mmostavi / 2_o_me_sitepred

Rep for predicting 2'O sites in RNA sequences with machine learning and deep learning models

Predicting precise location of 2'O methylations with Machine Learning and Deep Learning models

Link to the first paper

Link to the second paper

This repository contains codes and dataset for two papers.

Requirments:

Link to dna2vec
deepexlain for extracting significant scores around 2'O sites

The impact of three key factors are investigated:

• Length of RNA sequences containing 2'O sites
• Embedding method (one-hot encoding vs dna2vec)
• Type of predictive models (SVM and CNN are chosen as the best Machine Learning and Deep Learning models, respectively)

Main results

1. Embedding has no impact on CNN performance however it makes a considerable difference in SVM models performance

2. Increasing input sequence length does not have any impact on performance of both SVM and CNN.

3. Significant nucleotides around 2'O sites are extracted by attention mechanism of CNN

• The location of 2'O motif reported in the nm_seq has got highest attention from CNN
• 2'O site is in the middle (23rd position)