FLOating-Window Projective Separator (FloWPS): A Data Trimming Tool for Support Vector Machines (SVM) to Improve Robustness of the Classifier

Abstract

Here, we propose a heuristic technique of data trimming for SVM termed FLOating Window Projective Separator (FloWPS), tailored for personalized predictions based on molecular data. This procedure can operate with high throughput genetic datasets like gene expression or mutation profiles. Its application prevents SVM from extrapolation by excluding non-informative features. FloWPS requires training on the data for the individuals with known clinical outcomes to create a clinically relevant classifier. The genetic profiles linked with the outcomes are broken as usual into the training and validation datasets. The unique property of FloWPS is that irrelevant features in validation dataset that don’t have significant number of neighboring hits in the training dataset are removed from further analyses. Next, similarly to the k nearest neighbors (kNN) method, for each point of a validation dataset, FloWPS takes into account only the proximal points of the training dataset. Thus, for every point of a validation dataset, the training dataset is adjusted to form a floating window. FloWPS performance was tested on ten gene expression datasets for 992 cancer patients either responding or not on the different types of chemotherapy. We experimentally confirmed by leave-one-out cross-validation that FloWPS enables to significantly increase quality of a classifier built based on the classical SVM in most of the applications, particularly for polynomial kernels.

Flexible Data Trimming Improves Performance of Global Machine Learning Methods in Omics-Based Personalized Oncology

“…We then applied the best settings previously found for BNB, MLP and RF methods using responder-equalized data for the new eleven datasets containing different proportions of treatment responders' and non-responders' samples. In addition, we also used linear SVM method ( Figure 1, Table 3) with penalty parameter C = 1 because our previous results [8] showed that C ≤ 1 minimizes the risk of overtraining for SVM. The output ML classifier quality metrics were obtained for these four methods, including AUC ( Figure 1a The application of FloWPS improved the classifier quality for these four ML methods, as the median AUC for the treatment response classifiers increased from 0.76-0.84 range to 0.83-0.89 ( Figure 1a-d, Table 3).…”

“…In turn, agnostic drug scoring approach, including machine learning (ML) methods can offer even a wider spectrum of opportunities by non-hypothesis-driven direct linkage of specific molecular features with clinical outcomes, such as responsiveness on certain types of treatment [7,8]. ML has a variety of methods that could be used for such agnostic approach, e.g., decision trees, DT [9,10], random forests, RF [11], linear [12], logistic [13], lasso [14,15], and ridge [16] regressions, multi-layer perceptron, MLP [10,17,18], support vectors machines, SVM [9,10,19], adaptive boosting [20][21][22].…”

“…For generating statistically significant predictions, extensive reduction of a pool of features under consideration is needed to make their number comparable with the number of individual samples available [10,[29][30][31]. To leverage the performance of ML in biomedicine, we recently developed an approach called flexible data trimming (Data trimming (DT) is the process of removing or excluding extreme values, or outliers, from a dataset [32]) [8,29,[33][34][35]. This approach is heuristic and based on a common geometrical sense ( Figure 1).…”

“…Thus, for every point of a validation dataset, the training dataset is adjusted to form a floating window. We, therefore, called the respective ML approach, floating window projective separator (FloWPS) [8].…”

“…In a pilot trial of this approach, it significantly enhanced robustness of the SVM classifier in all ten clinical gene expression datasets totally representing 992 cancer patients either responding or not on the different types of chemotherapy [8]. FloWPS demonstrated surprisingly high performance (the ROC (receiver-operator curve) is a widely used graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied.…”