AdaSampling for Positive-Unlabeled and Label Noise Learning With Bioinformatics Applications

Abstract

Class labels are required for supervised learning but may be corrupted or missing in various applications. In binary classification, for example, when only a subset of positive instances is labeled whereas the remaining are unlabeled, positive-unlabeled (PU) learning is required to model from both positive and unlabeled data. Similarly, when class labels are corrupted by mislabeled instances, methods are needed for learning in the presence of class label noise (LN). Here we propose adaptive sampling (AdaSampling), a framework for both PU learning and learning with class LN. By iteratively estimating the class mislabeling probability with an adaptive sampling procedure, the proposed method progressively reduces the risk of selecting mislabeled instances for model training and subsequently constructs highly generalizable models even when a large proportion of mislabeled instances is present in the data. We demonstrate the utilities of proposed methods using simulation and benchmark data, and compare them to alternative approaches that are commonly used for PU learning and/or learning with LN. We then introduce two novel bioinformatics applications where AdaSampling is used to: 1) identify kinase-substrates from mass spectrometry-based phosphoproteomics data and 2) predict transcription factor target genes by integrating various next-generation sequencing data.

PhosR enables processing and functional analysis of phosphoproteomic data

“…To this end, PhosR implements a multi-step kinase-substrate scoring method where first the likelihood of a kinase to regulate a phosphosite is scored by combining both kinase recognition motifs and the dynamic phosphorylation profiles of sites. The combined scores across all kinases are then integrated using an adaptive sampling-based positiveunlabelled learning method (Yang et al, 2018) to prioritise the kinase most likely to regulate a phosphosite (see Methods). The application of the proposed scoring method to the myotube phosphoproteome uncovers potential kinase-substrate pairs ( Figure 3A; row dendrogram) and global relationships between kinases ( Figure 3A; column dendrogram).…”

“…To identify potential kinases that could be responsible for the phosphorylation change of a phosphorylation site, we implement a multi-step framework that contains two major components including (i) a kinaseSubstrateScore function which scores a given phosphosite using kinase recognition motif and phosphoproteomic dynamics, and (ii) a kinaseSubstratePred function which synthesise the scores generated from (i) for predicting kinase-substrate relationships using an adaptive sampling-based positiveunlabelled learning method (Yang et al, 2018). The kinase-substrate scoring function combines both kinase recognition motif (i.e.…”

“…This is difficult for most kinases because the numbers of known substrates are prohibitively small for training predictive models. To this end, we implemented in PhosR the AdaSampling-based positive-unlabelled ensemble of support vector machines (SVMs) as described previously (Yang et al, 2018). For each kinase the top 30 highly ranked phosphosites (based on the combined scores) are used initially as positive examples for training SVMs for predicting substrates of that kinase and the AdaSampling procedure is used to subsequently update the training examples based on the model confidence on each phosphosite.…”

Probing lncRNA–Protein Interactions: Data Repositories, Models, and Algorithms

“…Positive-unlabeled (PU) learning (de Campos et al, 2018;Sansone et al, 2018;Yang et al, 2018) is applied to various situations. In PU learning, a supervised learning-based method is designed to learn a classification model from a positive sample set and an unlabeled dataset from an unknown class.…”

“…In PU learning, a supervised learning-based method is designed to learn a classification model from a positive sample set and an unlabeled dataset from an unknown class. Yang et al (2018) designed an adaptive sampling framework with class label noise based on PU learning and introduced two new bioinformatic applications: identifying kinase-substrates and identifying transcription factor target genes. Therefore, PU learning may be one strong way to solve the problem of lacking negative LPIs.…”