Asmitha Rathis Why Bioinformatics Protein structure Genetic Variants Anomaly classification Protein classification SegmentationSplicing Why is Deep Learning beneficial scalable with large datasets and are effective in identifying complex patterns from featurerich datasets ID: 932044
Download Presentation The PPT/PDF document "Deep Learning in Bioinformatics" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Deep Learning in Bioinformatics
Asmitha Rathis
Slide2Why Bioinformatics?
Protein structure
Genetic Variants
Anomaly classification Protein classificationSegmentation/Splicing
Slide3Why is Deep Learning beneficial?
scalable with large datasets and are effective in identifying complex patterns from feature-rich datasets
learn high levels of abstractions from multiple layers of non-linear transformations.
Slide4Terms
What are Motifs?
short, recurring patterns in DNA that are presumed to have a biological function
What is non-coding DNA? DNA that do not encode protein sequences.
Slide5Papers
DanQ
: a hybrid convolutional and recurrent deep neural network for quantifying the function of DNA sequences
- Daniel Quang and Xiaohui Xie [2016]Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning -
Babak
Alipanahi
et al [2015]
Exploiting the past and the future in protein secondary structure prediction -
Pierre Bald et al [1999]
Slide6DanQ:a hybrid convolutional and recurrent deep neural network for quantifying the function of DNA sequences
A predictive model for the function of non-coding DNA has enormous benefit for translation research
98% of human genome is non coding DNA and 93% of disease variants lie in this region
Previous work:
DeepSea
model
Propose a novel hybrid convolutional and bi-directional long short-term memory recurrent neural network framework
Slide7Network Model
Convolution for motifs
Recurrent layer for capturing dependency between the motifs and grammar
Slide8Training Details
Random initialization and initialize kernels from known motifs
Dropout is included
RMSprop algorithm with a minibatch size of 10060 epochs to fully train and each epoch of training takes ∼6 h
Slide9Results
Calculated ROC for each of the 919 binary targets on the test set
Predicted probability was the average of the forward and reverse complement sequence pairs
Slide10Results
Precision recall curve
Slide11Future Work
Better initialization techniques
Half are initialized with known motifs from JASPAR dataset
Datasets from more cell types
Slide12Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning: DeepBind
DNA- and RNA-binding proteins play a central role in gene regulation, including transcription and alternative splicing.
In the field of transcription, sequence specificity of DNA usually means how specific a protein, usually a transcription factor, recognizes its target DNA motif.
Slide13Challenges
Data come in qualitatively different forms,
eg
: microarray and sequencing data Quantity is very largeNeed to overcome the biases of existing technologies
Slide14Data
For training,
DeepBind
uses a set of sequences and, for each sequence, an experimentally determined binding score.
Slide15Binding score :
Slide16Training/Testing Details
training on
in vitro
data and testing on
in vivo
data.
vitro :
refers to the technique of performing a given procedure in a controlled environment outside of a living organism
Vivo :
tested on whole, living organisms or cells, usually animals, including humans, and plants,
Slide17Slide18Results
Slide19Analysis of potentially disease-causing genomic variants
Use binding models to identify, group and visualize variants that potentially change protein binding
Importance of each base based on the height of the letter
The mutation map indicating how much each possible mutation will increase or decrease the binding score.
A cancer risk variant in a
MYC
enhancer weakens a TCF7L2 binding site.
Slide20Analysis of Splicing Patterns
Slide21Exploiting the past and the future in protein secondary structure prediction
Predicting the secondary structure of a protein (alpha-helix, beta sheet, coil) is an important step towards understanding its three dimensional structure as well as its function.
Old methods : ML models that don’t capture variable long ranged information, Increasing size of window leads to overfitting
Slide22Slide23Slide24Results
Slide25Results
Overall performance close to 76% correct classification with 6 BRNNs
Use a range to limit the size of the window
Size of window
Slide26Questions
Based on the more recent models and technologies seen in class, which of them can be applied to these problems?
Can these techniques be applied to other bioinformatics tasks?