M. A. Basher, A. R. | bioRxiv Science

bioRxiv2020

Multi-label pathway prediction based on active dataset subsampling

Metabolic pathways are composed of reaction sequences catalyzed by enzymes. The set of reactions within and between cells comprises a reactome. Pathways and reactomes can be predicted from organismal or multi-organismal genomes using rule-based or machine learning methods. While machine learning methods overcome issues of probability and scale associated with rule-based methods, several complications remain that can degrade performance including inadequately labeled training data, missing feature information, and inherent imbalances in the distribution of pathways within a dataset. Here, we present leADS (multi-label learning based on active dataset subsampling), a machine learning method, that uses subsampling to reduce the negative impact of training loss due to class imbalance. We demonstrate leADs performance using organismal and multi-organismal datasets in relation to other machine learning pathway prediction methods. Availability and implementationleADS is available under the GNU license at github.com/hallamlab/leADS. A wiki, including a tutorial, is available at github.com//hallamlab/leADS/wiki Contactshallam@mail.ubc.ca

bioinformatics↗

bioRxiv2020

reMap: Relabeling Multi-label Pathway Data with Bags to Enhance Predictive Performance

Metabolic pathway inference from genomic sequence information is an integral scientific problem with wide ranging applications in the life sciences. As sequencing throughput increases, scalable and performative methods for pathway prediction at different levels of genome complexity and completion become compulsory. In this paper, we present reMap (relabeling metabolic pathway data with groups) a simple, and yet, generic framework, that performs relabeling examples to a different set of labels, characterized as groups. A pathway group is comprised of a subset of statistically correlated pathways that can be further distributed between multiple pathway groups. This has important implications for pathway prediction, where a learning algorithm can revisit a pathway multiple times across groups to improve sensitivity. The relabeling process in reMap is achieved through an alternating feedback process. In the first feed-forward phase, a minimal subset of pathway groups is picked to label each example. In the second feed-backward phase, reMaps internal parameters are updated to increase the accuracy of mapping examples to pathway groups. The resulting pathway group dataset is then be used to train a multi-label learning algorithm. reMaps effectiveness was evaluated on metabolic pathway prediction where resulting performance metrics equaled or exceeded other prediction methods on organismal genomes with improved predictive performance.

bioinformatics↗