1. Channel-Independent Viterbi Algorithm (CIVA) for DNA Sequencing
Xiaohua (Edward) Li
Department of Electrical and Computer Engineering
State University of New York at Binghamton

2. Outline Introduction CIVA Use CIVA for base-calling Simulations
Conclusions

3. Introduction: DNA sequencing
DNA sequencing (base-calling)
Procedure
template, PCR, electrophoresis, gel image, trace file
Base-caller

4. Introduction: Base-caller
Base-calling: detect DNA base sequence
Approaches
Manual reading, automated by heuristic knowledge
Image processing with signal models (ABI, Phred)
Deconvolution with communication (ISI) signal model, e.g., MLSE, MAP

5. Proposed Method: CIVA
Our method: with ISI model, robust to signal irregularity
Difficulty comes from irregular trace signal
Amplitude and position jitter
Short signal, limited samples, yet time-varying
Solution: CIVA
joint symbol/position optimization
without channel estimation

6. CIVA: Basic Idea List all possible symbol matrices S(n),
Find a probe for each possible S(n)
Use all probes to determine S(n) from X(n)

7. CIVA: Properties
CIVA: a trellis searching algorithm where metrics are calculated by probes
Properties
Near optimal for even ill-conditioned channels
No channel estimation, channel independent
High computational complexity
Applications
Direct application: system with simple signaling and short channel, e.g., GSM, sensor networks, base-calling
Future: more application with complexity reduction

8. CIVA for Base-calling Model trace signal with communication system
Channel effect introduces ISI

9. Symbol Matrix Structure

10. Probe Construction

11. Probe Construction Example

12. Trellis Metric Calculation

13. CIVA Trellis Search

14. Special Consideration for DNA Trace Signal
Amplitude jitter
solved inherently
Limited trace samples and time varying
fast convergence of CIVA
Timing jitter
looking for best timing for each sample

15. Simulations: Experiment 1
A trace file with reference bases from Staden Package
Normalize trace, find approximate base interval, apply CIVA with M=P=1 (2-tap channel. 25 trellis states, 125 transitional paths)
Results: less than 3% error compared with reference

16. Simulations: Experiment 1
Two zoom-in sections
#1. with confident base detections
#2. with undetermined N

17. Simulations: Experiment 2
A gel image from Prof. S. Gal with low quality
Scanning to trace signal

18. Simulations: Experiment 2
Apply CIVA for base-calling
A zoom-in section

19. Conclusions CIVA algorithm proposed for DNA sequence base-calling
Robust to signal irregularity with affordable computational complexity
Experiments show positive performance
More experiments are required for evaluation