Transfer String Kernel for Cross-Context Sequence Specific DNA-Protein Binding Prediction. by Ritambhara Singh IIIT-Delhi June 10, 2016

Transcription

1 Transfer String Kernel for Cross-Context Sequence Specific DNA-Protein Binding Prediction by Ritambhara Singh IIIT-Delhi June 10,

2 Biology in a Slide DNA RNA PROTEIN CELL ORGANISM 2

3 DNA and Diseases Down Syndrome Parkinson s Disease Autism Muscular Atrophy DNA RNA PROTEIN Sickle Cell Disease... CELL ORGANISM 3

4 Transcription Factors Transcription Factor ATCGCGTAGCTAGGGATGACAGACACACATAATTCTAGATA DNA RNA PROTEIN Transcription Factor Binding Site CELL Gene ORGANISM 4

5 ChIP-seq Maps TF binding Transcription Factor Transcription Factor Binding Site Gene DNA CHIP-SEQ Peak ChIP-seq Map for TF ATATCGTATCTTTTAAACCGGGTTGGCCACTAGA ATATCGTATCTAAACCGCCTCGG Genome 5

6 TF Binding Differs Across Contexts ATATCGTATCTTTTAAACCGGGTATGTAATGCAT ATATCGTATCTAAACCGCCCGTGT ATATCGTATCTTTTAAACCGGGTTGGCCAGTATA ATATCGTATCTAAACCGCCCTGCA 6

7 Source : 7 Current Challenge: ENCODE Data Gap (Blood Cell) (Stem Cell) (Leukemia) (Lung Cancer) (Immunity related)?? (Nerve Cell) (Cervical Cancer)

8 Case for Computational Tools 8

9 Existing Computational Tools Generative Approaches MEME CISFINDER Discriminative Approaches STRING KERNEL+SVM 9

10 Generative : PWM Based approach Peak ATATCGTATAACAATAACCGGGAACTAATAGC ATATCGTATCTAACAAATCCTACT ChIP-seq Map for TF Genome A T C G Position Weight Matrix Sequence Logo 10

11 Generative Approach : Output?? ATATCGTATCTTTTAAACCGGGTTGGCCAATAGC ATATCGTATCTAAACCGCCCTACT Genome Source : 11

12 Generative Approach: Limitations Output: Long list of potential TFs Work well for only well preserved motifs or large training datasets PWMs for all ~2000 TFs not available Lower prediction performance than discriminative approaches 12

13 Discriminative Approach : Output?? ATATCGTATCTTTTAAACCGGGTTGGCCAATAGC ATATCGTATCTAAACCGCCCTACT Genome Peak ATATCGTATCTTTTAAACCGGGTTGGCCAATAGC ATATCGTATCTAAACCGCCCTACT Genome

14 Discriminative : String Kernel Approach Support Vector Machine 14

15 Discriminative Approach : Limitation Assumption: Training/test data follow same distribution regardless of context. 15

16 Aim Improve prediction of Transcription Factor Binding sites across contexts using knowledge transfer. 16

17 Proposed Solution : Cross-Context Knowledge Transfer 17

18 Transfer String Kernel : Overview Source Context Xs ATCGAT GTATAC Xt ATACAT GCTTAC Target Context Feature Conversion Feature Conversion Knowledge Transfer (KMM) Training Classification 18

19 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 19

20 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 20

21 String Kernel : Spectrum Kernel Feature map indexed by all k-length subsequences ( k-mers ) from alphabet Σ of amino acids, Σ =20 21

22 String Kernel : Mismatch Kernel For k-mer s, the mismatch neighborhood N (k,m) (s) is the set of all k-mers t within m mismatches from s. 22

23 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 23

24 Support Vector Machine w. x + b -1 w. x + b = 0 w. x + b +1 Negative Instances (y = -1) Positive Instances (y = +1) 24

25 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 25

26 Transfer Learning (KMM) True densities p tr (x) Ratios r(x) p te (x) r(x) 26

27 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 27

28 Importance Re-weighting Original Weights KMM Weights 28

29 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 29

30 Transfer String Kernel (TSK) Source! Context! ATCGATCG ATCGATCG% CCCGATCG CTCGCTCC% Target! Context! Mismatch String Kernel Feature Conversion! Feature Conversion! Mismatch String Kernel Kernel Mean Matching (KMM) Knowledge Transfer! Importance Reweighting Training! SVM Classification! 30

31 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 31

32 Experimental Setup 14 Transcription Factors (ENCODE ChIP-seq) Top 1000 positive sequences (500 training and 500 testing) 1000 random negative sequences Hyper-parameter tuning for k=(8,10,12) and m=(1,2,3) Dictionary size = 4 {A,T,C,G} 32

33 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 33

34 Results 34

35 Results Cross Context AUC Score Sin3a Max Mxi1 Chd2 Ctcf Transcription Factors TSK SK 35

36 Outline Method String Kernel Support Vector Machine Transfer Learning (KMM) Importance re-weighting Transfer String Kernel Evaluation Experimental Setup Cross-context TFBS prediction Cross-context Protein Binding prediction 36

37 Results Cross context 37

38 Summary TSK overall improves the cross-context TFBS predictions; String kernel based approaches perform better than the state-ofthe-art Position Weight/Frequency Matrix based TFBS tools; TSK approach is generalizable for performance improvement of any cross-context sequence prediction task. Presented in BIOKDD 15 38

39 Acknowledgements Dr. Mazhar Adli Adli Lab : Department of Biochemistry and Molecular Nipun Batra IIIT-Delhi 39

40 Machine Learning UVa Dr. Yanjun Qi (Advisor) Beilun Wang Weilin Xu Jack Lanchantin Ji Gao 40

41 Future Directions Deep Learning : Gene expression prediction using histone modification data (ECCB 2016) Improving TFBS prediction using DNA sequences (ICLR Workshop 2016, ICML Workshop 2016) String Kernels: Improving efficiency!! (on-going work) 41

42 Thank You 42