@article{LausserSchaeferSchirraetal.2019, author = {Lausser, Ludwig and Sch{\"a}fer, Lisa M. and Schirra, Lyn-Rouven and Szekely, Robin and Schmid, Florian and Kestler, Hans A.}, title = {Assessing phenotype order in molecular data}, volume = {9}, pages = {11746}, journal = {Scientific Reports}, publisher = {Springer Nature}, address = {London}, issn = {2045-2322}, doi = {https://doi.org/10.1038/s41598-019-48150-z}, year = {2019}, abstract = {Biological entities are key elements of biomedical research. Their definition and their relationships are important in areas such as phylogenetic reconstruction, developmental processes or tumor evolution. Hypotheses about relationships like phenotype order are often postulated based on prior knowledge or belief. Evidence on a molecular level is typically unknown and whether total orders are reflected in the molecular measurements is unclear or not assessed. In this work we propose a method that allows a fast and exhaustive screening for total orders in large datasets. We utilise ordinal classifier cascades to identify discriminable molecular representations of the phenotypes. These classifiers are constrained by an order hypothesis and are highly sensitive to incorrect assumptions. Two new error bounds, which are introduced and theoretically proven, lead to a substantial speed-up and allow the application to large collections of many phenotypes. In our experiments we show that by exhaustively evaluating all possible candidate orders, we are able to identify phenotype orders that best coincide with the high-dimensional molecular profiles.}, language = {en} } @article{LausserSzekelyKlimmeketal.2020, author = {Lausser, Ludwig and Szekely, Robin and Klimmek, Attila and Schmid, Florian and Kestler, Hans A.}, title = {Constraining classifiers in molecular analysis}, volume = {17 (2020)}, pages = {20190612}, journal = {Journal of the Royal Society Interface}, subtitle = {invariance and robustness}, number = {163}, publisher = {The Royal Society}, address = {London}, issn = {1742-5662}, doi = {https://doi.org/10.1098/rsif.2019.0612}, year = {2020}, abstract = {Analysing molecular profiles requires the selection of classification models that can cope with the high dimensionality and variability of these data. Also, improper reference point choice and scaling pose additional challenges. Often model selection is somewhat guided by ad hoc simulations rather than by sophisticated considerations on the properties of a categorization model. Here, we derive and report four linked linear concept classes/models with distinct invariance properties for high-dimensional molecular classification. We can further show that these concept classes also form a half-order of complexity classes in terms of Vapnik-Chervonenkis dimensions, which also implies increased generalization abilities. We implemented support vector machines with these properties. Surprisingly, we were able to attain comparable or even superior generalization abilities to the standard linear one on the 27 investigated RNA-Seq and microarray datasets. Our results indicate that a priori chosen invariant models can replace ad hoc robustness analysis by interpretable and theoretically guaranteed properties in molecular categorization.}, language = {en} } @article{LausserSzekelyKestler2020, author = {Lausser, Ludwig and Szekely, Robin and Kestler, Hans A.}, title = {Chained correlations for feature selection}, volume = {14}, journal = {Advances in Data Analysis and Classification}, number = {4}, publisher = {Springer}, address = {Berlin}, issn = {1862-5355}, pages = {871 -- 884}, year = {2020}, abstract = {Data-driven algorithms stand and fall with the availability and quality of existing data sources. Both can be limited in high-dimensional settings (n≫m). For example, supervised learning algorithms designed for molecular pheno- or genotyping are restricted to samples of the corresponding diagnostic classes. Samples of other related entities, such as arise in differential diagnosis, are usually not utilized in this learning scheme. Nevertheless, they might provide domain knowledge on the background or context of the original diagnostic task. In this work, we discuss the possibility of incorporating samples of foreign classes in the training of diagnostic classification models that can be related to the task of differential diagnosis. Especially in heterogeneous data collections comprising multiple diagnostic categories, the foreign ones can change the magnitude of available samples. More precisely, we utilize this information for the internal feature selection process of diagnostic models. We propose the use of chained correlations of original and foreign diagnostic classes. This method allows the detection of intermediate foreign classes by evaluating the correlation between class labels and features for each pair of original and foreign categories. Interestingly, this criterion does not require direct comparisons of the initial diagnostic groups and therefore, might be suitable for settings with restricted data access.}, language = {en} } @article{LausserSzekelySchmidetal.2022, author = {Lausser, Ludwig and Szekely, Robin and Schmid, Florian and Maucher, Markus and Kestler, Hans A.}, title = {Efficient cross-validation traversals in feature subset selection}, volume = {12}, pages = {21485}, journal = {Scientific Reports}, publisher = {Springer Nature}, address = {London}, issn = {2045-2322}, doi = {https://doi.org/10.1038/s41598-022-25942-4}, year = {2022}, abstract = {Sparse and robust classification models have the potential for revealing common predictive patterns that not only allow for categorizing objects into classes but also for generating mechanistic hypotheses. Identifying a small and informative subset of features is their main ingredient. However, the exponential search space of feature subsets and the heuristic nature of selection algorithms limit the coverage of these analyses, even for low-dimensional datasets. We present methods for reducing the computational complexity of feature selection criteria allowing for higher efficiency and coverage of screenings. We achieve this by reducing the preparation costs of high-dimensional subsets O(nm2) to those of one-dimensional ones O(m2). Our methods are based on a tight interaction between a parallelizable cross-validation traversal strategy and distance-based classification algorithms and can be used with any product distance or kernel. We evaluate the traversal strategy exemplarily in exhaustive feature subset selection experiments (perfect coverage). Its runtime, fitness landscape, and predictive performance are analyzed on publicly available datasets. Even in low-dimensional settings, we achieve approximately a 15-fold increase in exhaustively generating distance matrices for feature combinations bringing a new level of evaluations into reach.}, language = {en} } @inproceedings{LausserSzekelyKessleretal.2018, author = {Lausser, Ludwig and Szekely, Robin and Kessler, Viktor and Schwenker, Friedhelm}, title = {Selecting Features from Foreign Classes}, booktitle = {Artificial Neural Networks in Pattern Recognition: 8th IAPR TC3 Workshop, ANNPR 2018, Siena, Italy, September 19-21, 2018: Proceedings}, editor = {Pancioni, Luca and Schwenker, Friedhelm and Trentin, Edmondo}, publisher = {Springer}, address = {Cham}, isbn = {978-3-319-99978-4}, doi = {https://doi.org/10.1007/978-3-319-99978-4_5}, pages = {66 -- 77}, year = {2018}, language = {en} } @article{LausserSzekelyKestler2024, author = {Lausser, Ludwig and Szekely, Robin and Kestler, Hans A.}, title = {Permutation-invariant linear classifiers}, volume = {113}, journal = {Machine Learning}, number = {10}, publisher = {Springer}, address = {Dordrecht}, issn = {1573-0565}, doi = {https://doi.org/10.1007/s10994-024-06561-8}, pages = {7195 -- 7221}, year = {2024}, abstract = {Invariant concept classes form the backbone of classification algorithms immune to specific data transformations, ensuring consistent predictions regardless of these alterations. However, this robustness can come at the cost of limited access to the original sample information, potentially impacting generalization performance. This study introduces an addition to these classes—the permutation-invariant linear classifiers. Distinguished by their structural characteristics, permutation-invariant linear classifiers are unaffected by permutations on feature vectors, a property not guaranteed by other non-constant linear classifiers. The study characterizes this new concept class, highlighting its constant capacity, independent of input dimensionality. In practical assessments using linear support vector machines, the permutation-invariant classifiers exhibit superior performance in permutation experiments on artificial datasets and real mutation profiles. Interestingly, they outperform general linear classifiers not only in permutation experiments but also in permutation-free settings, surpassing unconstrained counterparts. Additionally, findings from real mutation profiles support the significance of tumor mutational burden as a biomarker.}, language = {en} }