diagnostic and prognostic assessment of tumors, will facilitate clinical decision-making including rational stratification of patient treatment. Equally, the inherent complexity of quantitative immunohistochemistry, in conjunction with protocol variation across laboratories, analysis of different immune cell types, inconsistent region selection criteria, and variable ways to quantify immune infiltration, all underline the urgent requirement to reach assay harmonization. In an effort to promote the Immunoscore in routine clinical settings, an international task force was initiated. This review represents a follow-up of the announcement of this initiative, and of the J Transl Med. editorial from January 2012. Immunophenotyping of tumors may provide crucial novel prognostic information. The results of this international validation may result in the implementation of the Immunoscore as a new component for the classification of cancer, designated TNM-I (TNM-Immune).

Patients with IDH1 wild type anaplastic astrocytomas exhibit worse prognosis than IDH1-mutated glioblastomas, and IDH1 mutation status accounts for the unfavorable prognostic effect of higher age: Implications for classification of gliomas

Abstract: WHO grading of human brain tumors extends beyond a strictly histological grading system by providing a basis predictive for the clinical behavior of the respective neoplasm. For example, patients with glioblastoma WHO grade IV usually show a less favorable clinical course and receive more aggressive first-line treatment than patients with anaplastic astrocytoma WHO grade III. Here we provide evidence that the IDH1 status is more prognostic for overall survival than standard histological criteria that differentiate high-grade astrocytomas. We sequenced the isocitrate dehydrogenase 1 gene (IDH1) at codon 132 in 382 patients with anaplastic astrocytoma and glioblastoma from the NOA-04 trial and from a prospective translational cohort study of the German Glioma Network. Patients with anaplastic astrocytomas carried IDH1 mutations in 60%, and patients with glioblastomas in 7.2%. IDH1 was the most prominent single prognostic factor (RR 2.7; 95% CI 1.6–4.5) followed by age, diagnosis and MGMT. The sequence from more favorable to poorer outcome was (1) anaplastic astrocytoma with IDH1 mutation, (2) glioblastoma with IDH1 mutation, (3) anaplastic astrocytoma without IDH1 mutation and (4) glioblastoma without IDH1 mutation (p < 0.0001). In this combined set of anaplastic astrocytomas and glioblastomas both, IDH1 mutation and IDH1 expression status were of greater prognostic relevance than histological diagnosis according to the current WHO classification system. Our data indicate that much of the prognostic significance of patient age is due to the predominant occurrence of IDH1 mutations in younger patients. Immunohistochemistry using a mutation-specific antibody recognizing the R132H mutation yielded similar results. We propose to complement the current WHO classification and grading of high-grade astrocytic gliomas by the IDH1 mutation status and to use this combined histological and molecular classification in future clinical trials.

Intracytoplasmic Sperm Injection May Increase the Risk of Imprinting Defects

Abstract: In germ cells and the early embryo, the mammalian genome undergoes widespread epigenetic reprogramming. Animal studies suggest that this process is vulnerable to external factors. We report two children who were conceived by intracytoplasmic sperm injection (ICSI) and who developed Angelman syndrome. Molecular studies, including DNA methylation and microsatellite and quantitative Southern blot analysis, revealed a sporadic imprinting defect in both patients. We discuss the possibility that ICSI may interfere with the establishment of the maternal imprint in the oocyte or pre-embryo.

The genetic history of Ice Age Europe

Abstract: Modern humans arrived in Europe ~45,000 years ago, but little is known about their genetic composition before the start of farming ~8,500 years ago. Here we analyse genome-wide data from 51 Eurasians from ~45,000-7,000 years ago. Over this time, the proportion of Neanderthal DNA decreased from 3-6% to around 2%, consistent with natural selection against Neanderthal variants in modern humans. Whereas there is no evidence of the earliest modern humans in Europe contributing to the genetic composition of present-day Europeans, all individuals between ~37,000 and ~14,000 years ago descended from a single founder population which forms part of the ancestry of present-day Europeans. An ~35,000-year-old individual from northwest Europe represents an early branch of this founder population which was then displaced across a broad region, before reappearing in southwest Europe at the height of the last Ice Age ~19,000 years ago. During the major warming period after ~14,000 years ago, a genetic component related to present-day Near Easterners became widespread in Europe. These results document how population turnover and migration have been recurring themes of European prehistory.