HLMethy: a machine learning-based model to identify the hidden labels of mA candidates.
Journal:
Plant molecular biology
Published Date:
Nov 13, 2019
Abstract
We developed a machine learning-based model to identify the hidden labels of mA candidates from noisy m6A-seq data. Peak-calling approaches, such as MeRIP-seq or mA-seq, are commonly used to map mA modifications. However, these technologies can only map mA sites with 100-200 nt resolution and cannot reveal the precise location or the number of modified residues in a transcript. To address this challenge, we developed a novel machine learning-based approach, named HLMethy, to assign labels to mA candidates from noisy mA-seq data. The multiple instance learning framework was adopted and two different training strategies were used to generate the classification model. To test the performance of our model, the mA sites with single-base resolution were used and our model achieved comparable performance against existing instance-level predictors, which suggest that our model has the potential to improve the data quality of mA-seq at reduced costs. What's more, our generic framework can be extended to other newly found modifications that are found by peak-calling approaches. The source code of HLMethy is available at https://github.com/liuze-nwafu/HLMethy.