Classifying Diseases Using Convolutional Neural Networks

Soham Chatterjee

Satya Avva

Archana Iyer

Malaikannan Sankarasubbu

About Us

  • Satya Avva, Ph.D. Life Sciences graduate from LSU
  • Archana Iyer, Soham Chatterjee 
  • Freshers from SRM University, and 
  • B. Tech. Electrical, Electronics 
  • Next Tech Lab
  • Saama Blog
  • Soham's Blog

A BIG THANKYOU

  •  Thank you to Dr. Anand Dubey, Mr. Nikhil Gopinath, and Prof Chandrabose for providing their technical inputs and for reviewing our work
  • Thank you to Malaikannan, for bringing us together as a team and leading our research from end to end

What is DNA Methylation?

Addition of a methyl group to Cytosine base in CpG context

Why Study DNA Methylation?

Gene in the Normal Cell

Gene in the Cancer Cell

Lam K, et al., 2016

Process for Solving Our Problem

Data 

CpG Site  Number
Methylated Value
  • 32 diseases

  • 10,000 samples

  • Rows: 365860
  • Converted into a matrix of shape (220,1663)

Text

Data Pre Processing

Convolution Neural Network

Model Training 

  • Adam optimizer with an initial learning rate of(1e-4)
  • All of the parameters were zero-centered
  • Normal distribution with standard deviation of 0.02
  • Training was halted if the model accuracy decreased twice by more than 0.5%

Accuracy

Loss

  • Base CNN- This was a vanilla CNN with 3X3 filters; 3 filter layers; 10 filters in each layer
  • CNN Width: Rectangular filter with window sizes 1, 3, 5, 7

  • CNN Height and Width: Combination of both height-wise and width-wise filter of windows 1, 3, 5, 7

  • CNN Height: Height-wise filters of window 1, 3, 5, 7.

Training Approach 

Model Accuracy

Conclusion

  • Model is able to learn the minute differences of changing Methylation patterns on a genome-wide scales.
  • Deep Learning based approach is faster, cheaper and more accurate than the traditional approach.
  • With more appropriate training data, it can also be possible to predict the different stages of cancer.

Contact us

  • Feedback- A link to give us a feedback-tinyurl.com/dnamethyltalk
  • We would love to hear from you and your responses will be anonymous!
  • You can get the link to the slides at the end.
  • Contact us:
  • Satya Avva 
    • satya.avva@saama.com
  • Archana Iyer:
    • varchanaiyer139@gmail.com
  • Soham Chatterjee:
    • 96soham96@gmail.com
    • csoham.wordpress.com 

DNA

By archana iyer

DNA

  • 1,014