Identification of RNA Pseudouridine Sites using Deep Learning Approaches

Supervised By

Dr. Md. Al Mamun

Professor

Dept. of Computer Science & Engineering

Rajshahi University of Engineering & Technology

​

​

Presented By

Abu Zahid Bin Aziz

Roll: 1503047

Dept. of Computer Science & Engineering

Rajshahi University of Engineering & Technology

​

​

Outline

• Introduction
• Motivation
• Objectives
• Literature Review
• Workflow
• Dataset collection
• Data preprocessing
• Architecture of our CNN model
• Hyperparameter tuning
• Results
• Webserver Implementation
• Future scope
• Publication
• References

2

Introduction

3

• Pseudouridine (Ψ) is the most prevalent RNA modification found in both prokaryotes and eukaryotes.
• They have been confirmed to occur in rRNA, mRNA, tRNA, and nuclear/nucleolar RNA.
• Identifying them has vital significance in drug development, gene therapy, academic research etc.
• In this work, we proposed a multi-stage convolutional neural network for identifying Ψ-sites.

14/02/2021 Pseudouridine Identification by Deep Learning 3

4

• Previous predictors that were developed using laboratory techniques had some demerits such as-
• Costly
• Time consuming
• Required experienced professionals for maintenance.

​

• These problems encouraged us for applying computational methods in this task.

​

​

Motivation

14/02/2021 Pseudouridine Identification by Deep Learning 4

5

• To find an effective computational method to identify Pseudouridine sites.
• To investigate CNN’s performance on biological sequence inputs.
• To investigate whether multi-stage CNN model can produce better result than a sequential CNN model.
• To employ a CNN model which will produce better performances than the existing ones.

14/02/2021 Pseudouridine Identification by Deep Learning 5

Objectives

6

• Ppus(2015): SVM classifier by Li et al.[1]
• iRNA-Pseu(2016): Improved from previous classifier by Chen et al.[2]
• PseUI(2018): SVM classifier by He et al.[3]
• iPseu-CNN(2019): CNN classifier by Tahir et al.[4]
• XG-Pseu(2019): Gradient boosting based method Liu et al.[5]
• iPseu-Layer(2020): An ensemble model Mu et al. [6]

​

​

​

​

14/02/2021 Pseudouridine Identification by Deep Learning 6

Literature Review

7

14/02/2021 Pseudouridine Identification by Deep Learning 7

Workflow

8

• Data were collected for three different species which are H. sapience, S. cerevisiae and M. musculus.

​

​

​

​

​

​

14/02/2021 Pseudouridine Identification by Deep Learning 8

Dataset Collection

• Our data preprocessing had only one step- Binary ‘one-hot’ encoding.
• We did it in two ways:
• General ‘one-hot’ encoding
• Merged-seq ‘one-hot’ encoding
• General ‘one-hot’ encoding converted the inputs into (N, 4) matrix.

​

​

​

​

​

​

​

​

9

14/02/2021 Pseudouridine Identification by Deep Learning 9

Data Preprocessing

• Merged-seq ‘one-hot’ encoding converted our inputs into (N, 12) matrix.

​

​

​

​

​

​

​

10

14/02/2021 Pseudouridine Identification by Deep Learning 10

Data Preprocessing(Contd.)

11

14/02/2021 Pseudouridine Identification by Deep Learning 11

CNN Architecture(General)

12

14/02/2021 Pseudouridine Identification by Deep Learning 12

CNN Architecture(Multi-stage)

13

• In our investigation we tuned a number of hyperparameters.
• We employed k-fold cross-validation(k=10) and grid search for that.

​

14/02/2021 Pseudouridine Identification by Deep Learning 13

Hyperparameter Tuning

14

14/02/2021 Pseudouridine Identification by Deep Learning 14

Hyperparameter Tuning(Contd.)

​

​

​

​

​

​

​

​

15

14/02/2021 Pseudouridine Identification by Deep Learning 15

Results

*AC= ACCURACY, SN=SENSITIVITY, SP=SPECIFICITY, MCC= MATTHEWS CORRELATION COEFFICIENT

16

14/02/2021 Pseudouridine Identification by Deep Learning 16

Results(Contd.)

​

​

​

​

​

​

​

​

17

14/02/2021 Pseudouridine Identification by Deep Learning 17

Results(Contd.)

​

​

​

​

​

​

​

​

18

14/02/2021 Pseudouridine Identification by Deep Learning 18

Webserver Implementation(Step-1)

​

​

​

​

​

​

​

​

19

14/02/2021 Pseudouridine Identification by Deep Learning 19

Webserver Implementation(Step-2)

​

​

​

​

​

​

​

​

20

• Taking account of RNAs of other species.
• Taking account of other RNA modifications.
• Investigating CNN’s performance in other biological datasets.
• Implementing other encoding techniques.

​

14/02/2021 Pseudouridine Identification by Deep Learning 20

Future Scopes

• Based on the findings of our work, a paper was accepted and presented in the iEEE Region 10 Symposium (TENSYMP) 2020.
• The paper is available in the iEEE Xplore digital library. �

21

14/02/2021 Pseudouridine Identification by Deep Learning 21

Publication

​

​

​

​

​

​

​

​

[1] Y.-H. Li, G. Zhang, and Q. Cui, “Ppus: a web server to predict pusspecific pseudouridine sites,” Bioinformatics, vol. 31, no. 20, pp. 3362–3364, 2015.

​

[2] W. Chen, H. Tang, J. Ye, H. Lin, and K.-C. Chou, “iRNA-Pseu: Identifying rna pseudouridine sites,” Molecular Therapy-Nucleic Acids, vol. 5, p.e332, 2016.

​

[3] J. He, T. Fang, Z. Zhang, B. Huang, X. Zhu, and Y. Xiong, “PseUI: pseudouridine sites identification based on rna sequence information,” BMC bioinformatics, vol. 19, no. 1, p. 306, 2018.

​

[4] M. Tahir, H. Tayara, and K. T. Chong, “iPseU-CNN: Identifying RNA pseudouridine sites using convolutional neural networks,” Molecular Therapy-Nucleic Acids, vol. 16, pp. 463–470, 2019.

​

[5] Liu K, Chen W, Lin H. XG-PseU: an eXtreme Gradient Boosting based method for identifying pseudouridine sites. Molecular Genetics and Genomics. 2020;295(1):13–21.

​

[6] Mu Y, Zhang R, Wang L, Liu X. iPseU-Layer: Identifying RNA Pseudouridine Sites Using Layered Ensemble Model. Interdisciplinary Sciences: Computational Life Sciences. 2020; p. 1–11.

​

14/02/2021 Pseudouridine Identification by Deep Learning 22

References

Thank You